Mariano Grosman

Anduril's pre-IPO position is structurally different from every other defense-tech name

Mariano Grosman — Sun, 21 Jun 2026 11:33:50 GMT

Hey everybody,

When allocators ask me about Anduril, they almost always reach for one of two mental models. Either they treat it like a defense contractor, and reach for the framework they would use on a legacy prime. Or they treat it like a high-growth technology company, and reach for the software playbook. In my view, both are wrong, and the reason they are wrong is the most interesting thing about the position.

Anduril is structurally a hybrid that does not fit either box. It sells into defense, which means its revenue, its customers, and its growth are shaped by a procurement system unlike anything in commercial technology. But it is built like a technology company, with a product-first model that is the opposite of how legacy defense primes are structured. That combination is rare, and it creates a structural profile that neither the defense framework nor the software framework captures correctly.

This is the same kind of structural analysis I applied to Anthropic and OpenAI in an earlier post. Not a view on whether the company is good. A view on how the position is actually built, because that structure is what determines how it should be analyzed. I am going to name Anduril throughout this post, because the whole point is the structure specific to this company. As always, naming a company to analyze its structure is a completely different thing from having anything to offer, and this post does neither buying nor selling. It is analysis.

→ Why the defense-contractor framework gets Anduril wrong

→ Why the software-startup framework gets it wrong too

→ The procurement structure that shapes everything about the position

→ The product-first model that separates it from legacy defense

→ The structural questions an allocator should actually be asking

By the end of this post, you will understand why Anduril sits in a structural category of its own, and the specific questions that category raises for anyone evaluating the private position.

Here is what is inside.

Why the defense-contractor framework is wrong
Why the software-startup framework is wrong
The procurement structure that shapes the position
The product-first model that sets it apart
The structural questions to actually ask
Why the category matters

Let’s begin.

The references and observations behind this post:

→ My own structural analysis of how defense-technology businesses are built and how that shapes their equity

→ Public, widely-reported information about how defense procurement works in the United States

→ The structural framework from my earlier post comparing two frontier AI companies, applied here to a different sector

→ Claude for pressure-testing the structural argument against the two conventional frameworks

→ Conversations with allocators about how they are categorizing defense-technology names

A note on figures: this post is entirely structural and contains no valuations, revenue figures, contract values, or other quantitative claims. Where I describe how the business is built or how procurement works, it is my analysis and widely-reported general knowledge, framed as such. Anduril is named here solely to analyze the structure of its position, not in connection with any transaction.

1. Why the defense-contractor framework is wrong

When an allocator reaches for the defense-contractor framework, they are usually thinking about the legacy primes. Long-dated cost-plus contracts. Slow, relationship-driven procurement. Revenue that is highly predictable but also highly constrained by the structure of government contracting. A business where the moat is incumbency and the relationships are measured in decades.

Applying that framework to Anduril gets several things wrong at once.

The legacy prime model is built around winning large, specific programs of record and then delivering against them over many years, often on a cost-plus basis where the economics are structured around reimbursed costs plus a margin. In my view, Anduril’s model is meaningfully different in structure. It is built around developing products first and then selling them, which is much closer to how a commercial technology company operates than how a traditional prime operates.

That structural difference matters for the equity. A legacy prime’s value is tied to its backlog of awarded programs and its entrenched position in a slow procurement cycle. If you analyze Anduril as though its value works the same way, you are looking for the wrong things. You are looking for backlog when you should be looking at product adoption. You are looking for incumbency when the entire thesis is about displacing incumbency. The framework points your attention in the wrong direction.

The defense-contractor framework is not wrong because defense is the wrong sector. It is wrong because it describes a structurally different kind of defense company than the one Anduril is built to be.

2. Why the software-startup framework is wrong

So allocators sometimes swing the other way and reach for the software-startup framework instead. Product-first, high growth, large addressable market, the familiar technology playbook. This gets closer to the product model, but it gets the customer and the revenue structure badly wrong.

A commercial software company sells into a market of many independent buyers who make relatively fast purchasing decisions on commercial terms. The growth dynamics, the sales cycles, the revenue recognition, all of it is shaped by that commercial reality. The software framework is built on those assumptions.

Anduril does not sell into that world. It sells into defense procurement, and defense procurement is structurally unlike any commercial market. The customer is the government. The buying process is governed by appropriations, program timelines, and a procurement apparatus that operates on its own logic and its own clock. The sales cycle is not a commercial sales cycle. The revenue does not behave like commercial software revenue. The addressable market is not a commercial addressable market, it is a defense budget allocated through a political and bureaucratic process.

In my view, this is where the software framework fails. It correctly identifies that Anduril is product-first, but it incorrectly assumes that product-first means commercial-market dynamics. The product is built like technology. The customer and the revenue are shaped by defense. An allocator using the pure software framework will model growth and sales cycles as though they behave commercially, and in my view they do not.

So the defense framework gets the product model wrong, and the software framework gets the customer and revenue wrong. Neither one describes the actual structure. That is what makes this a hybrid, and the hybrid is the whole point.

3. The procurement structure that shapes the position

To analyze Anduril correctly, in my view you have to start with the thing that shapes everything: defense procurement, and how it differs from a commercial market.

The customer is singular and structured. In commercial markets, you sell to many buyers. In defense, the customer is effectively the government, buying through a structured appropriations and procurement process. That changes the concentration profile of the revenue and the nature of the customer relationship in ways that neither the pure software framework nor the legacy prime framework captures the same way.

The buying cycle runs on its own clock. Defense procurement is governed by budget cycles, program timelines, and political processes. The implication for an allocator is that the growth path is shaped by how a company moves through that procurement apparatus, which is structurally different from how a commercial company grows through a market.

The barriers to entry are unusual. Selling into defense requires clearances, compliance, relationships, and the ability to navigate a procurement system that is itself a formidable barrier. In my view, this is one of the most distinctive structural features. The barrier protecting a successful defense-technology company is not only the product. It is the demonstrated ability to sell through a system that is genuinely hard to sell through.

The displacement thesis is structural, not just competitive. The interesting structural question with Anduril is whether a product-first company can displace the legacy prime model inside this procurement system. That is not a normal competitive question about a better product. It is a question about whether the structure of how defense buys can shift toward a different kind of supplier. In my view, that structural question is the actual thesis, and it is invisible to both conventional frameworks.

If you understand the procurement structure, you understand why Anduril cannot be analyzed like either a software company or a legacy prime. The procurement system is the environment that shapes the entire position, and it behaves like neither a commercial market nor the slow incumbency game of the old primes.

4. The product-first model that sets it apart

The other half of the hybrid is the product-first model, and this is where Anduril structurally departs from the legacy defense world even though it operates in the same procurement system.

The legacy prime model, in broad structural terms, is oriented around winning a specified program and building to that specification, frequently on cost-plus economics. The government defines the requirement, and the contractor delivers against it. The structural orientation is toward the contract.

A product-first model inverts this. The company develops products on its own initiative and then sells them into the procurement system, rather than waiting for a program to be specified and bidding to build it. In my view, this is the genuinely different structural feature, because it changes what the company is at its core. It is oriented toward the product, not the contract.

This product-first orientation is why the software framework is tempting. The orientation does resemble a technology company. But the difference, again, is the customer. A commercial technology company builds products for a commercial market. Anduril builds products for the defense procurement system. Same product-first orientation, completely different buyer.

In my view, the equity implication of the product-first model is significant. It means the value is tied to product adoption and the ability to keep building products the procurement system wants, rather than to a backlog of awarded programs. That is a structurally different value driver than a legacy prime, and analyzing it requires looking at different things. This is the structural insight that neither framework delivers on its own. You need the product lens from technology and the customer lens from defense, held at the same time.

5. The structural questions to actually ask

If Anduril is a structural hybrid, then the right diligence questions are different from the ones either conventional framework would generate. In my view, here are the structural questions an allocator should actually be asking. None of them require me to have a view on the company’s quality, and none of them touch any specific transaction. They are the questions the hybrid structure raises.

→ How concentrated is the revenue across government customers and programs, and how should that concentration be understood given the singular nature of the defense customer? This is a structural question the software framework would not even think to ask.

→ How does the company actually move through the procurement apparatus, and what does its demonstrated ability to navigate that system tell you about the durability of its position? The procurement barrier is part of the moat, so understanding how the company clears it is central.

→ Is the value driven by product adoption or by awarded programs, and which model does the company’s structure actually follow? This is the question that separates the product-first read from the legacy-prime read, and it determines what you should be tracking.

→ What is the displacement thesis, structurally? Is the bet that a product-first model can shift how defense procures, and what would have to be true structurally for that to happen? This is the actual thesis hiding underneath the position.

→ How do the structural features of defense revenue interact with the standard pre-IPO structural checks? Everything from my earlier diligence-checklist post still applies, but the defense context changes how some of those checks should be read.

These are the questions the hybrid structure generates. In my view, an allocator asking these is analyzing the actual position. An allocator asking the standard software questions or the standard defense-prime questions is analyzing a company that does not exist, because Anduril is neither of those things.

6. Why the category matters

The reason all of this matters comes back to a simple point I keep returning to in these posts. You cannot analyze a position correctly with the wrong framework, and the wrong framework is exactly what most allocators reach for when a company does not fit a familiar box.

Anduril does not fit a familiar box. In my view it is structurally a hybrid: a product-first company selling into a procurement system that behaves like neither a commercial market nor the slow incumbency game of the legacy primes. The defense framework gets the product model wrong. The software framework gets the customer wrong. Only a framework that holds both lenses at once describes the actual structure.

Three things to take from this:

→ Match the framework to the structure, not to the sector. Anduril is in defense, but the legacy-defense framework does not fit it. The sector label is not the same as the structural model, and in my view conflating the two is the most common analytical error on names like this.

→ The hybrid structure is the thesis, not a complication. The interesting question, whether a product-first model can shift defense procurement, only becomes visible once you stop forcing the company into one box. The structure is where the actual opportunity and the actual risk both live.

→ The right questions come from the structure. Once you see Anduril as a hybrid, the diligence questions change, and in my view they become much sharper than anything either conventional framework would generate.

The interesting question is not whether Anduril is a good defense company or a good technology company. In my view that framing is the error. The question is whether you are analyzing the structure that actually exists, and that structure belongs to a category Anduril largely occupies by itself. The discipline is in seeing the position for what it is structurally, not for which box is easiest to reach for.

Curious what this means for your portfolio?

Book a 20-minute call. We can walk through the structural read on defense-technology names and how to think about the category.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

Rainmaker Securities, LLC (“RMS”) is a FINRA (FINRA.org) registered broker-dealer and SIPC (SIPC.org) member. Find this broker-dealer and its agents at brokercheck.finra.org. Our relationship summary can be found at rainmakersecurities.com/disclosures.

RMS is engaged by its clients to make referrals to buyers or sellers of private securities (“Securities”). If such client closes a Securities transaction with a buyer or seller so referred, RMS is entitled to a success fee from the client. Such success fee may be in the form of cash or in warrants to purchase securities of the client or client’s affiliate. RMS or RMS representatives may hold equity in its issuer clients or in the issuers of securities purchased or sold by the parties to a transaction.

This communication is confidential and is addressed only to its intended recipient. This communication does not represent an offer or solicitation to buy or sell Securities. Such an offer must be made via definitive legal documentation by the seller of securities.

RISK FACTOR LANGUAGE

Investments in the Securities are speculative and involve a high degree of risk. Companies engaging in private placements may be early stage and high risk. You should be able to afford the increased risk of loss with such investments, including the potential of a total loss.

An investor in the Securities should have little to no need for liquidity in the foreseeable future. Unlike an investment purchased on a stock exchange, an investment in a private placement is highly illiquid. You will mostly likely be investing in restricted securities, may have difficulty finding a buyer for the securities when you can resell and, as a result, may need to hold the securities indefinitely.

Limited disclosure Information. Companies engaging in private placements are not required to provide the disclosure that would be required in a registered offering. You may have less information to make an informed investment decision than, for example, stock purchased on a stock exchange, including information that may help you determine whether the price asked for the investment is a fair price. Potential buyers or sellers of the Securities should seek professional counsel prior to entering into any transaction.

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

The pre-IPO due diligence checklist most allocators skip

Mariano Grosman — Fri, 19 Jun 2026 11:08:46 GMT

Hey everybody,

Most allocators I work with are good at financial diligence. They read the financials. They model the growth. They pull the comparable IPOs. They stress the revenue assumptions. By the time they call me, the financial work is usually done and done well.

Then they stop. And the place they stop is exactly where, in my experience, the expensive surprises live.

The financial diligence tells you whether the business is good. It does not tell you what you actually own, who can block your exit, or what the terms do to your proceeds when something goes sideways. That second category, the structural and behavioral diligence, lives outside the deal materials, and almost nobody runs it with the same rigor they bring to the financials.

This post is the six checks I never skip. None of them show up in the pitch deck. All of them change what you actually own when the exit comes.

→ Why financial diligence is necessary but not sufficient

→ The right-of-first-refusal language that can block your exit

→ The transfer mechanics that decide whether you can even sell

→ The 409A and lockup terms you inherit without realizing it

→ The SPV manager track record and information rights most allocators never check

By the end of this post, you will have a six-point checklist for the structural diligence that financial analysis does not cover, and a sense of why each one matters at exit.

Here is what is inside.

Why financial diligence is not enough
The right-of-first-refusal check
The transfer mechanics check
The 409A and lockup check
The SPV manager and information rights check
The six-point checklist

Let’s begin.

The references and observations behind this post:

→ My own broker book and direct experience of where deals create surprises after the financial work is done

→ The structural frameworks from my earlier posts, applied here as a practical checklist

→ Standard private-securities transaction documents, where these terms actually live

→ Claude for pressure-testing the checklist against real deal structures

→ Conversations with allocators about the surprises they did not see coming

A note on figures: this post is observational and structural, drawn from my own experience. It contains no valuations or market statistics. Where I describe what a check catches, it is my account, framed as such.

1. Why financial diligence is not enough

Financial diligence answers one question: is this a good business at this price? That is a necessary question. It is not the only one.

In pre-IPO, you are not buying a share of a business in a clean, liquid market. You are buying a specific position, with specific rights, governed by specific documents, that sits in a specific place in a cap table waterfall. Two allocators can do identical financial diligence on the same company, buy positions at the same valuation, and end up with materially different outcomes, because what they actually bought was different in ways the financial analysis never examined.

The financial work tells you whether the company will be worth more later. The structural work tells you whether you will actually capture that value, whether you can exit when you want, and what happens to your proceeds when the terms get tested. In my experience, the allocators who lose money in pre-IPO rarely lose it because they misjudged the business. They lose it because they missed something structural that the financial diligence was never designed to catch.

The six checks below are the ones I run on every position, after the financial work is done, precisely because the financial work does not cover them.

2. The right-of-first-refusal check

The first thing I check is who can block or interfere with your eventual exit, and the most common mechanism is the right of first refusal.

A right of first refusal, or ROFR, gives the company or existing investors the right to match any offer you receive before you can sell to an outside buyer. On the surface this sounds harmless. In practice, in my experience, it can significantly affect both the timing and the price of your exit.

Here is why it matters. When you go to sell your position, a ROFR means you cannot simply transact with the buyer you found. You have to take that offer back to the company or the existing holders and give them the chance to match it. That process takes time, and during that time the buyer you lined up may walk. Worse, the existence of a ROFR can suppress the offers you get in the first place, because sophisticated buyers know the deal is not certain to close and may not want to do the work of pricing a position they might lose to the ROFR holder.

The check is simple to describe and easy to skip. Read the operating agreement and the shareholder agreement for the ROFR language. Understand exactly who holds the right, how long they have to exercise it, and what the process looks like. In my experience, an aggressive ROFR is one of the most common reasons a pre-IPO exit takes far longer than the allocator expected, and it is almost never discussed during the financial diligence.

3. The transfer mechanics check

Closely related, but distinct, is the question of whether you can transfer the position at all, and what the mechanics actually require.

Many pre-IPO positions come with transfer restrictions that go beyond the ROFR. The company may have to approve any transfer. There may be restrictions on who you can sell to, requiring the buyer to meet certain qualifications. There may be blackout periods during which no transfers are allowed. There may be a requirement that the company’s board formally consents, which means your exit depends on getting on a board agenda.

In my experience, allocators routinely assume that owning a position means they can sell it when they choose. In pre-IPO, that assumption is often wrong. The transfer mechanics can mean that selling requires the cooperation of parties who have no obligation to move quickly, and sometimes have an incentive to move slowly.

The check is to map the actual transfer process before you buy. Who has to approve a transfer. How long approvals typically take. What qualifications a buyer must meet. Whether there are blackout periods. In my experience, understanding the transfer mechanics before you enter a position is the difference between an exit you control and an exit that depends on other people’s goodwill.

4. The 409A and lockup check

The next two checks are about terms you inherit, often without realizing it.

The 409A consideration.

The 409A valuation is the company’s official valuation of its common stock for tax purposes. It matters to a secondary buyer because it sets a reference point that can diverge significantly from the price you are paying. When the 409A valuation and the secondary price diverge, it can create tax complexity and, in some situations, affect how the position is treated. In my experience, this is a detail that financial diligence skips entirely, because it is a tax and structural matter, not a business-quality matter. It is worth understanding the current 409A status of the common stock and how it relates to the price you are being asked to pay.

The lockup consideration.

Pre-IPO positions often carry lockup terms inherited from the company’s financing rounds, and these terms roll forward to you as the new holder. A lockup restricts when you can sell after a liquidity event, including after an IPO. The allocator who assumes they can sell immediately after the company goes public may find they are locked up for a period inherited from terms they never examined. In my experience, the lockup terms that come attached to a position are one of the most commonly overlooked features, and they directly affect when you can actually realize your investment.

The check for both is to read the documents for the 409A status and any lockup provisions, and to understand exactly what rolls forward to you when you take the position. Neither of these is visible in the financials. Both can materially change your timeline to liquidity.

5. The SPV manager and information rights check

The last two checks are about the structure you are buying through and the visibility you get afterward.

The SPV manager track record.

Many pre-IPO positions are held through a special purpose vehicle, an SPV, managed by a third party. When you buy into an SPV, you are not just buying exposure to the company. You are also relying on the SPV manager to handle the position, manage the wind-down, and distribute the proceeds when the exit comes. In my experience, allocators focus entirely on the underlying company and almost never examine the SPV manager. That is a mistake. How the manager has handled prior wind-downs, how they distribute shares, how transparent and competent they have been on past vehicles, all of this affects what actually happens to your money at exit. The check is to look at the manager’s track record on prior SPVs before you rely on them for this one.

The information rights.

Finally, understand what you will actually be able to see after you buy. Information rights determine whether you get ongoing financial updates, cap table changes, and material developments, or whether you are buying a position and then going dark until the exit. In my experience, the level of information rights varies widely, and allocators rarely check before committing. The check is to understand exactly what reporting and visibility you are entitled to post-transaction, because there is a meaningful difference between a position you can monitor and one you cannot.

6. The six-point checklist

Here is the full checklist in one place. Run it on every position after the financial diligence is done.

1. Right of first refusal. Who can match or block your exit. How long they have. What the process does to your timing and price.

2. Transfer mechanics. Whether you can transfer at all. Who approves. How long it takes. What qualifications a buyer must meet. Whether there are blackout periods.

3. 409A status. The official common-stock valuation, how it relates to your purchase price, and the tax and structural implications of any divergence.

4. Lockup terms. What lockup provisions roll forward to you, and how they affect when you can sell after a liquidity event.

5. SPV manager track record. If you are buying through an SPV, how the manager has handled prior wind-downs and distributions.

6. Information rights. What ongoing reporting and visibility you are entitled to after the transaction closes.

None of these six are answered by the financial diligence. Every one of them affects what you actually own, whether you can exit, and what you receive when you do. In my experience, the allocators who run this checklist alongside their financial work are the ones who avoid the surprises, and the allocators who skip it are the ones who discover these terms at exactly the moment they wish they had known earlier.

The financial diligence tells you whether the business is good. This checklist tells you whether the position is. You need both.

Want to go deeper?

Book a 20-minute call. We can walk through the structural diligence on whatever you are evaluating.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

The SpaceX IPO headline was the valuation. The lesson is the lock-up

Mariano Grosman — Thu, 18 Jun 2026 10:45:15 GMT

Hey everybody,

On June 12, SpaceX went public, and the headline wrote itself. Priced at $135 a share, roughly $75 billion raised, a valuation around $1.75 trillion, the biggest IPO in history by a wide margin (SpaceX pricing announcement, June 11, and Reuters). The stock closed its first day near $161, up about 19% (CNBC). Every outlet led with the size, and fair enough. The size is historic.

But the number is not the part I would tell a pre-IPO investor to study. The part worth studying is the lock-up.

Here is the detail almost none of the headlines led with, and the one that matters most for anyone who holds private shares. The offering itself was primary only. SpaceX sold new shares as a company. No existing holders, not the early employees, not the venture investors who waited more than two decades, cashed out at the listing (as reported by Quartz from the S-1). The people who had been holding illiquid SpaceX paper for years did not get their liquidity on IPO day. They got a staggered, conditional schedule that runs for months, and in the largest holder’s case, more than a year.

That is the lesson. For a pre-IPO investor, the IPO is not the exit. It is the start of the exit. And the structure of that exit was decided long before the bell rang.

→ What actually happened on June 12

→ Why the lock-up, not the valuation, is the real story for pre-IPO holders

→ How the staggered lock-up actually works, tranche by tranche

→ Why “eligible to transfer” is not the same as “able to sell”

→ What all of this teaches you about your own eventual exit

By the end of this post, you will understand why the most instructive part of the biggest IPO in history is the part that decides when insiders can sell, and what it means for how you should think about your own private positions.

Here is what is inside.

What actually happened on June 12
Why the lock-up is the real story
How the staggered lock-up works
“Eligible to transfer” is not “able to sell”
What this teaches you about your own exit
The structure was always the story

Let’s begin.

The references and sources behind this post:

→ SpaceX’s S-1 registration statement and its June 11 IPO pricing announcement, the primary documents for the offering terms

→ Reporting on the lock-up structure from CNBC, Quartz, Investing.com, and Darrow Wealth Management, each summarizing the S-1 lock-up terms

→ NPR and CNBC for the first-day trading figures

→ Claude for organizing the structural argument and cross-checking the lock-up mechanics against the reporting

A note on figures: every figure in this post (the offering price, the amount raised, the valuation, the first-day move, and the lock-up percentages and dates) is drawn from SpaceX’s own filings and pricing announcement or from the major financial press reporting on the S-1, and is attributed inline. Where the cumulative unlock path depends on a performance condition, I have said so rather than assuming it.

1. What actually happened on June 12

The facts, briefly, because the size is not the point but it sets the stage.

SpaceX priced its IPO at $135 a share on about 556 million shares of Class A common stock, raising roughly $75 billion and listing on Nasdaq under the ticker SPCX (SpaceX pricing announcement, June 11). At that price the company was valued at roughly $1.75 trillion, which made it the largest IPO on record, more than twice the previous record (Reuters, and NPR). The stock opened around $150 and closed its first session near $161, up about 19% (CNBC).

Two structural features of the offering matter more than the headline number.

First, the offering was primary. The shares sold in the IPO were new shares issued by SpaceX as a company. Existing shareholders did not sell into the offering (as reported by Quartz, summarizing the S-1). That means the venture funds and the employees who held SpaceX equity, some of them for the better part of two decades, did not get liquidity at the listing. Their ability to sell is governed entirely by the lock-up.

Second, the company carries a dual-class structure. Musk controls roughly 42% of the equity and about 85% of the voting power (Investing.com). That governance fact is its own discussion, and it drew a public letter from Senator Elizabeth Warren urging the SEC to look harder at the offering (Financer). But the feature I want to focus on is the one that decides when all those long-time holders can actually sell. The lock-up.

2. Why the lock-up is the real story

For a pre-IPO investor, the lock-up is the part of any IPO that should command the most attention, because it is the part that determines when an investment actually becomes cash.

Most people think of the IPO as the finish line. The company goes public, and the holders are free. That is not how it works, and SpaceX is a clean illustration of why. The holders who had been waiting for liquidity did not get it on June 12. They got a schedule. And until that schedule releases their shares, their position is worth a great deal on paper and nothing in the bank.

This is the single most underappreciated idea in pre-IPO investing, and it is the same idea I wrote about in the due-diligence checklist post, where lock-up terms were one of the structural checks most allocators skip. The SpaceX IPO just made the point at the largest possible scale. Even in the biggest, most heavily demanded offering in history, getting to the IPO and getting to liquidity were two completely different events, separated by months and conditions.

The valuation tells you what the company is worth. The lock-up tells you when, and under what conditions, a holder can realize any of it. For anyone holding private shares, the second question is the one that touches your actual life, because it decides when the position turns into money you can use. SpaceX designed an unusually elaborate answer to that question, and the design is worth understanding.

3. How the staggered lock-up works

Most IPO lock-ups are simple: a single 180-day wall, after which insiders can sell. SpaceX did something more elaborate. It built a staggered release, a series of tranches that free up portions of insider stock at different points across the 180-day window, with one of those tranches tied to how the stock performs. The terms below come from the S-1, as reported by CNBC, Darrow Wealth Management, and Investing.com.

Here is the sequence.

The first release comes after Q2 earnings. Once SpaceX reports its first quarterly results as a public company, covering the April to June period (the call is expected between mid-July and September), up to 20% of eligible insider shares are released.

A performance kicker can add to that first release. If the stock trades at least 30% above the $135 offering price on 5 of 10 consecutive trading days before that first earnings release, up to an additional 10% is freed early. Thirty percent above the offer price is about $175.50 a share. The stock closed its first day near $161, so as of the debut that condition was in play but not yet met. Whether it triggers depends on where SPCX trades in the weeks ahead, which is exactly why a cumulative unlock chart can only show you a scenario, not a certainty.

Five time-based tranches release through the middle of the window. At 70, 90, 105, 120, and 135 days after the offering, each date frees up to a further 7% of eligible shares. Those dates fall roughly across late August through late October.

A larger release comes after Q3 earnings. When SpaceX reports its July to September results, up to 28% more is freed.

The remainder comes off at 180 days. The full balance of eligible insider shares is released at the end of the 180-day period, which lands around early to mid December 2026.

Musk is carved out of all of it. The accelerated schedule does not apply to him. He is subject to a separate, longer restriction reported at 366 days, with his shares unlocking around June 2027 (Investing.com, and reporting referencing his 2027 unlock). The largest holder waits the longest.

One more nuance the structure includes: a 5% friends-and-family allocation carries no lock-up at all (Investing.com). So a small slice was free immediately, while the overwhelming majority moved onto the staggered schedule above.

That is a genuinely unusual design. A normal lock-up concentrates all the selling pressure on a single day. This one spreads it across roughly six months, with the early portion conditional on performance. Whether that smoothing works as intended is one of the most-watched questions in the market right now. But the mechanics are the point for our purposes, because they show how much structure sits between a holder and their money.

4. “Eligible to transfer” is not “able to sell”

Here is the distinction that separates someone who understands lock-ups from someone who has only read the headline number. When a tranche releases, the shares become eligible to transfer. That is not the same as being able to sell freely, and the gap between the two is where people get surprised.

Several things sit in that gap.

Company trading policies still apply. A release date in the lock-up agreement frees the shares from the lock-up. It does not override the company’s own insider trading policy, blackout windows, or the rules that apply to employees who may hold material information. An employee whose shares are technically released may still be restricted from selling by the company’s own policy. Eligible, but not yet able.

Eligible does not mean sold. A tranche releasing a percentage of shares does not mean those shares hit the market. It means they can. How many holders actually sell, and when, is a separate question. The release sets the ceiling on possible selling, not the floor.

The conditions can move the schedule. Because one tranche depends on the stock trading above a threshold, the actual path of releases is not fixed in advance. The schedule you can draw on day one is a projection that resolves into a real path only as the conditions are met or missed.

This is why I am always cautious with a clean unlock chart, including the one in this very post. A chart like that is a useful map, and it gets the structure right. But the line it draws is one scenario among several, and the difference between “eligible to transfer” and “sold and settled in your account” is exactly the kind of detail that the structure, not the headline, controls. The holder who understands that distinction plans around it. The holder who does not assumes the IPO was the finish line and gets surprised by how long the real exit takes.

5. What this teaches you about your own exit

You are probably not holding SpaceX. But if you hold any private position, the SpaceX lock-up is teaching you something directly about your own eventual exit, and it is worth taking the lesson now rather than at your own liquidity event.

→ The IPO is the start of your exit, not the end of it. Build your expectations around a staggered, conditional release that can run six months or more, not around a single day when everything becomes cash. The biggest IPO in history just showed that even there, the holders waited.

→ Read the lock-up terms before you enter, not after. The structure of your eventual exit is written into the documents you sign on the way in. A staggered schedule, a performance condition, a longer restriction for certain holders. All of it is decided early. By the time you want to sell, the terms are fixed. This is the same point as the diligence checklist: the exit structure is part of what you are buying, and it deserves the same scrutiny as the business.

→ Distinguish eligible from liquid in your own planning. When you model when a position turns into money you can use, do not model the release date. Model the date you can actually sell, after trading policies, conditions, and the practical mechanics of finding a buyer. Those are different dates, and the difference can be months.

→ The largest holders often wait the longest. Concentrated or insider positions frequently carry the longest restrictions, as Musk’s 366-day lock-up illustrates. If you hold a large or specially situated position, your timeline may be longer than the headline lock-up suggests.

The SpaceX lock-up is, in effect, a free and very public lesson in exit structure, delivered at the largest scale the market has ever seen. The smart move is to study it now, while it is someone else’s exit, and apply it to how you evaluate your own positions.

6. The structure was always the story

The valuation got the headline because the valuation is a big, simple number, and big simple numbers travel. But a valuation is a snapshot of what the company is worth on one day. The lock-up is the mechanism that decides when any of that worth becomes real for the people who hold the shares. For a pre-IPO investor, the second thing is the one that touches your life.

Three things to take from this:

→ Study the lock-up, not just the valuation. The valuation is what the company is worth. The lock-up is when you can act on it. The second number is the one you live with.

→ The exit structure is decided on the way in. SpaceX’s staggered release, its performance kicker, and its longer restriction for the largest holder were all written into the filing before the stock ever traded. Your own exit terms work the same way. Read them early.

→ Eligible is not liquid. The single most useful habit in pre-IPO investing is to stop treating a release date as a payday and start treating it as the beginning of a process. The gap between the two is where the surprises live, and closing it is just a matter of reading the structure honestly.

The interesting question was never whether SpaceX was worth $1.75 trillion. In my view, that framing is the distraction. The interesting question, the one that actually teaches you something about your own positions, is how the people who held the shares get to the money, and on what schedule, and under what conditions. The discipline is in studying the structure that answers that question, not the headline that ignores it.

Curious what this means for pre-IPO investing?

Book a 20-minute call. We can walk through how exit structure and lock-up terms should factor into the way you evaluate a private position.

Book a call

RISK FACTOR LANGUAGE

STANDARD RMS DISCLOSURE LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

AI is not coming for the pre-IPO broker. It is coming for the broker's analyst

Mariano Grosman — Mon, 15 Jun 2026 14:03:01 GMT

Hey everybody,

The narrative I hear most often about AI and my industry is some version of “AI is going to replace pre-IPO brokers.” It gets repeated at conferences, in articles, in nervous conversations between brokers who have watched what happened to travel agents and stock-trading desks.

In my view, it is the wrong unit of analysis.

AI is not coming for the broker. The broker’s job is relationships, judgment, access, and the out-of-data signals I have written about in earlier posts. Those are exactly the things AI cannot do. If anything, the broker who uses AI well becomes more valuable, not less.

What AI is coming for is a different role entirely. The role inside the brokerage that does the structural analysis, the comparable sourcing, the document review, the first-pass diligence. The analyst role. That work is exactly what AI does well, and in my experience it is being compressed faster than almost anyone in the industry is willing to say out loud.

This distinction matters because it changes how a brokerage should be built over the next several years. The firms that understand which role is at risk are restructuring now. The firms that think “AI replaces brokers” is the whole story are preparing for the wrong change.

→ Why “AI replaces brokers” is the wrong frame

→ What the analyst role actually does, and why AI does it well

→ What this means for how a brokerage is staffed and structured

→ Why this is good for allocators, not bad

→ What it means for anyone building a career in this industry

By the end of this post, you will understand why the role at risk is not the one in the headlines, and what the change actually means for brokerages, allocators, and careers.

Here is what is inside.

Why “AI replaces brokers” is the wrong frame
What the analyst role actually does
Why AI does that work well
How this changes the way a brokerage is built
Why this is good for allocators
What it means for a career in this industry

Let’s begin.

The references and observations behind this post:

→ My own experience running deal flow and the analytical work behind it over 11 years

→ Direct observation of how the structural-analysis workload has changed since AI tools became capable

→ Claude and other AI tools that now do work that used to require a dedicated analyst

→ Conversations with other brokers about how they are restructuring their teams

A note on figures: this post is observational and structural, drawn from my own experience. Where I describe how the work has changed, it is my account, framed as such, not a measured study.

1. Why “AI replaces brokers” is the wrong frame

When people say AI will replace brokers, they are usually picturing the broker as a single undifferentiated job. In reality, the work a brokerage does splits into two very different kinds of tasks, and AI affects them in opposite directions.

The first kind is relationship and judgment work. Knowing the founders. Reading the seller’s motivation. Sensing when a sector is about to turn. Deciding which structural risk actually matters in a specific situation. Maintaining the trust that makes an allocator pick up the phone. This is the part of the job that took me 11 years to build, and in my view AI does not touch it.

The second kind is analytical work. Reconstructing the cap table. Building the comparable set. Reviewing the documents. Producing the first-pass structural analysis. Pulling the data together into something a decision can be made from. This is real, skilled work, and historically it required people. It is also exactly what AI is good at.

When you split the job this way, “AI replaces brokers” stops making sense. AI is not replacing the broker. It is replacing, or at least dramatically compressing, the analytical layer that used to sit underneath the broker. The broker on top becomes more leveraged. The analyst underneath becomes less necessary.

The headline is not wrong because AI is weak. It is wrong because it points at the wrong role.

2. What the analyst role actually does

To see why this matters, it helps to be precise about what the analytical layer in a pre-IPO brokerage actually does. In my experience, it is four things.

It reconstructs the cap table. Taking the raw documents and building a clear picture of the preferred classes, the liquidation preferences, the SPV layers, and what a given position actually owns. Skilled, slow, detail-intensive work.

It builds the comparable set. Finding the companies that have exited in the relevant window, pulling their valuations and outcomes, and assembling the comp analysis that anchors a pricing view.

It reviews the documents. Reading the operating agreement, the subscription documents, the transfer terms, and pulling out what matters. The kind of work where a missed detail on page 40 changes the answer.

It produces the first-pass analysis. Taking all of the above and turning it into a structured starting point that the senior broker can then apply judgment to.

This is the work that, in a traditional brokerage, justified a team of analysts underneath the senior people. It is genuinely valuable work. It is also, every single piece of it, the kind of work AI now does at a fraction of the time and cost.

That is the uncomfortable part. The analyst role was not low-skill. It was the training ground where the next generation of brokers learned the craft. And it is exactly the role most exposed to AI compression.

3. Why AI does that work well

The reason AI is so well-suited to the analytical layer is the same reason I found in my own experiment running deal flow through an AI agent. AI is excellent at the parts of the work that depend on rigor, consistency, and breadth, and the analyst role is almost entirely rigor, consistency, and breadth.

Cap table reconstruction rewards consistency. AI does not get tired on the fortieth cap table of the month. It applies the same care to the last document as the first. In my experience, that consistency catches things a fatigued human analyst misses.

Comparable sourcing rewards breadth. AI can surface more comparable companies, faster, than a human analyst working manually. The breadth of the comp set goes up, and the time to build it goes down.

Document review rewards thoroughness. AI reads every page with the same attention. The page-40 detail that a human might skim past gets the same treatment as page 1.

First-pass analysis rewards speed. The structured starting point that used to take an analyst days now takes the AI a fraction of the time, leaving the senior broker more time for the judgment layer.

None of this means AI replaces the judgment. It means AI replaces the preparation that fed the judgment. And since the analyst role was mostly preparation, the analyst role is what gets compressed.

This is not a knock on analysts. It is a statement about which kind of work AI does well, and the analyst role happens to sit right in the center of it.

4. How this changes the way a brokerage is built

If the analytical layer compresses, the structure of a brokerage changes. In my view, here is how.

The pyramid flattens. The traditional brokerage was a pyramid: a few senior brokers on top, a larger base of analysts underneath doing the preparation. When AI does the preparation, the base shrinks. The firm needs fewer analysts to support the same number of senior brokers, because each senior broker can now lean on AI for the analytical layer.

The senior broker gets more leveraged. A senior broker who used to need three analysts to support their deal flow can now support more deal flow with AI handling the preparation. The leverage per senior person goes up. This is the same dynamic I wrote about in the live-pricing and compliance posts. The capable individual covers more ground.

The training path has to be rebuilt. This is the part the industry is not talking about enough. The analyst role was where brokers learned the craft. You did the cap table reconstruction for years, and in doing it you learned to see the structures. If AI does that work, the next generation does not get that training the same way. In my view, brokerages that want to develop talent are going to have to deliberately rebuild a training path that the old analyst role used to provide automatically.

The hiring profile shifts. The firm needs fewer people who are good at the preparation and more people who are good at the judgment, the relationships, and using the AI tools well. That is a different hiring profile than the industry has optimized for.

In my view, the firms that restructure around this early will be leaner, faster, and more leveraged than the firms still staffing the old pyramid. The change is not coming for the broker on top. It is coming for the shape of everything underneath.

5. Why this is good for allocators

It would be easy to read all of this as bad news. In my view, for allocators it is the opposite.

→ The analysis gets more consistent. AI does not have bad days. The structural analysis underneath a recommendation is more thorough and more consistent than a stretched analyst team could produce. In my experience, that means fewer things slip through.

→ The senior broker has more time for judgment. When AI handles the preparation, the senior broker spends more time on the part of the work that actually serves the allocator: the judgment, the out-of-data signals, the correction of what the analysis missed. The allocator gets more of the broker’s most valuable attention.

→ The cost structure improves. A leaner brokerage with more leverage per senior person can, in principle, serve allocators more efficiently. In my view, that efficiency eventually shows up in how well allocators are served.

The allocator was never paying for the analyst layer. They were paying for the judgment on top of it. AI compressing the analyst layer means more of what the allocator pays for goes to the judgment they actually wanted.

6. What it means for a career in this industry

If you are building a career in pre-IPO, this changes the path, and in my view it is worth being honest about how.

→ Do not build your value on the analytical layer alone. The cap table reconstruction, the comp sourcing, the document review. These were the traditional entry points, and they are exactly what AI compresses. Being excellent at only these is, in my view, building on ground that is shifting.

→ Build the judgment and the relationships as early as you can. The parts of the job AI cannot do are the parts that compound into a durable career. The sooner you move toward judgment, relationships, and the out-of-data signals, the more durable your position.

→ Become the person who uses the AI tools best. There is a real and growing edge in being the broker who understands exactly where the AI tools break and how to direct them. In my view, that skill is becoming as valuable as the analytical skill it replaces.

The career path that ran through years of analytical apprenticeship is changing. In my view, the people who recognize that early and orient toward judgment and relationships will be fine. The people who assume the old path still exists may find the bottom rungs of the ladder have been quietly removed.

The interesting question is not whether AI changes my industry. In my view it already has. The question is whether you are watching the right role. The headline points at the broker. The change is happening underneath.

Curious how this works?

Book a 20-minute call. I am happy to walk through how the market and the work are changing.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

How AI is changing the broker-client conversation, and what the new conversation actually sounds like

Mariano Grosman — Fri, 12 Jun 2026 10:30:43 GMT

Hey everybody, welcome to Post 09.

A few years ago, the typical first call with a new allocator went one direction. They asked, I answered. What is the company. What is the valuation. What is the structure. What is the comparable set. I was the source of the information, and the conversation was shaped around me providing it.

That call sounds completely different now.

Today, a new allocator shows up to the first call having already run the company through three AI tools. They have a summary. They have a comparable set. They have a rough valuation range. They have a list of questions the AI generated for them. They are not coming to me for the information anymore. They are coming to me to find out what the information means and what they should not trust about it.

The shift happened quietly, over roughly 18 months, and in my experience it has changed what the job actually is. The broker who has not noticed the change is still answering questions nobody is asking. The broker who has noticed it is having a completely different, and much more valuable, conversation.

→ What the old conversation sounded like

→ What the new conversation sounds like

→ The three questions allocators ask now that they never asked before

→ Why this shift favors brokers who adapt and punishes brokers who do not

→ What it means for how allocators should use their broker going forward

By the end of this post, you will recognize whether your own broker conversations have made this shift, and what the high-value version of the conversation actually sounds like.

Here is what is inside.

What the old conversation sounded like
What the new conversation sounds like
The three questions allocators ask now
Why the shift favors brokers who adapt
What it means for how you use your broker
The conversation worth having

Let’s begin.

The references and observations behind this post:

→ My own client conversations across the US and Latin America over the past several years

→ Direct experience of how first calls with new allocators have changed

→ Claude and other AI tools that allocators now routinely use before they ever call a broker

→ Conversations with other brokers about how their own client interactions have shifted

A note on figures: this post is entirely observational, drawn from my own experience. Where I describe how conversations have changed, it is my account of what I have seen, framed as such, not a measured study.

1. What the old conversation sounded like

For most of my career, the broker held an information advantage, and the conversation was built around it.

An allocator interested in a pre-IPO name would come to me because I had access they did not. I knew where the last secondary cleared. I had the structural detail on the cap table. I could source the comparable set. I knew which deals were real and which were noise. The information lived with the broker, and the allocator came to the broker to get it.

The old first call sounded like a briefing. The allocator asked the foundational questions, and I provided the foundational answers. What does the company do. What is the most recent valuation. How is the cap table structured. What are the lockup terms. Who else is in the deal. The value I provided was the information itself.

That was a good business for a long time. The problem is that the foundational information is exactly what AI is now best at providing. The briefing I used to deliver is the briefing an allocator can now generate themselves in twenty minutes before they ever pick up the phone.

The old conversation has not become less friendly. It has become less necessary.

2. What the new conversation sounds like

The new first call sounds nothing like a briefing. It sounds like a second opinion.

The allocator arrives having already done the foundational work with AI. They have the summary. They have the comparable set. They have a valuation range. What they want from me is not the information. It is the interpretation, the correction, and the judgment about what the AI got wrong.

In my experience, the new conversation opens in a completely different place. Instead of “what is this company,” the allocator says “here is what I found, here is what I think, tell me where I am wrong.” The starting point has moved from zero to the middle of the analysis. We skip the briefing entirely and go straight to the hard part.

This is a better conversation. It is also a more demanding one. When the allocator has already done the foundational work, the broker cannot add value by repeating it. The broker has to add value above it. The judgment about which structural risk actually matters. The correction of the comparable the AI chose badly. The signal the AI could not see because it was not in the data. The reason the valuation range the AI produced is too optimistic.

In my experience, the brokers who thrive in the new conversation are the ones who treat the allocator’s AI-generated analysis as the starting point and immediately go to work on what it missed. The brokers who struggle are the ones still trying to deliver the briefing the allocator already has.

3. The three questions allocators ask now

The clearest sign of the shift is in the specific questions allocators ask. Three questions come up constantly now that almost never came up before.

Question 1: “Where is my AI analysis wrong?”

This is the single most common new question. The allocator has run the company through an AI tool and produced an analysis. They do not fully trust it, and they are right not to. They come to me to find the errors. In my experience, this is one of the most valuable things I do now. I know where AI analysis tends to break on pre-IPO names, and I can find the specific errors fast.

Question 2: “What can you see that the AI cannot?”

Allocators have figured out that AI is excellent at the parts of the analysis that live in the data and poor at the parts that do not. So they ask directly about the out-of-data signals. What do I know about the founder. What is the real story behind the last round. What is the seller’s actual motivation. These are the questions where the broker relationship still holds all the value, and allocators now ask them explicitly.

Question 3: “Is this the right framework, or is the AI using the wrong one?”

The more sophisticated allocators have realized that AI often applies a public-market framework to a private-market problem, the exact issue I wrote about in an earlier post. They come to me asking whether the framework their AI used is the right one for this specific situation. This is a higher-order question than anything I used to get, and it reflects how much the baseline has risen.

These three questions have one thing in common. None of them are answerable by the AI the allocator already used. They are all questions about the limits of that AI, and that is precisely where the broker now lives.

4. Why the shift favors brokers who adapt

This change is not neutral. In my view, it sharply favors one kind of broker and punishes another.

It favors the broker who knows where AI breaks.

If the allocator’s first question is now “where is my AI analysis wrong,” then the broker’s most valuable knowledge is a precise understanding of where AI analysis breaks on pre-IPO names. The broker who has done the work to understand the limits of these tools, who has tested them against real deals, can answer that question with specificity. In my experience, that is the single most useful thing I can offer a new allocator now.

It favors the broker who leans into the out-of-data signals.

The questions allocators ask now are concentrated in exactly the area AI cannot reach. The founder relationships, the seller motivations, the sector timing, the deal momentum. The broker who has invested in those relationships and that judgment has more to offer in the new conversation than ever, because the allocator now asks for exactly that and nothing else.

It punishes the broker whose only value was the information.

If a broker’s entire value proposition was access to information the allocator could not get, that broker is in trouble. The allocator can now get a usable version of that information themselves. The broker who has not moved up the value chain to judgment and out-of-data signal is competing with a free tool, and losing.

In my view, this is the quiet sorting happening in the industry right now. The conversation changed, and it revealed which brokers were adding judgment and which were just relaying information. The ones adding judgment are having better conversations than ever. The ones relaying information are watching their calls get shorter.

5. What it means for how you use your broker

If you are an allocator, the shift has a direct implication for how you should be using your broker relationships.

→ Stop using your broker for the foundational briefing. You can get that from AI faster. Using a broker call to ask what the company does is, in my view, a waste of the most valuable resource in the relationship.

→ Start using your broker for the three questions AI cannot answer. Bring your AI analysis to the call and ask where it is wrong. Ask what the broker can see that the AI cannot. Ask whether the framework is right. These are the questions that justify the relationship.

→ Judge your broker on the quality of the correction, not the quality of the briefing. The broker who can take your AI analysis and immediately show you the three things it got wrong is worth far more than the broker who can recite the same summary your AI already gave you. In my view, that is the test of a broker worth keeping in the new environment.

The best way to use a broker in 2026 is to do the foundational work yourself with AI, then bring it to the broker for the judgment layer. The allocators who use the relationship that way are getting far more out of it than the ones still asking for the briefing.

6. The conversation worth having

The broker-client conversation changed quietly, but in my view it changed permanently. The information advantage that defined the relationship for decades is gone, and what replaced it is more demanding for the broker and more valuable for the allocator who knows how to use it.

Three things to take from this:

→ If you are an allocator, notice whether your broker conversations have made the shift. In my view, if your broker is still delivering briefings you could generate yourself, you are not getting the value the relationship should provide.

→ The high-value conversation starts where the AI analysis ends. Bring the AI work to the table and spend the call on what it missed. That is where the judgment lives, and judgment is what you are actually paying for.

→ The relationship is worth more now, not less, if you use it correctly. AI did not make brokers obsolete. In my experience, it made the good ones more valuable by stripping away the part of the job that was never really the point.

The interesting question is not whether AI changed the broker-client conversation. In my view it already has, completely. The question is whether you have updated how you use the relationship to match. The discipline is in spending the conversation on the part that AI cannot touch.

Want to learn more?

Book a 20-minute call. Bring whatever analysis you have already done, and we can go straight to the part that matters.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

Why AI is going to change pre-IPO access in Latin America before it changes anything in the US

Mariano Grosman — Wed, 10 Jun 2026 14:01:54 GMT

Hey everybody,

There is a quiet assumption in most conversations about AI and private markets: that the US, with the most developed pre-IPO infrastructure, is where AI will have the biggest impact. In my view, that assumption is backwards.

The US pre-IPO market is already efficient. Not perfectly efficient, but efficient enough that AI mostly compresses time. It makes a fast process faster. That is valuable, but it is incremental.

Latin America is a different story. The pre-IPO market across the region is so structurally inefficient that AI does not just compress time. It changes who can access the market at all. That is not incremental. That is a different category of change.

I work both sides of this. My book includes allocators across the US and across Latin America, and the contrast between how AI lands in each market is, in my experience, one of the most underappreciated dynamics in private markets right now.

→ Why the US market only gets incremental change from AI

→ The three structural inefficiencies that define Latin American pre-IPO access

→ Why AI removes those specific barriers in a way nothing else has

→ What this means for allocators on both sides of the border

→ Why I believe the next decade of LatAm pre-IPO growth comes from AI-enabled brokers, not new brokers

By the end of this post, you will understand why I believe the most important AI-driven change in pre-IPO access is going to happen in Latin America first, and what that means whether you allocate from New York or from São Paulo.

Here is what is inside.

Why the US market only gets incremental change
The three structural inefficiencies in Latin American pre-IPO
Why AI removes those specific barriers
What this means for allocators on both sides
Why the next decade belongs to AI-enabled brokers
The positioning question for the next 24 months

Let’s begin.

The references and observations behind this post:

→ My own book of allocator relationships across the US and Latin America

→ Direct experience of how pre-IPO transactions actually get sourced and closed across both markets

→ Public information on AI translation and language-model capability improvements over the past 18 months

→ Claude for pressure-testing the asymmetry argument

→ Conversations with family offices and allocators in multiple Latin American countries

A note on figures: this post makes structural and directional claims drawn from my own experience working both markets. Where I describe a trend, it is my observation, framed as such, not a published statistic.

1. Why the US market only gets incremental change

The US pre-IPO market has, over the past two decades, built real infrastructure. Specialized brokers. Secondary marketplaces. Research providers. A dense network of allocators who already know how to evaluate private positions. Legal and compliance frameworks that, while complex, are well-established.

In a market like that, what does AI actually change? In my view, it compresses time. The analysis that took two weeks now takes two hours. The diligence that required a team now requires a person with the right tools. The comparable-company sourcing that was slow is now fast.

That is genuinely valuable. I wrote a whole post about running my own deal flow through an AI pressure-test. But notice what it is: it is making an already-functional process more efficient. The allocator who could access the US pre-IPO market before AI can still access it after AI. They just do it faster.

The barriers to access in the US were never really about information processing speed. They were about relationships, capital, and qualification. AI does not remove those barriers. It optimizes the work that happens once you are already inside.

That is why I call the US change incremental. Real, but incremental.

2. The three structural inefficiencies in Latin American pre-IPO

Latin America is different because the barriers are different. In my experience working with allocators across the region, three structural inefficiencies define the market, and all three are the kind of barrier that AI is uniquely suited to remove.

Inefficiency 1: The language barrier.

The overwhelming majority of high-quality institutional pre-IPO research is published in English. The deal documentation is in English. The comparable-company analysis, the sector research, the structural analysis. All English.

A Spanish-speaking or Portuguese-speaking allocator in the region has historically faced a real disadvantage. Not because they lack sophistication, but because the entire information layer of the market is in a language that is not their first. In my experience, this slows everything down and filters out allocators who would otherwise be active.

Inefficiency 2: The coverage barrier.

In the US, a pre-IPO broker can specialize because the market is deep enough to support specialization. In Latin America, the market is spread across many countries, each with its own regulatory environment, its own currency considerations, its own allocator base.

Historically, that fragmentation meant a single broker could realistically serve allocators in only one or two countries well. Serving allocators across six countries with the same quality of analysis was operationally impossible for one person. The market was too fragmented and the work did not scale.

Inefficiency 3: The information-access barrier.

In the US, an allocator has access to research providers, broker networks, and data platforms as a matter of course. Across much of Latin America, that infrastructure is thinner. The allocator who wants to evaluate a US pre-IPO position has historically had fewer tools, fewer sources, and less institutional support to do it well.

These three inefficiencies compound. Language slows the information. Fragmentation limits the coverage. Thin infrastructure limits the tools. The result, in my view, is a market where serious capital exists but the access to deploy it into pre-IPO has been structurally constrained.

3. Why AI removes those specific barriers

Here is why this matters. AI is unusually well-suited to remove exactly these three barriers, in a way that no previous technology was.

AI removes the language barrier.

This is the big one. In my experience, AI translation crossed a quality threshold over the past 18 months. It is now good enough that a Spanish-speaking or Portuguese-speaking allocator can work with English-language pre-IPO research at close to the same speed and comprehension as a native English speaker. The entire English-language information layer of the market just became accessible to a much wider allocator base.

That is not an incremental change. That is a barrier that defined the market for decades, removed.

AI removes the coverage barrier.

The reason a single broker could only serve one or two countries was that the work did not scale. The analysis, the document review, the comparable sourcing, all of it was manual and time-bound. With AI handling the structural preparation, a single broker can now realistically support allocators across many countries with the same quality of analysis. The work scales in a way it never did before.

In my own practice, this is the change I feel most directly. The operational ceiling that used to limit how many markets I could serve well has moved.

AI removes the information-access barrier.

The allocator in a market with thin research infrastructure now has access to the same analytical capability as the allocator in New York, because the capability lives in the tool, not in the local infrastructure. AI does not care whether you are sitting in Manhattan or in Bogotá. The analytical floor just rose everywhere at once.

Stack these three together and you get the asymmetry at the heart of this post. In the US, AI optimizes a market that already worked. In Latin America, AI removes the specific barriers that kept serious capital out. Same technology, completely different magnitude of effect.

4. What this means for allocators on both sides

The implication is different depending on where you sit. Both are worth naming.

For Latin American allocators:

→ The information disadvantage you have lived with is closing fast. In my view, the English-language research layer is now accessible in a way it was not two years ago. The allocators who internalize this early will be active in pre-IPO before their peers.

→ Do not wait for local infrastructure to catch up. The analytical capability is available now through the tools, regardless of what exists locally. In my experience, the allocators who move first on this will have a multi-year head start.

→ The quality bar has not dropped. AI removes the access barrier, but the structural diligence still matters as much as it did. The frameworks I have written about in earlier posts apply just as much to an allocator in Santiago as to one in San Francisco.

For US allocators:

→ The pool of capital competing for the best pre-IPO positions is about to widen. In my view, allocators who were structurally locked out of the market are entering it. That changes the competitive dynamics for everyone.

→ The geographic edge that US allocators had, simply by being closer to the infrastructure, is eroding. The edge that remains is relationships and judgment, not access. In my view, that is worth recognizing now rather than later.

→ Cross-border deal flow becomes more viable. As the access barrier falls, the opportunity to transact across the border in both directions grows. The allocator who builds relationships across the region early benefits from that widening.

5. Why the next decade belongs to AI-enabled brokers

Here is the thesis stated plainly, as my own view. The next decade of pre-IPO growth in Latin America is not going to come from a wave of new brokers entering the market. It is going to come from AI-enabled brokers who can now serve the region in a way that was operationally impossible before.

The reasoning, in my assessment:

The barrier to serving Latin American pre-IPO allocators well was never demand. The capital exists. The interest in US pre-IPO exposure exists. The barrier was the operational impossibility of one broker covering a fragmented, multi-language, multi-country market at high quality.

AI removes that operational impossibility. A broker who can now handle the language layer, scale the analysis across countries, and bring institutional-grade tools to allocators regardless of local infrastructure can serve the region in a way that simply was not feasible before.

In my view, that broker captures a market that has been underserved for structural reasons, not for lack of demand. And because the advantage compounds, relationships built now in a market that is just opening compound for years.

This is the opportunity I am most focused on, and it is the reason I think the LatAm pre-IPO story over the next decade is going to surprise people who are only watching the US.

6. The positioning question for the next 24 months

The asymmetry between how AI lands in the US versus Latin America is, in my view, one of the most underappreciated dynamics in private markets. It is also one that rewards moving early.

Three things to consider now:

→ If you allocate from Latin America, start treating the English-language research layer as accessible. In my view, the language barrier that constrained you is closing, and the allocators who act on that early will be ahead.

→ If you allocate from the US, recognize that the competitive pool is widening. The geographic access edge is eroding. The edge that lasts is relationships and judgment. In my view, that is where to invest your attention.

→ On both sides, the relationships built now in a just-opening market compound. The broker, the allocator, the family office that builds cross-border relationships while the market is opening is positioned for a decade, not a quarter.

The interesting question is not whether AI will change pre-IPO access. In my view it already is. The question is where it changes access first, and my answer is the place almost nobody is watching. The discipline is in positioning where the change is largest, not where it is loudest.

Want to talk it through?

Book a 20-minute call. Whether you allocate from the US or from Latin America, I am happy to walk through how the market is changing.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

The compliance bottleneck is the next pre-IPO opportunity

Mariano Grosman — Tue, 09 Jun 2026 16:02:25 GMT

Hey everybody, welcome to Post 07.

Ask any allocator what the most frustrating part of a pre-IPO transaction is, and almost nobody says the analysis. They say the wait.

The wait between agreeing on a deal and actually closing it. The compliance review. The buyer qualification. The source-of-funds verification. The transfer approval. The escrow setup. In my experience, this is where pre-IPO deals slow to a crawl, and it is the part of the process nobody writes about because it is unglamorous.

Here is the thesis I want to lay out: the compliance bottleneck is not just an operational annoyance. It is the single biggest structural inefficiency left in the pre-IPO market, and AI is starting to collapse it. In my view, the firms that move first on AI-assisted compliance are going to reshape who wins the secondary market over the next several years.

This is a forecast and an argument, not a certainty. I will lay out the reasoning so you can decide whether you agree.

→ Why compliance is the real bottleneck, not analysis

→ What AI actually compresses in the compliance workflow

→ Why faster compliance changes the competitive structure of the market

→ Where the risk in AI-assisted compliance actually sits

→ What allocators and brokers should watch over the next 24 months

By the end of this post, you will understand why I believe the compliance layer, not the analysis layer, is where the next pre-IPO advantage gets built.

Here is what is inside.

Why compliance is the real bottleneck
What AI actually compresses in the workflow
Why faster compliance reshapes the market
Where the risk actually sits
What this means for allocators and brokers
The 24-month positioning question

Let’s begin.

The references and observations behind this post:

→ My own broker book and direct experience of where pre-IPO transactions slow down

→ The published regulatory frameworks that govern private securities transfers (Reg D, Reg A+, Reg CF), available at sec.gov

→ The FINRA rules governing broker-dealer suitability and anti-money-laundering obligations, available at finra.org

→ Claude for pressure-testing the structural argument

→ Conversations with compliance officers and platform operators building in this space

A note on figures: this post deliberately avoids putting specific time-savings percentages or dollar figures on the compliance compression, because I do not have a verifiable published source for those numbers. Where I describe the direction of the change, it is my observation from my own practice, framed as such.

1. Why compliance is the real bottleneck

Most of the conversation about AI in pre-IPO focuses on the analysis. Can AI evaluate a deal, model a cap table, source comparables. I wrote a whole post about that. But in my experience, analysis was never the slow part. A sharp broker with the right tools can analyze a deal quickly. The slow part comes after the decision to proceed.

Here is what actually happens between yes and closed.

Buyer qualification. The selling party and the broker-dealer have to verify that the buyer is accredited or qualified under the applicable exemption. This is a regulatory requirement, not a formality.

Source-of-funds verification. Anti-money-laundering obligations require verifying where the buyer’s capital is coming from. For a family office or an institutional buyer with a complex structure, this can take real time.

Transfer approval. The company whose shares are being transferred usually has to approve the transfer, and may have right-of-first-refusal rights that have to be cleared first.

Documentation and escrow. The purchase agreement, the transfer documentation, and the escrow arrangements all have to be prepared, reviewed, and executed.

In my experience, each of these steps is sequential, each involves multiple parties, and each is governed by real regulatory obligations under FINRA and SEC rules. None of them can be skipped. The result is that the compliance phase, not the analysis phase, is where pre-IPO deals lose the most time.

This is the bottleneck. And in my view, a bottleneck that everyone tolerates is exactly where the next opportunity hides.

2. What AI actually compresses in the workflow

Let me be precise about what AI does and does not do here, because the distinction matters and because overclaiming on compliance automation is genuinely dangerous.

AI does not replace the compliance judgment. A registered compliance officer still has to make the final determinations, and the regulatory obligations do not go away because a tool got faster. What AI compresses is the preparation, the document review, and the information gathering that feeds the compliance decision.

In my experience, here is where the compression actually happens.

Document review. Pulling the relevant terms out of an operating agreement, a subscription document, or a transfer agreement is something AI does quickly. What used to require a person reading every page can be compressed to a first-pass extraction that a human then verifies.

Information gathering for buyer qualification. Assembling the documentation that supports an accreditation determination is faster with AI handling the collection and organization, with a human making the actual determination.

Consistency checking. AI is good at catching inconsistencies across documents. A name that does not match, a figure that does not reconcile, a date that conflicts. In my experience, these inconsistencies are exactly what slow down a compliance review when they are caught late, and catching them early speeds everything up.

Drafting first-pass documentation. The standard transaction documents have predictable structures. AI can produce a first draft that a compliance professional then reviews and corrects, rather than starting from a blank page.

The pattern here is the same one I found in my own analysis workflow. AI compresses the preparation and the consistency work. The judgment stays human. But because the preparation was the slow part, compressing it changes the total timeline meaningfully.

3. Why faster compliance reshapes the market

This is the part of the thesis that matters for allocators. Faster compliance is not just a convenience. In my view, it changes the competitive structure of the entire secondary market.

Here is the reasoning.

Faster compliance means more transactions can clear. When the compliance phase compresses, the same broker-dealer can process more transactions in the same period. Throughput goes up.

Higher throughput means the firms that move first capture more volume. In a market where deals are abundant but closing capacity is the constraint, the firm that can close faster wins more business. This is the same dynamic I wrote about in the live-pricing post. Execution speed becomes the competitive edge.

More volume means better data, which means a compounding advantage. A firm processing more transactions sees more of the market, which improves its pricing intelligence, which wins it more business. The advantage compounds.

In my view, this is how the next reshaping of the pre-IPO industry happens. Not through better analysis, which is being commoditized for everyone. Through faster, AI-assisted compliance and execution, which only some firms will build well. The firms that treat compliance as a competitive capability rather than a cost center are the ones I expect to win the next cycle.

The uncomfortable implication for incumbents: a firm whose compliance process still takes weeks is going to lose volume to a firm whose process takes days, even if their analysis is identical. Compliance speed, historically the least glamorous part of the business, becomes a primary competitive axis.

4. Where the risk actually sits

I want to spend real time here, because a post that argues for AI in compliance without addressing the risk would be irresponsible. Compliance exists for a reason, and the risk of getting this wrong is not theoretical.

In my view, here is where the genuine risk sits.

The risk is over-automation, not under-automation. The temptation, once AI makes compliance preparation fast, is to let the tool make determinations it should not make. The accreditation determination, the suitability judgment, the AML conclusion. These are regulatory obligations that require human judgment and accountability. A firm that lets AI make these calls is not moving fast. It is creating regulatory exposure that will eventually surface.

The audit trail has to be human-verifiable. Anything AI produces in a compliance context has to be reviewable by a human compliance officer and defensible to a regulator. If the process cannot be audited and explained, it does not matter how fast it is. The speed is worthless if the work cannot withstand examination.

The regulatory obligations do not change because the tool got faster. FINRA and SEC rules governing private securities transactions apply regardless of how the work gets done. AI is a tool for meeting those obligations more efficiently, not a reason to meet them less rigorously. In my view, any firm that frames AI compliance as a way to do less compliance has misunderstood the entire opportunity.

The firms that get this right, in my assessment, are the ones that use AI to make rigorous compliance faster, not to make compliance lighter. That distinction is the whole game. The opportunity is speed with the same rigor, not speed instead of rigor.

5. What this means for allocators and brokers

The implication is different for each side. Here is how I would think about it.

For allocators:

→ When you evaluate a broker, ask about their closing timeline, not just their analysis. In my view, the broker who can close cleanly in days rather than weeks is signaling an operational capability that matters more every year.

→ Do not mistake speed for corner-cutting, and do not mistake slowness for rigor. A slow compliance process is not inherently more thorough. Ask how the broker maintains rigor while moving quickly. The good answer involves AI handling preparation while humans hold judgment.

→ Watch which platforms and firms are investing in this capability. In my view, they are the ones positioned to win the next cycle, and the allocators who build relationships with them early will benefit.

For brokers:

→ Treat compliance as a competitive capability, not a cost center. In my assessment, this is the reframe that separates the firms that win the next cycle from the ones that lose volume.

→ Invest in AI-assisted compliance preparation while keeping human judgment and accountability firmly in place. The firms that get the balance right move fast and stay defensible. The firms that over-automate create exposure. The firms that under-invest lose on speed.

→ Build the audit trail into the workflow from the start. In my view, a fast compliance process that cannot be examined is a liability waiting to surface, not an advantage.

6. The 24-month positioning question

The compliance bottleneck has been tolerated for so long that most participants treat it as a fixed cost of doing business. In my view, that assumption is about to break, and the firms that see it first will have an advantage that compounds.

Three things to consider now:

→ Map where your own transactions actually slow down. In my experience, most brokers and allocators have never measured this precisely. The honest measurement usually reveals that compliance and closing, not analysis, is the real drag.

→ Distinguish the parts of compliance that require human judgment from the parts that are preparation. The preparation is where AI helps. The judgment stays human. Knowing the difference is what lets a firm move fast without creating exposure.

→ Watch the firms treating compliance speed as a competitive edge. In my assessment, they are building the capability that defines the next cycle. Whether you are an allocator choosing a broker or a broker deciding where to invest, this is the capability to track.

The interesting question is not whether AI will speed up compliance. In my view it already is. The question is which firms will use that speed to win, and whether they will do it while keeping the rigor that compliance exists to protect. The discipline is in moving fast without ever cutting the corner that matters.

New to pre-IPO investing?

Book a 20-minute call. I am happy to walk through how the market works, no jargon, no pressure.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

Anthropic's secondary market is structurally different from OpenAI's

Mariano Grosman — Fri, 05 Jun 2026 17:04:19 GMT

Most allocator conversations I have about frontier AI exposure treat Anthropic and OpenAI as substitutes for each other. Same market opportunity. Same competitive dynamic. Same kind of position to hold.

The structural mechanics of the two secondary markets tell a very different story.

The two companies have different cap table architectures, different transfer regimes, different corporate purpose constructs, and different strategic-investor concentrations. Each one of those differences changes what an allocator actually owns when they take a position. Stack the differences together and you get two markets that look similar from a distance and behave nothing alike up close.

This post walks through the structural lens on both names, side by side, so allocators can decide whether they want frontier AI exposure broadly or one specific name for a specific reason.

→ Why the corporate purpose construct matters more than most allocators realize

→ The transfer regime differences that change liquidity timelines

→ The strategic-investor concentration question

→ The cap table architecture differences

→ What allocators should ask before taking either position

By the end of this post, you will have a structural framework for evaluating frontier-AI secondary exposure that goes beyond “they both build large language models.”

Here is what is inside.

Why the structural comparison matters
Corporate purpose: PBC vs. capped-profit
Transfer regime and secondary liquidity
Strategic investor concentration
Cap table architecture
The six questions to ask before taking either position

Let’s begin.

The references and observations behind this post:

→ Public filings from both companies including charter amendments and S-1 drafts where available

→ Reporting in financial press on each company’s recent secondary activity over the last 24 months

→ Conversations with brokers and allocators who have transacted in both names

→ Claude for pressure-testing the structural comparison against the available public record

→ Independent equity research on both names where third-party analysts have built models

1. Why the structural comparison matters

The most common framing I hear from allocators is “I want frontier AI exposure, so I am looking at Anthropic and OpenAI.” That framing is fine as a starting point. It is dangerous as a stopping point.

The reason: frontier AI as a thesis is one bet. Owning Anthropic shares is a second bet on top of it. Owning OpenAI shares is a different second bet on top of it. Treating the second bet as identical to the first means assuming the cap table, the corporate structure, the transfer mechanics, and the strategic-investor positioning do not affect what you eventually receive at an exit.

They affect everything.

The structural lens from Post 03 applies here in a specific way. Two investors with the same dollar exposure to “frontier AI” through different cap-table positions can see materially different net outcomes at IPO, at acquisition, or at any liquidity event in between. The frontier AI thesis is the easy part. The structural call on which company is the harder part.

2. Corporate purpose: PBC vs. capped-profit

The most underappreciated structural difference between the two companies is their underlying corporate purpose construct.

Anthropic is a Public Benefit Corporation.

The PBC structure means Anthropic’s board is legally required to balance shareholder returns against its stated public benefit purpose, which is the responsible development of AI for the long-term benefit of humanity. The board’s fiduciary duty is not pure profit maximization. It is profit maximization weighed against the public benefit purpose.

What this means in practice: certain commercial decisions that a traditional C-corp would be required to take to maximize shareholder value, Anthropic’s board may decline to take if those decisions conflict with the public benefit purpose. Examples could include declining a lucrative commercial contract that creates safety risks, slowing a product release for additional safety testing, or accepting a lower-margin enterprise deal in exchange for a safer deployment model.

The PBC structure is also harder to convert. A shareholder vote is required, and the threshold is higher than for an ordinary corporate change.

OpenAI’s structure is different.

OpenAI was originally a nonprofit. It later created a capped-profit subsidiary that allows commercial investors to participate in returns up to a cap, with any value above the cap flowing back to the nonprofit parent. The exact cap multiples have evolved over time, and the company has been in a multi-year process of restructuring this arrangement.

What this means in practice: investors in the capped-profit entity have a return ceiling above which they do not participate. The structure is complex, the caps have changed, and the relationship between the nonprofit parent and the commercial entity is a recurring subject of legal and governance attention.

The implication for allocators:

Anthropic’s PBC structure is unusual but legally clean. The board’s duties are clearly defined. The conversion process is clearly defined. Investors know what they are buying.

OpenAI’s capped-profit structure is genuinely unique. The legal precedent is thinner. The governance dynamics are more complex. The path to a conventional IPO requires resolving the structure first, and how that resolution happens is a material question for any allocator considering exposure.

Two investors with identical dollar exposure to frontier AI through the two different structures are buying very different governance and exit profiles.

3. Transfer regime and secondary liquidity

Both companies have active secondary markets. The mechanics are not the same.

Anthropic’s transfer regime.

Historically, Anthropic has run relatively structured secondary processes when they have allowed transfers. Tender-style programs at defined windows, with clear pricing and clear qualification criteria. The secondary activity has been less frequent than OpenAI’s but more orderly when it occurs.

The implication for allocators: when Anthropic transfers happen, the price is more discoverable and the process is cleaner. The frequency is the constraint.

OpenAI’s transfer regime.

OpenAI has historically run larger and more frequent tender offers, often in connection with primary financing rounds. The volume of secondary activity in OpenAI shares is meaningfully higher than in Anthropic. Pricing is more visible because the volume is higher.

The implication for allocators: more liquidity, more price discovery, more frequent opportunities to enter or exit. The structural complexity from Section 2 sits on top of all of this.

The asymmetry matters.

An allocator who needs to size a position quickly is going to find more opportunity in OpenAI. An allocator who values orderly process and clean price discovery on each transaction is going to find more comfort in Anthropic. Neither is better in the abstract. They serve different allocator needs.

4. Strategic investor concentration

This is where the two cap tables diverge most sharply.

Anthropic’s strategic investor profile.

Anthropic has taken meaningful capital from multiple large strategic investors over the past several years, with Amazon and Google as the most notable participants. The strategic relationships involve compute commitments, distribution arrangements, and commercial partnerships that go beyond pure financial investment.

The relevant question for allocators: how does strategic-investor concentration affect what minority secondary holders eventually receive at exit? Strategic investors often have negotiated rights, preferred terms, and operational influence that the secondary market does not.

OpenAI’s strategic investor profile.

OpenAI’s most prominent strategic relationship has been with Microsoft, with the relationship including significant compute commitments, distribution through Azure, and what has been reported as a complex multi-tier investment structure tied to specific profit-sharing arrangements.

The relevant question for allocators: how does Microsoft’s structural position in the cap table affect what other shareholders receive? The complexity of the OpenAI-Microsoft relationship is well-known, and the resolution of that relationship in any future IPO or restructuring is a material governance event.

The structural lens applied.

In both cases, the strategic investor is not just a passive holder. They have negotiated rights and operational influence that affect the cap table waterfall at any exit event. A secondary holder buying in at the current valuation is buying behind these strategic positions, and the structural lens from Post 03 requires modeling what the waterfall actually looks like.

For Anthropic, the strategic concentration is split across multiple parties with overlapping but distinct relationships. For OpenAI, the strategic concentration is more singular and the structural complexity is higher.

5. Cap table architecture

Beyond the strategic-investor question, the underlying cap table architectures differ.

Anthropic’s cap table is closer to a conventional venture cap table.

Multiple preferred share classes from successive financing rounds. Common stock held by founders and employees. Standard liquidation preferences. Standard conversion mechanics. The PBC overlay sits on top of this, but the underlying cap table behaves the way a venture cap table behaves at exit.

OpenAI’s cap table reflects the capped-profit structure.

The capped-profit subsidiary issues profit participation interests that operate differently from conventional preferred shares. The mechanics by which value flows from the operating entity to the various tiers of investors and back to the nonprofit parent are more complex than a conventional venture cap table. The cap on returns above certain multiples is the most visible feature, but the underlying mechanics also affect timing, governance, and the conversion path to public-company shares.

The implication for the structural lens.

When you model the cap table waterfall on Anthropic, you can use the standard preferred-stock-meets-common-stock framework with the PBC overlay as a governance consideration. When you model the waterfall on OpenAI, the capped-profit structure requires its own framework that does not map cleanly to conventional venture math.

Two allocators using a single shared model template on both names will produce wrong answers on at least one of them.

6. The six questions to ask before taking either position

The structural lens produces a checklist. Six questions any allocator should answer before committing to either name.

Question 1: Are you buying frontier AI as a thesis, or this specific company?

If the thesis, build a basket across multiple frontier-AI exposures. If the specific company, the structural reasons for choosing that one need to be explicit.

Question 2: How does the corporate purpose construct affect your exit timeline?

For Anthropic, the PBC structure is unusual but stable. For OpenAI, the capped-profit structure is in active evolution and its resolution is a material event. Allocators with shorter time horizons should treat the OpenAI structure resolution as a primary risk factor.

Question 3: Where do you sit in the cap table waterfall?

Both companies have multiple preferred share classes, multiple strategic investors with negotiated rights, and complex transfer regimes. The minority secondary position you are buying needs to be modeled against the full waterfall, not just the headline valuation.

Question 4: What is your liquidity assumption?

Anthropic transfers are less frequent and more structured. OpenAI transfers are more frequent and more visible. Match your assumption to the actual transfer regime, not to a generic “private market liquidity” estimate.

Question 5: How does the strategic investor concentration affect you?

Both companies have strategic investors with operational influence. Model the scenarios in which those strategic investors exercise their rights in ways that affect minority holders. The right answer in both cases is “this matters and most allocators do not model it.”

Question 6: What is your exit thesis?

IPO. Acquisition by a strategic. Continued operation as a private company with periodic structured liquidity. Each exit path has different mechanics for each company, and the structural differences from this post affect which paths are realistic.

If you cannot answer all six questions for the specific position you are considering, the position is not ready to take. Not because frontier AI is a bad thesis. Because the structural call on which company at what entry point is the harder part, and it deserves its own analysis.

Want a second set of eyes?

Book a 20-minute call to talk through pre-IPO investing and deal structures. Bring your questions.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

What I learned from running a Claude-based simulation of a pre-IPO deal flow

Mariano Grosman — Thu, 04 Jun 2026 15:31:03 GMT

A few weeks ago, I ran an experiment I had been wanting to run for almost a year.

I took 12 months of my actual pre-IPO deal flow, anonymized the counterparties, stripped out the names and prices, and fed the entire dataset to a Claude-based AI agent. I gave it the role of a junior analyst on my team. I asked it to evaluate each deal, surface the ones it would have recommended, surface the ones it would have flagged, and explain its reasoning.

Then I compared the agent’s output to what I actually did with each deal in real life. The hits, the misses, and the cases where the agent was right but I was wrong.

This is the most honest post I have written about AI in the pre-IPO workflow, because it is the only one based on a controlled experiment.

→ How the simulation was set up

→ Which AI tools work for which part of the workflow

→ Where the agent got things right and where it surprised me

→ Where the agent got things spectacularly wrong

→ The pattern that emerged across 47 deals

→ What I changed in my workflow after running this

By the end of this post, you will have a clearer mental model of which parts of the pre-IPO workflow are ready to be augmented by AI, which parts are not, and which parts probably never will be.

Here is what is inside.

The setup and the data
Which LLM to use for which part of the workflow
Where the agent got things right
Where the agent got things wrong
The pattern that emerged across 47 deals
What I changed in my workflow after running this

Let’s begin.

The tools and references behind this experiment:

→ 12 months of anonymized deal flow from my own broker book (47 deals total)

→ Claude running as the primary agent with custom instructions

→ Cross-tested against Gemini, Perplexity, and ChatGPT on specific subtasks

→ A scoring rubric I built before running the experiment to evaluate the agent’s recommendations against my actual decisions

→ A separate scoring pass run two weeks later to control for hindsight bias

1. The setup and the data

The 47 deals spanned the typical mix in my book. Late-stage AI names. Defense-tech secondaries. A few fintech opportunities. One space company. Two crypto-adjacent positions. A handful of consumer-tech names that have since IPO’d. Several that have not and probably will not in the next 24 months.

For each deal, I fed the agent the same inputs I had when I evaluated the deal in real life:

→ The company description and recent financial trajectory

→ The cap table and SPV structure (anonymized counterparty identities)

→ The proposed transaction terms (ticket size, price per share, lockup terms, transfer mechanics)

→ My internal notes on the deal as they existed at the time of evaluation

→ The market context for the 30 days before and after the deal

The agent received no information about what I actually decided. It had to evaluate cold.

The output format was simple. For each deal, the agent had to give a recommendation (proceed, decline, or investigate further), a confidence level (low, medium, high), and a written rationale of 200 to 400 words explaining the call.

I then scored the agent’s output against my actual decisions and the outcomes that had since played out for each deal. The results were not what I expected.

2. Which LLM to use for which part of the workflow

I tested the same workflow across four different LLMs over a six-week period. Each one has parts of the job it does meaningfully better than the others. The honest answer is to use more than one.

Claude for structural analysis and long-context reasoning.

When the input is a 50-page research deck plus internal notes plus a cap table, Claude is the strongest at holding all of it in working memory and producing coherent analysis across the whole dataset. It is also the best at the second-pass step where you ask the model to identify the assumptions it made on the first pass. Pattern recognition across long inputs is where it leads.

Best for: the main per-deal evaluation, the follow-up assumption check, structural waterfall modeling.

Gemini for sourcing and market context.

Gemini has the cleanest live web integration for pulling current market context. When the question is about recent comparable IPOs in a vertical and how they priced, Gemini’s output is consistently more current and more verifiable than the alternatives.

Best for: comparable company sourcing, current market conditions, recent IPO performance lookups.

Perplexity for source-cited research.

When you need a research answer and you also need the sources behind it, Perplexity is built for that workflow. Every claim comes with a footnote you can click. For pre-IPO work, where verifying the source of a claim is as important as the claim itself, this saves hours.

Best for: regulatory questions, recent news about specific companies, anything where you need to verify the source before relying on the answer.

ChatGPT for translation and committee-ready language.

When you have a finished analysis and you need to convert it into a one-page memo for an investment committee that does not specialize in pre-IPO, ChatGPT’s output reads cleanest. The default style is professional, accessible, and stays in plain English.

Best for: turning broker-speak into committee language, generating one-page summaries, drafting client communication.

The honest combination.

For my own workflow, the loop is Claude for the main analysis, Perplexity for source-verified research lookups during the analysis, Gemini for comparable-company sourcing, and ChatGPT for the final translation step. Four tools, each used for what it is best at, with the broker judgment connecting them.

The single biggest leverage point in this workflow is not picking one tool. It is connecting the tool to the data sources you already use (Google Drive, Notion, Gmail) through the native integrations the tools now offer. Once your AI can read your deal materials directly, the workflow changes from copying and pasting documents to pointing the AI at a folder. The native integration menus in Claude, Gemini, and ChatGPT walk you through this. No technical setup required.

3. Where the agent got things right

The agent matched my real-world decision on 38 of the 47 deals. An 81 percent agreement rate, which by itself was higher than I would have predicted before running the experiment.

The interesting cases were not the agreements. They were the patterns inside the agreements.

Pattern 1: The agent was excellent at flagging structural red flags.

On every deal where the cap table had a stacked SPV structure that compressed the actual ownership economics, the agent surfaced it before I did in my own notes. Not by a lot. Sometimes by a single day, sometimes by a few hours. But the agent never missed a stacked structure. I had missed one in real life early in the test period and had to go back and correct my own decision later. The agent caught it the first time.

Pattern 2: The agent was surprisingly good at comparable analysis.

On every deal where I had used a recent IPO as a comparable, the agent surfaced two or three additional comparables I had not considered. In about a third of those cases, the additional comparables would have changed my pricing call by 10 to 20 percent if I had factored them in. The agent was not getting smarter than me on the company itself. It was getting more thorough on the comp set.

Pattern 3: The agent was rigorous about disclosure gaps.

On 11 of the 47 deals, the agent flagged that critical information was missing from the materials I had been given. In real life, I had noticed the missing information on 9 of those 11. The agent caught the two I had missed. Both of them turned out to be deals where the missing information was a meaningful signal of seller motivation, and one of them I should have walked away from but did not.

These three patterns are the part of the workflow where the agent is genuinely transformational. Not because it is smarter. Because it is consistently rigorous in a way that humans are not when they are tired or busy or have seen the same kind of deal a hundred times.

4. Where the agent got things wrong

On 9 of the 47 deals, the agent disagreed with me. On 6 of those 9, the agent was wrong and I was right. On 2 of the 9, the agent was right and I was wrong. On 1, the outcome is still unresolved.

The 6 where the agent was wrong fell into three clusters.

Cluster 1: The agent missed founder credibility signals.

Three of the 6 deals where the agent recommended proceeding were deals I had declined because of something specific about the founder or management team. In each case, the agent had access to the same written information I did. The information just did not contain the signal I had picked up from a five-minute conversation with another broker who knew the founder personally. The agent could not see what was not in the data.

Cluster 2: The agent did not weight market timing correctly.

Two of the 6 deals were ones where the deal economics looked fine on paper, but the market timing was wrong. I declined them because I had a sense that the sector was about to soften. The agent did not have access to my sector-timing intuition, which I myself could not fully articulate at the time. In both cases, the sector did soften within 60 days and the deals would have been bad ones.

Cluster 3: The agent failed at one specific high-velocity case.

The one most surprising miss. A deal where the secondary buyer had to commit in 36 hours or lose the allocation entirely. The agent gave a careful, balanced analysis recommending we investigate further. The right answer was either to commit immediately or pass. There was no investigate-further option in real life. The agent could not factor velocity constraints because nothing in its training had taught it that some pre-IPO situations are time-bounded in a way that fundamentally changes the decision space.

The 2 cases where the agent was right and I was wrong are the most instructive. In both, the agent flagged a structural risk I had underweighted. In one case, I had been too eager because I had a relationship with the seller. In the other, I had let a comparable company’s recent IPO outcome bias me toward optimism. Both times, the agent did not have my biases, so it saw the structural risk clearly. Both times, my position would have lost money if I had gone in. I caught both before committing, but only after the agent’s analysis prompted me to reread my own notes.

5. The pattern that emerged across 47 deals

When I stepped back and looked at the agent’s strengths and weaknesses together, a pattern emerged that I had not expected.

The agent was excellent at every part of the job that depended on rigor, consistency, and breadth of comparison. It was poor at every part that depended on signals not present in the data.

This is obvious in hindsight. But what was not obvious was the implication. The parts of the pre-IPO workflow that depend on rigor are the parts that scale. The parts that depend on out-of-data signals are the parts that do not.

If you build a broker operation that relies on rigor, consistency, and comparison, AI will compress that operation by 5x to 10x. The same broker can now cover 5x to 10x the deal flow with the same quality of structural analysis.

If you build a broker operation that relies on out-of-data signals (founder relationships, sector timing intuition, velocity-bound situations), AI cannot compress that operation at all. The same broker can cover roughly the same deal flow as before. The advantage is qualitative, not quantitative.

The implication is uncomfortable for incumbents. Brokers whose competitive advantage was rigor are about to face brokers who use AI for rigor and have time left over to build the out-of-data signals. The combination is more powerful than either piece alone.

6. What I changed in my workflow after running this

The experiment changed three things in how I run my own work.

Change 1: Every deal now gets an AI second-opinion pass before I make a final call.

Not because the AI is smarter. Because it is consistent. On the 11 deals where the agent caught something I had missed in real life, the agent did not catch it because it was clever. It caught it because it never gets tired and never assumes. Building that second-opinion pass into the standard workflow has caught two structural issues in the four weeks since I started running it.

Change 2: I spend less time on structural analysis and more time on the out-of-data parts.

The relationship calls. The 15-minute conversations with brokers who know the founder. The reading of body language in seller meetings. The sector-timing reads. These are now where I spend the highest-leverage hours of my week. The structural analysis happens in parallel via the AI workflow.

Change 3: I stopped trying to use AI for the parts where it does not help.

For 18 months, I had been trying to use AI for things like sourcing new deals, predicting IPO timing, and reading relationship dynamics. None of these worked. The experiment confirmed that none of them will work in their current form. I have stopped trying. The hours I was spending on those failed AI use cases now go to the parts of the workflow where the second-opinion pass is high-leverage.

The interesting question is not whether AI will change pre-IPO work. It already has. The interesting question is whether you are pointing it at the right parts of the workflow.

Thinking about pre-IPO investing?

Book a 20-minute call. We can talk through how it works, common pitfalls, and what makes sense for your situation.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

Why the secondary market will be priced in real time by 2028

Mariano Grosman — Tue, 02 Jun 2026 14:31:48 GMT

Twenty years ago, a Bloomberg terminal was a luxury item reserved for institutions. Today, most retail investors have real-time public equity prices in their pocket for free. The information that was once locked behind institutional gates is now ambient.

My view is that the same shift is about to happen to pre-IPO secondaries. This is my forecast, not a certainty, and I will lay out the reasoning so you can decide whether you agree.

Right now, the secondary market for private company shares operates the way the bond market operated decades ago. Prices clear in private. Bid-ask spreads exist only in brokers’ inboxes. The same share class can trade at materially different prices on the same day depending on who you call.

My forecast is that within roughly 24 months, that breaks.

By 2028, I expect the late-stage names with the most secondary activity to have continuous live pricing visible to the entire market. Not perfect pricing. Not always-on auction-style trading. But a real-time bid-ask spread an allocator can look at without picking up the phone.

If I am right, this changes the broker model, the allocator workflow, and the structural moats that have defined the industry for two decades.

→ Why the technical infrastructure is finally ready

→ What live pricing actually looks like in practice

→ The three industry moats I believe disappear

→ The two new moats I believe replace them

→ What this means for every allocator and broker watching the next 24 months

By the end of this post, you will understand the reasoning behind my 2028 forecast, and which side of the change to position on if you find the reasoning persuasive.

Here is what is inside.

Why live pricing was impossible until now
The infrastructure that just got built
The three moats I believe disappear
The two moats I believe replace them
What this means for allocators and brokers
The 24-month positioning question

Let’s begin.

The references and observations behind this post:

→ My own broker book and direct observation of how secondary transactions clear today

→ Public information from emerging pre-IPO marketplaces and OTC platforms

→ The published SEC exemption frameworks (Reg A+, Reg CF) available at sec.gov

→ Claude for pressure-testing the reasoning behind the forecast

→ My own read of how AI-powered pricing has restructured adjacent markets like corporate bonds and private credit

A note on figures: where I do not have a verifiable published source for a specific number, I have left the number out and described the direction of the trend instead. The forecast is mine. The reasoning is laid out so you can judge it.

1. Why live pricing was impossible until now

The pre-IPO secondary market has had three structural barriers to live pricing. In my assessment, all three are coming down at the same time, which is why I believe the shift will feel sudden when it arrives.

Barrier 1: Data fragmentation.

Pre-IPO transactions clear across many broker-dealers, multiple marketplaces, and a large number of direct bilateral deals. Historically, no single source aggregated this data in real time. Each broker saw their own slice and nothing else.

In my view this is changing, because specialized data platforms are now aggregating secondary transaction data across the industry, and AI tools are normalizing the formats fast enough to serve it closer to live.

Barrier 2: Regulatory friction.

Pre-IPO transactions sit in a complex regulatory environment. Securities rules were written for public markets, and the rules for private transfers vary by issuer, by jurisdiction, and by buyer qualification.

The relevant regulatory frameworks here are real and published: Regulation A+ and Regulation Crowdfunding (Reg CF) are SEC exemption frameworks documented at sec.gov, and the SEC has modernized its accredited investor definition in recent years. My read is that this accumulated regulatory clarity makes it more feasible for platforms to build live-pricing infrastructure than it was a few years ago. That read is my interpretation, not an SEC statement.

Barrier 3: Issuer cooperation.

Most pre-IPO companies have historically restricted secondary trading. Transfer approvals were slow. Right-of-first-refusal clauses were aggressive. In my experience, companies have often preferred to limit price discovery.

This is the slowest-moving piece, but my observation from the market is that it is moving. Some late-stage companies now run their own structured secondary programs. Those programs, by definition, produce price discovery. My forecast is that once enough large names run regular structured secondaries, the market will have enough data to support live pricing across the late-stage category.

2. The infrastructure that just got built

In my assessment, three technical advances together make 2028 a realistic inflection point. I am describing capabilities I observe in the market, not citing measured adoption statistics.

Advance 1: AI normalization of transaction data.

Pre-IPO transaction data has always been messy. Different brokers report different fields, in different formats, with different completeness. AI tools can now normalize this kind of data far faster than a manual process. In my own workflow, work that used to take days now takes a fraction of that.

Advance 2: Specialized matching engines.

Marketplaces including Forge, EquityZen, and Hiive have built order-matching infrastructure for private shares. These are real, operating platforms. I am not citing their volumes because I do not have verified public figures for them. What I can say from observation is that the infrastructure exists and the category is growing.

Advance 3: Compliance automation.

The slowest part of any pre-IPO transaction has historically been the compliance review. Buyer qualification, source-of-funds verification, transfer approval, escrow setup. In my own practice, AI-assisted compliance tooling has compressed parts of this meaningfully. Faster compliance means more transactions, which means more data, which in my view feeds better live pricing over time.

Stack these on top of the regulatory and data shifts from Section 1, and my conclusion is that the infrastructure for live pricing is already partly built. I believe the next 24 months are about adoption, not invention. That conclusion is my forecast.

3. The three moats I believe disappear

This is where the forecast gets uncomfortable for incumbents. In my view, the pre-IPO industry has been built around three structural moats, and all three lose most of their value in a live-pricing world. What follows is my analysis, not a measured finding.

Moat 1: Information asymmetry on price.

For decades, the broker who knew where recent transactions cleared had an edge. They could quote a tighter bid-ask than competitors and capture more spread.

In a live-pricing market, every allocator has access to the same transaction data. My view is that the broker who used to win on price information will need to win on something else, or lose the business.

Moat 2: Synthesis and analysis speed.

The broker who could turn around a structural analysis quickly used to win against the slower broker. AI has already compressed this for everyone. My forecast is that by 2028 the speed differential between brokers on the analysis side is effectively gone, and synthesis becomes a baseline expectation rather than an edge.

Moat 3: Relationship density as a substitute for transparent pricing.

Allocators who could not see the market had to rely on brokers they trusted. Trust took years to build, which kept new entrants out.

In a live-pricing world, allocators can verify a broker’s pricing against the market in real time. My view is that this makes trust faster to build because it is more verifiable, and it lowers the barrier to entry for the industry. The relationship moat does not vanish, but in my assessment it weakens.

4. The two moats I believe replace them

Some moats survive. In my analysis, two new ones emerge. This is my opinion on where the durable advantage moves.

New moat 1: Execution speed at scale.

In a live-pricing market, my view is that the broker who can clear a large block quickly wins against the broker who needs weeks. Speed becomes the new edge.

Execution speed at scale requires deep counterparty networks, automated compliance infrastructure, and the operational discipline to close without endless back-and-forth. I believe the brokers who build this capability now will have an advantage by 2028.

New moat 2: Structural advisory on what to buy and what to avoid.

When everyone can see the price, my view is that the broker who can explain why the price is what it is wins. The structural lens, the cap table mechanics, the lockup terms, the SPV layer fees. The advisory layer on top of the live-pricing layer is where I believe the new margin lives.

This is harder to commoditize than execution speed. AI can compress the analysis, but in my experience the judgment about which structural risk matters in which scenario is still a human craft.

5. What this means for allocators and brokers

The implication is different for each side of the table. The recommendations below follow from my forecast. If you do not share the forecast, weight them accordingly.

For allocators:

→ Start valuing the broker who explains the price over the broker who merely quotes it. If pricing becomes public, the explanation is where the value concentrates.

→ Pay attention to broker execution speed. In my view, brokers who cannot close quickly will struggle in the next cycle.

→ Build internal capability to read live transaction data as it becomes available. My forecast is that this becomes a standard part of any allocation decision.

For brokers:

→ Stop competing primarily on price information. In my assessment, that moat is weakening. Compete on execution speed and structural advisory.

→ Invest in compliance automation. My view is that the brokers who close fastest will own the next cycle.

→ Build the analytical layer that sits on top of live pricing. Everyone will be able to quote the price. In my opinion, only some will be able to explain what it means.

6. The 24-month positioning question

The shift from opaque to live pricing is, in my view, the most important structural change in pre-IPO secondaries in two decades. I believe it is arriving faster than most participants expect. That belief is my forecast, and the next 24 months will test it.

Three things to consider now:

→ Map your current dependencies on price opacity. If your model relies on knowing what recent transactions cleared at while competitors do not, my view is that edge is on a clock.

→ Identify the structural advisory layer you can build. What analysis do you do that AI cannot easily replicate and that allocators will pay for in a transparent market? In my opinion, that is the value proposition to lock in now.

→ Talk to the platform operators building this infrastructure. Not to compete with them. To understand how their data layer will integrate with your workflow.

The interesting question is not whether AI will change pre-IPO work. In my view it already has. The question I am putting to you is whether the pricing transparency I am forecasting arrives on the timeline I expect. The discipline is in positioning before the answer is obvious.

Want to go deeper?

Book a 20-minute call to talk through pre-IPO investing, deal structures, and what to watch for.

Book Now

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

The valuation framework most allocators are using is wrong

Mariano Grosman — Sat, 30 May 2026 13:30:55 GMT

A few months back, a family office sent me their internal pre-IPO valuation model. It was beautiful. Discounted cash flows. Public-market comparables. Sensitivity tables. Twenty tabs of Excel that would make any banker proud.

In my assessment, the valuation it produced was far too high. My read was that it overstated the position by more than half. That is my judgment on that specific model, not a measured figure.

The analysts were not bad. They were sharp, well-credentialed, and methodical. The model failed because they were using a framework built for public companies and applying it to private ones. Those are two different problems.

This is the most common mistake I see in pre-IPO work, and almost nobody is talking about it. In my experience, allocators lose real money because they price private positions with public-market math.

→ Why DCF breaks the moment you cross into pre-IPO

→ Why public-market comparables miss the structural adjustments

→ Why the bid-ask in private markets carries information that public quotes do not

→ The three-lens framework I use to price private positions

→ How to test whether your current valuation method is helping or hurting

By the end of this post, you will have a clear test for whether your existing pre-IPO valuation framework is producing real numbers or sophisticated fiction.

Here is what is inside.

Why public-market valuation methods break in private markets
The three structural reasons standard models fail
The three-lens framework that actually works
How to apply each lens in practice
The test for your current valuation method
What this means for your next position

Let’s begin.

The references and frameworks behind this post:

→ Aswath Damodaran’s published work on illiquidity discounts and control, available at pages.stern.nyu.edu

→ Restricted-stock and lack-of-marketability studies cited in the valuation literature

→ Real pre-IPO transactions across my own broker book over the last 18 months

→ Claude for pressure-testing the framework against the published literature

→ Conversations with allocators who have run both public-market and private-market models

A note on figures: where I cite a discount range, I have attributed it to its published source. Where I give my own number, I have flagged it as my practitioner view rather than a published statistic.

1. Why public-market valuation methods break in private markets

Every standard valuation framework assumes three things that are true about public companies and false about private ones.

Assumption 1: You can sell when you want.

In public markets, liquidity is a given. The bid-ask is tight, the volume is deep, and you can exit a position in minutes. Every public-market valuation method silently assumes this. The “fair value” the model produces is the price at which a willing buyer and willing seller would transact today.

In private markets, liquidity is the entire problem. You might wait 12 months for a secondary buyer. You might wait 36 months for the IPO. You might wait indefinitely if the company stays private. Any valuation that does not adjust for this is overstating the value.

How much of an adjustment? The published literature gives a clear starting point. Aswath Damodaran of NYU Stern, whose valuation work is among the most widely used in the field, puts the illiquidity discount for a private firm at 20 to 30 percent as a rule of thumb, with a 25 percent base case for a profitable firm (source: Damodaran, “Private Company Valuation,” pages.stern.nyu.edu). In my experience, most allocators apply something closer to 10 to 15 percent on pre-IPO positions, which is below even the low end of the published range. That gap, in my view, is where they lose money.

Assumption 2: The market price is information.

Public-market valuation methods anchor on the current trading price as a reality check. If the DCF says $200 per share and the stock trades at $150, the analyst either adjusts the assumptions or accepts that the market knows something they do not.

Private markets have no continuous price. The last secondary trade might be weeks old at one valuation, and the current bid might be materially different. There is no single “market” to anchor against. There is a range of prices different counterparties would accept for the same shares on the same day.

If your valuation method requires a reference price to calibrate, in my view it does not work in private markets.

Assumption 3: The cap table is what it looks like.

In public companies, the cap table is typically one share class, fully diluted, disclosed quarterly.

In private companies, the cap table is a layered stack of preferred classes, common, convertibles, SAFEs, warrants, and SPV positions, with liquidation preferences and ratchets that change which dollar of exit value goes to which holder. Two investors with positions of equal book value can end up with materially different proceeds at the same exit price.

Any valuation method that does not model the cap table waterfall is, in my assessment, producing a number that is structurally meaningless.

2. The three structural reasons standard models fail

Each of the three assumptions corresponds to a structural reason the standard models break. The failure mode is different in each case.

Reason 1: The illiquidity discount is not optional.

DCF and public-market comparables both produce a gross valuation. To convert that to a private-market value, you have to apply a liquidity discount. The discount is not a rounding adjustment. It is a structural correction that the published evidence puts in a meaningful range.

The empirical restricted-stock studies that underpin much of this literature found that restricted, and therefore illiquid, stocks traded at discounts of 25 to 35 percent relative to their unrestricted counterparts (source: Damodaran, “Marketability and Value: Measuring the Illiquidity Discount,” pages.stern.nyu.edu). Broader practitioner surveys cite a wider 25 to 50 percent range, with 35 percent given as a common median data point (source: Financier Worldwide, “Valuation and liquidity discounts”).

Here is where I will give my own practitioner view, clearly flagged as such: for late-stage pre-IPO names with active secondary markets, I tend to work in the middle of the published range, around 25 to 35 percent. For early-stage names with no secondary activity at all, my own practice is to go higher than the standard published ranges, because in my experience the illiquidity is more severe than the restricted-stock studies capture. That higher number is my judgment, not a published statistic, and you should treat it as one practitioner’s view rather than an established figure.

If your current model is using a discount below the 20 to 30 percent published rule of thumb on a pre-IPO position, my view is the model is understating the discount.

Reason 2: The control component runs against the minority holder.

Damodaran addresses the control component directly: a controlling stake in a private business is more liquid than a minority stake, because the controlling holder can force a sale or a liquidity event and the minority holder cannot (source: Damodaran, “Private Company Valuation,” pages.stern.nyu.edu).

The implication for pre-IPO secondary buyers: you are almost always buying a minority position with no control. In my practitioner experience, this warrants an additional adjustment on top of the base illiquidity discount. I am not going to attach a precise percentage to it, because I have not found a published figure I trust for the specific case of late-stage pre-IPO minority secondaries, and inventing one would be exactly the kind of false precision this whole post argues against. The direction is well-supported: minority positions are worth less than control positions, all else equal. The magnitude is a judgment call.

Reason 3: The bid-ask spread is information.

In public markets, the bid-ask is a few cents wide and contains almost no information beyond market-maker mechanics.

In private markets, in my experience, the bid-ask can be wide, and it carries signal. A wide spread means the market disagrees on value. A narrow spread means consensus. A spread that has widened over recent weeks means new information has entered the market. A spread that has narrowed often means a transaction is close. Most valuation methods ignore the bid-ask entirely. In my view, it is where a lot of the real information lives.

3. The three-lens framework that actually works

After 11 years of broker work, here is the framework I have settled on. It does not produce a single point estimate. It produces a range with three lenses, and I only get comfortable with a position when the three lenses converge.

Lens 1: The structural lens.

Start with the cap table waterfall. What does this position actually own, after preferences, after ratchets, after the SPV layer fees? What does it own at the IPO, at a sale, at a recap, at a dilution event? The structural lens produces a range of net proceeds at different exit valuations. Not a single value. A range.

Lens 2: The market lens.

Look at the secondary market for this company over the last 24 months. Where have actual transactions cleared? Where is the current bid and ask? How wide is the spread, and is it widening or narrowing? The market lens produces a “what would actually clear today” number. Not a model output. A market observation.

Lens 3: The exit lens.

Look at comparable companies that have exited in the last 36 months. Same vertical, similar revenue range, similar growth profile. What multiples did they print at IPO or acquisition? What was the gap between the last private valuation and the public-market valuation 90 days post-listing? The exit lens produces a “what would this be worth on the other side” number, adjusted for the time and risk it takes to get there.

In my approach, the three lenses should agree within a reasonable range. When they do not, the position is mispriced in one of three directions, and the disagreement itself tells you where to look.

4. How to apply each lens in practice

The framework is only useful if you can run it. Here is how I build each lens.

Building the structural lens.

You need the full cap table: all preferred classes, the liquidation preferences, any participation rights, the SPV layer structure if applicable, and the current fully-diluted share count. Most of this requires the company to share the information or your broker’s relationship with the company to surface it.

Once you have the structural data, model the waterfall at several exit valuations, from well below the current mark to well above it. The output is a table showing net proceeds at each scenario, after fees and carry. That table is the structural lens.

Building the market lens.

You need secondary market data over the last 24 months. The most useful sources are broker networks, specialized pre-IPO platforms, and direct conversations with counterparties who have transacted in the name. Look at the most recent transactions, note the price, the volume, and the date, and look at the current bid-ask spread. The market lens is, in my practice, the median of recent prints adjusted toward the current bid-ask midpoint.

Building the exit lens.

Identify three to five comparable companies that have exited in the last 36 months. Comparable means similar vertical, revenue scale, growth rate, and capital structure. Pull the last private valuation before each company’s exit, and the public-market valuation about 90 days after listing. The exit lens is the relationship between those two figures across the comp set, applied to your current private valuation and discounted back for time and risk.

5. The test for your current valuation method

Three questions to ask about whatever framework you are using today.

Question 1: Does your model produce a single number or a range?

In my view, if it produces a single number, the model is wrong. Private-market valuations are uncertain by structure. A point estimate hides the uncertainty instead of representing it.

Question 2: Is your illiquidity discount inside the published range?

The published rule of thumb is 20 to 30 percent, with restricted-stock evidence at 25 to 35 percent and a broader practitioner range up to 50 percent (sources: Damodaran, NYU Stern; Financier Worldwide). If your discount sits below that range on a pre-IPO position, you are, in my assessment, using a public-market liquidity assumption on a private-market asset.

Question 3: Does your method use the bid-ask spread as input?

If the spread is not in your model, in my view you are ignoring one of the most informative data points in private markets. Your number is a model output, not a market observation.

If your current method fails these tests, my honest assessment is that the valuation it produces is not reliable.

6. What this means for your next position

The pre-IPO market is more transparent than it used to be, but in my view the valuation frameworks most allocators use were built for a market that did not have secondary trading data, broker networks, or AI tools to synthesize it all.

Three things to do now:

→ Re-run your most recent pre-IPO valuation through the three-lens framework. Compare the answer to what your current model produced. The gap is your blind spot.

→ Audit your illiquidity discount across the book against the published ranges. If you are running a discount below the 20 to 30 percent rule of thumb on any pre-IPO position, in my view it deserves a second look. Do the same audit on your minority-position adjustment.

→ Subscribe to two or three secondary market data sources. Broker networks, specialized platforms, anything that gives you actual transaction data. Without it, the market lens is a guess.

The framework you use determines the answer you get. If the framework is built for public markets, in my view the answer is wrong. If it accounts for the three things that make private markets different, the answer becomes useful.

Have a question this post raised?

Book a 20-minute call. We can dig into whatever you want to understand better about pre-IPO investing.

Book Now

Important information

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

Why most AI tools fail at private-market analysis

Mariano Grosman — Thu, 28 May 2026 13:49:28 GMT

Last month I watched a family office run their AI tool against a pre-IPO research deck. The output looked beautiful. Clean summary, neat bullet points, confident tone.

Almost every number in it was wrong.

The cap table summary missed two layers of preferred stock. The revenue figure was last year’s projection, not this year’s actual. The competitive analysis named a company that had been acquired six months earlier.

The output was confident, fluent, and unreliable.

This is the gap most allocators have not internalized yet. Public market AI tools work because the data is clean, current, and abundant. Private market AI tools fail in the same workflow because the data is messy, stale, and sparse.

After eighteen months of using AI as a core part of my deal pressure-testing, here is what I have learned about where it works, where it breaks, and what to actually use it for.

→ The three reasons most AI tools fail on private market data

→ The four tasks AI does better than any human analyst

→ The four tasks where AI will quietly mislead you

→ How to build a workflow that uses AI for synthesis and keeps the judgment human

By the end of this post, you will have a clear test for whether your current AI workflow is actually saving you time, or just helping you make confident wrong decisions faster.

Here is what is inside.

Why public market AI tools fail on private market data
The four tasks AI does better than any analyst
The four tasks where AI will mislead you
The hybrid workflow that actually works
The test to know if your AI workflow is helping or hurting
What this means for your next deal

Let’s begin.

The tools and data sources I tested this on:

→ Claude for structural analysis and synthesis

→ ChatGPT for general research and drafting

→ Perplexity for source-cited research

→ Specialized pre-IPO platforms across my own deal book

→ Eighteen months of real pre-IPO transactions in my own broker workflow as the validation set

1. Why public market AI tools fail on private market data

The reason is mechanical, not mysterious. Every AI tool you have ever used was trained on public market data. SEC filings. Earnings transcripts. Press releases. Analyst reports. Bloomberg terminals. Yahoo Finance.

Public markets generate roughly 10,000 data points per company per day, all of it timestamped, all of it structured, all of it available to scrape.

Private markets generate maybe 10 data points per company per quarter. Some of it is in deal rooms behind NDAs. Some is in broker networks behind relationships. Some is in private databases behind subscriptions. None of it is timestamped consistently.

Three structural problems make this gap worse.

Problem 1: Stale data. A pre-IPO company’s last public valuation might be 18 months old. The most recent secondary trade might be 6 weeks old. The current bid-ask might exist only in a broker’s email inbox. AI tools default to whatever they can find, which is usually the 18-month-old number. They present it with the same confidence as a real-time stock quote.

Problem 2: Missing context. Public companies disclose their cap table structure quarterly. Private companies do not disclose it at all. AI tools fill that gap with what they can infer, which is usually wrong. Preferred stock layers, ratchet provisions, transfer restrictions, and lockup terms are the kind of details that change the entire economics of a position, and they live nowhere in the training data.

Problem 3: No feedback loop. When an AI tool is wrong about a public stock price, the market corrects it within seconds. When an AI tool is wrong about a pre-IPO company, nobody knows until the deal closes or the IPO prices. The errors compound silently.

The result: a tool that looks like it is helping, but is actually generating confident output on the wrong inputs.

2. The four tasks AI does better than any analyst

This is the part most operators get wrong in the other direction. AI is genuinely transformational for some parts of the pre-IPO workflow. Just not the parts most people are using it for.

Task 1: Structural map building.

Take an entire pre-IPO research dossier (200 pages, multiple PDFs, internal memos, press clippings) and reduce it to a structured map of every entity mentioned, every dollar figure quoted, every date referenced. Then ask the AI which entities, figures, and dates contradict each other across the documents.

This is the highest-leverage thing AI does. A human analyst doing the same task takes 8 hours. AI does it in 4 minutes. The contradictions it surfaces are where the real diligence work begins.

Task 2: Comparable company sourcing.

Ask AI to surface every company that has gone public in the last 36 months with similar characteristics: same vertical, similar revenue range, similar growth profile, similar capital structure. Then ask it to pull the post-IPO performance of each one.

The output is not perfect, but it is 80 percent right and it is 100 percent faster than building the comp set manually.

Task 3: Bear case generation.

Give AI the bull case for a position and ask it to construct the strongest possible bear case using only the same source material. This forces it to find the weaknesses in the analysis you have already done.

The trick: tell it the bear case will be presented to a hostile investment committee. The tone shifts. The pushback gets sharper. The arguments get more concrete.

Task 4: Translation.

Take broker-speak (ratchets, drag-along, weighted-average anti-dilution, qualified financing thresholds) and translate it into committee-ready language that a non-specialist board member can follow. This is the most underrated AI use case in finance. Allocators have been doing this manually for years. AI does it in seconds, and the output is usually cleaner than the manual version.

3. The four tasks where AI will mislead you

Task 1: Current valuation estimation.

If you ask AI to estimate a pre-IPO company’s current valuation, it will give you a number. The number will be wrong. Sometimes by 20 percent, sometimes by 80 percent. There is no way to tell without independent verification.

Why: the training data is too old, the secondary market data is too sparse, and the AI tool will not flag its own uncertainty. It will state the wrong number with the same confidence it states the right one.

Use case where this matters: never ask AI for a current valuation. Always ask a broker, then ask AI to pressure-test the broker’s number.

Task 2: Cap table reconstruction.

If you ask AI to map a private company’s cap table from public sources, it will reconstruct something that looks right. It will almost always miss the preferred stock layers, the convertible notes, the SAFE conversions, and the recent secondary trades that change the underlying ownership math.

Use case where this matters: the cap table is where 70 percent of pre-IPO surprises hide. Do not let AI build it for you. Use AI to check a cap table you have already built from primary sources.

Task 3: Predicting IPO timing.

If you ask AI when a company will go public, it will give you a window. The window will be wrong. Companies go public on regulatory windows, market conditions, and internal politics, none of which are predictable from training data.

Use case where this matters: IPO timing is a coin flip even for the people inside the company. AI estimates of timing are noise.

Task 4: Sourcing specific deal flow.

If you ask AI to find pre-IPO opportunities for you, it will surface companies that are already widely known. It cannot access the actual deal flow, which lives in broker networks and private platforms behind logins.

Use case where this matters: AI is downstream of the source. The source is still the same as it was fifteen years ago.

4. The hybrid workflow that actually works

The brokers who get this right are not using AI to replace any part of the workflow. They are using it to compress specific parts so they have more time for the parts that need human judgment.

Here is the actual loop I run on every new pre-IPO position I evaluate:

Step 1: Build the structural map manually.

Pull the primary sources. Read them yourself. Take notes. This is the part you cannot outsource and the part that builds the judgment you will need for every subsequent step.

Step 2: Use AI to find what you missed.

Hand the AI everything you read in Step 1 and ask it to surface every contradiction, every gap, and every unverified assumption in your notes. This is where AI earns its keep.

Step 3: Use AI to build the comp set.

Ask AI to surface comparable companies that have gone public recently and pull their post-IPO performance. Verify the AI output against your own primary sources before using any of it.

Step 4: Use AI to construct the bear case.

Give AI the bull case you have built and ask it to construct the bear case. Read both cases side by side. If the bear case is more concrete than the bull case, that is a signal.

Step 5: Pressure-test with humans.

Take the synthesized output and run it past two or three brokers who actually know the company. Their pushback is the final filter.

The pattern: AI compresses Steps 2, 3, and 4. Steps 1 and 5 stay human, because they have to.

5. The test to know if your AI workflow is helping or hurting

Three questions to ask about any AI tool you are using in your pre-IPO workflow:

Question 1: When the AI is wrong, can you tell?

If you cannot independently verify the output, the AI is not helping you, it is making you confident in unverified output. That is worse than no AI at all.

Question 2: What part of the analysis are you skipping because the AI told you the answer?

If the AI shortcut is replacing a step where the judgment lives, you are losing the part of the job that has the highest value.

Question 3: How often do you go back and check whether the AI was actually right?

If the answer is never, you have no idea whether the tool is working. The feedback loop is the only thing that tells you which AI outputs to trust.

If your current workflow fails any of these three tests, the workflow is not actually helping. It is producing confident output faster, which is not the same thing.

6. What this means for your next deal

The pre-IPO market is about to be flooded with AI-generated research, AI-generated valuations, and AI-generated investment memos. Most of it will be wrong in subtle ways that are not visible until after the position is taken.

Three things to do now:

→ Audit your current AI workflow. Pick a deal you closed in the last six months. Re-run the analysis through your AI tool. Compare the AI output to what actually happened. The gaps tell you which tasks to keep automating and which to take back manually.

→ Build the human-only checkpoints. Decide which two or three steps in your workflow cannot be done by AI, no matter how good the tool gets. Lock those as the parts you do yourself, every time.

→ Talk to brokers who use AI well. Not to copy their prompts. To understand which parts of their judgment they refuse to outsource. That refusal is where the actual edge is.

Curious about pre-IPO investing?

Book a 20-minute call. Happy to walk through how the market works, what to look for, and how to think about it for your portfolio.

Book Now

Important information

STANDARD RMS DISCLOSURE LANGUAGE

RISK FACTOR LANGUAGE

EQUITY DISCLOSURE LANGUAGE

EQUITY DISCLOSURE: RMS and/or its registered representatives hold an equity interest in the seller/issuer of the Securities.

Built for qualified investors. Spoken plainly.

Mariano Grosman

Managing Director, Rainmaker Securities

Mariano Grosman

Anduril's pre-IPO position is structurally different from every other defense-tech name

Table of contents

1. Why the defense-contractor framework is wrong

2. Why the software-startup framework is wrong

3. The procurement structure that shapes the position

4. The product-first model that sets it apart

5. The structural questions to actually ask

6. Why the category matters

The pre-IPO due diligence checklist most allocators skip

Table of contents

1. Why financial diligence is not enough

2. The right-of-first-refusal check

3. The transfer mechanics check

4. The 409A and lockup check

5. The SPV manager and information rights check

6. The six-point checklist

The SpaceX IPO headline was the valuation. The lesson is the lock-up

Table of contents

1. What actually happened on June 12

2. Why the lock-up is the real story

3. How the staggered lock-up works

4. “Eligible to transfer” is not “able to sell”

5. What this teaches you about your own exit

6. The structure was always the story

AI is not coming for the pre-IPO broker. It is coming for the broker's analyst

Table of contents

1. Why “AI replaces brokers” is the wrong frame

2. What the analyst role actually does

3. Why AI does that work well

4. How this changes the way a brokerage is built

5. Why this is good for allocators

6. What it means for a career in this industry

How AI is changing the broker-client conversation, and what the new conversation actually sounds like

Table of contents

1. What the old conversation sounded like

2. What the new conversation sounds like

3. The three questions allocators ask now

4. Why the shift favors brokers who adapt

5. What it means for how you use your broker

6. The conversation worth having

Why AI is going to change pre-IPO access in Latin America before it changes anything in the US

Table of contents

1. Why the US market only gets incremental change

2. The three structural inefficiencies in Latin American pre-IPO

3. Why AI removes those specific barriers

4. What this means for allocators on both sides

5. Why the next decade belongs to AI-enabled brokers

6. The positioning question for the next 24 months

The compliance bottleneck is the next pre-IPO opportunity

Table of contents

1. Why compliance is the real bottleneck

2. What AI actually compresses in the workflow

3. Why faster compliance reshapes the market

4. Where the risk actually sits

5. What this means for allocators and brokers

6. The 24-month positioning question

Anthropic's secondary market is structurally different from OpenAI's

Table of contents

1. Why the structural comparison matters

2. Corporate purpose: PBC vs. capped-profit

3. Transfer regime and secondary liquidity

4. Strategic investor concentration

5. Cap table architecture

6. The six questions to ask before taking either position

What I learned from running a Claude-based simulation of a pre-IPO deal flow

Table of contents

1. The setup and the data

2. Which LLM to use for which part of the workflow

3. Where the agent got things right

4. Where the agent got things wrong

5. The pattern that emerged across 47 deals

6. What I changed in my workflow after running this

Why the secondary market will be priced in real time by 2028

Table of contents

1. Why live pricing was impossible until now

2. The infrastructure that just got built

3. The three moats I believe disappear

4. The two moats I believe replace them

5. What this means for allocators and brokers