Accessing AI Leaders Before the IPO: A Framework for Late-Stage Private Names

Introduction

A wealthy investor who decides, in 2026, that artificial intelligence is the defining investment theme of the decade faces an immediate and frustrating problem. The companies that most clearly embody the theme, the frontier model laboratories, the leading applied-AI platforms and the specialised infrastructure providers that sit beneath them, are overwhelmingly private. They have raised extraordinary sums, they command valuations that rival mature listed businesses, and yet their shares cannot be bought on any exchange. The investor who wishes to own them must therefore navigate the private secondary market, a market that is opaque by design, uneven in quality and unforgiving of the unprepared.

This is not a temporary state of affairs. The structural drift towards companies staying private for longer has been visible for two decades, and the current AI cohort has taken it further than any group of companies before it. Abundant private capital, from growth funds, sovereign investors and a new generation of dedicated AI vehicles, has removed much of the financing pressure that once pushed companies towards a listing. The result is that an enormous amount of value is being created, and in some cases realised through secondary sales, entirely outside the public markets. For an allocator, the practical consequence is stark: to participate in the theme at all, one must learn to operate in the private secondary market or accept being left on the sidelines until a listing that may be years away.

The difficulty is that the private secondary market for marquee AI names is not a single, orderly venue. It is a patchwork of routes, each with its own economics, its own intermediaries and its own pitfalls. The same exposure to the same company can be obtained by buying shares directly from a departing employee, by purchasing a unit in a special-purpose vehicle that someone else has assembled, by committing to a secondary fund that holds a basket of such names, or by entering a synthetic forward that references the share price without ever placing a share on the register. These routes are not interchangeable. They differ in what they cost, in what they let the investor see, in what rights they confer and in how cleanly they can be exited. An investor who chooses the wrong route, or who chooses the right route but diligences the wrong shares, can end up paying a full price for a stale mark, an impaired claim or no claim at all.

The Framework: Names, Routes, Diligence And Structure

This section develops the framework in the order of the four decisions. It begins with the name screen, proceeds to the access routes and the comparison among them, then sets out the diligence sequence specific to pre-IPO shares, and closes with the structuring and governance choices. Throughout, the figures are illustrative devices for reasoning, not sourced evidence.

Screening the Names

The first decision is which names belong in the screen at all. It is tempting to treat this as a stock-picking exercise, a hunt for the single company that will define the cycle. That framing is a trap. In a cohort of marquee AI names, most of the candidates are, on the merits, good companies; the category leadership, the revenue trajectory and the strategic importance that put them on the radar are real. The differentiating question is therefore not whether a company is good but whether the investor can enter it on disciplined terms. The screen must hold two dimensions in view at once: the durability and quality of the franchise, and the valuation discipline available at the point of entry.

Franchise quality, the vertical axis of the screen, captures the durability of the company's advantage: the depth of its moat, the quality and recurrence of its revenue, its position in the value chain and the credibility of its path to sustained profitability. A frontier model laboratory with a defensible technical lead and a fast-growing enterprise revenue base scores high on this axis; a promising but undifferentiated application-layer company scores lower, because its advantage is more easily competed away. Entry discipline, the horizontal axis, captures whether the investor can buy at a price that embeds a sensible margin of safety: a meaningful discount to the last round, a valuation that does not require heroic assumptions to justify, and an illiquidity premium commensurate with the lock-up. A name can be excellent on the first axis and uninvestable on the second if the only available shares are priced at or above a stretched last-round mark.

The screen-in zone, shown in Figure 3, is the region where both axes are adequate: a franchise of genuine quality available at a price that embeds discipline. Names that are high on quality but low on discipline are not rejected outright but parked, to be revisited if and when supply appears at a better price. Names that are low on quality are screened out regardless of price, because a discount on a weak franchise is not a margin of safety but a warning. The discipline the screen enforces is precisely the discipline the literature identifies as the pre-IPO buyer's principal source of edge: it refuses to let the quality of the name override the discipline of the price.

Figure 3. Name-Screening Map: Franchise Quality against Entry Discipline

Illustrative. Each bubble is a stylised late-stage AI category; bubble size denotes relative scale. The screen-in zone (upper right) holds names of genuine quality available on disciplined terms. Positions are judgemental and not based on any specific company.

The Access Routes

The second decision, and the structural heart of the framework, is which route to use to obtain the exposure. The same economic interest in the same company can be reached through several distinct routes, and they are not interchangeable. Four principal routes, with two further hybrids, span the field.

The direct secondary is the purchase of existing shares from an existing holder, typically a former employee, an early investor or a founder seeking partial liquidity. It offers the most direct exposure and, in principle, the cleanest claim, because the buyer ends up on the company's register or its equivalent. Its drawbacks are that it is the hardest route to access, since attractive supply is scarce and relationship-gated; that it requires the buyer to clear transfer restrictions and rights of first refusal; and that it places the full diligence and execution burden on the buyer. It suits a large, well-advised investor with strong intermediary relationships and the infrastructure to execute.

The single-name special-purpose vehicle is a pooled structure assembled by a sponsor to hold shares in one company, into which several investors buy units. It lowers the access barrier, because the sponsor sources the shares and clears the transfer process, and it allows a smaller investor to participate in a name that would otherwise be out of reach. Its costs are a layer of fees and a layer of separation: the investor owns a unit in a vehicle, not the shares themselves, and the rights, information and exit terms are only as good as the vehicle's documentation. The quality of the sponsor and the terms of the structure therefore matter as much as the underlying company.

The multi-name secondary fund is a blind or semi-blind pool that buys a basket of late-stage stakes across several companies. It offers diversification, professional sourcing and a single relationship in place of many, at the cost of the least direct exposure, the longest separation from the underlying shares, and a full layer of fund-level fees and carry. It suits an investor who wants exposure to the theme rather than to a specific name, and who values diversification and delegation over directness and control.

Implementation: A Six-To-Twelve-Month Roadmap

This section turns the framework into an ordered, practical sequence. The roadmap assumes an investor moving from a standing start, with a mandate to build pre-IPO AI exposure but without an existing pipeline, and it runs over roughly six to twelve months from mandate to first close and into ongoing monitoring. The phases overlap, as Figure 7 shows, but the logic is sequential: each phase depends on the decisions made in the one before.

Illustrative timeline in months from mandate. Phases overlap but follow a logical sequence from sizing and screening through access-route selection, diligence, pricing and documentation to ongoing monitoring.

Mandate and Sleeve Sizing

The first phase fixes the boundaries of the exercise before any name is considered. It sets the total capital allocated to pre-IPO AI exposure as a sleeve within the broader portfolio, the maximum position in any single name, the acceptable holding horizon, and the liquidity the investor is willing to surrender. These constraints are not bureaucratic; they are the inputs that make every later decision tractable. An investor who has not decided how much illiquidity they can bear, or how concentrated they are willing to become, cannot rationally choose between a direct secondary and a diversified fund, because the choice depends precisely on those tolerances.

Name Shortlist and Screening

The second phase applies the name screen of Section 4.1 to assemble a shortlist. The output is not a single pick but a ranked set of candidates, each placed on the quality-against-discipline map, with the screen-in names prioritised and the high-quality but expensive names parked for later. This phase is deliberately separated from the search for supply, because conflating the two, screening only the names for which shares happen to be available, lets the intermediary's inventory drive the portfolio rather than the investor's judgement.

Access-Route Selection

The third phase matches each shortlisted name to an access route using the comparison of Section 4.3. For some names a direct secondary will be reachable; for others only a vehicle or a fund will provide access; for a few only a synthetic route may be available. The output is a route plan that pairs each target with the route best suited to the investor's constraints and to what the market actually offers for that name. This is also the phase in which the investor selects and engages intermediaries, applying Proposition 5: the brokers, bankers and platforms chosen here will shape the supply the investor sees and the terms on which they see it.

Counterparty and Provenance Diligence

The fourth phase begins the diligence stack at its foundation. Before any fundamental or price analysis, the investor establishes the provenance of the specific shares on offer: the identity and motivation of the seller, the class and rights of the shares, and whether the transfer is permitted under the company's restrictions and rights of first refusal. This phase frequently eliminates transactions that looked attractive on the surface, and doing so early, before time and money are sunk into fundamental analysis, is one of the principal efficiencies the ordered sequence delivers.

Pricing and Negotiation

The fifth phase applies the fundamental, price and structure layers of the diligence stack to the transactions that have cleared provenance, and it negotiates the entry. The discipline of Section 4.4 governs: the reference mark is the opening figure, the target is an entry that embeds both a discount and an illiquidity premium, and the willingness to walk away is the investor's strongest card. This phase is where the durable edge identified in the literature is captured or surrendered.

Documentation and Funding

The sixth phase executes the structuring and governance decisions of Section 4.6: the legal review of the vehicle and the share documentation, the tax and custody analysis, the negotiation of information and protective rights, and the funding of the purchase. This is the phase in which specialist advice is indispensable, and in which the investor ensures that the exposure documented matches the exposure intended.

Post-Close Monitoring

Conclusion

The defining companies of the artificial-intelligence cycle are staying private, and for an investor who wishes to own them the binding constraint is access rather than selection. This paper has argued that the access problem, far from being an impenetrable mystery, can be reduced to four sequential and tractable decisions: which names to screen in, which route to use, how to diligence the specific shares, and how to structure and govern the holding. The marquee names are, for the most part, good companies; the edge available to a disciplined buyer lies not in picking them but in reaching them on the right terms, at a price that embeds a margin of safety, through an instrument that confers a clean claim, sourced through an intermediary whose interests are aligned.

The framework's central counsel is to make each of the four decisions deliberately and in order, rather than allowing the first intermediary's first offer to make all four at once. Begin with a screen that holds franchise quality and entry discipline in view together. Choose the access route that matches the investor's own constraints rather than the one most readily offered. Diligence the shares before the company, because the most expensive errors are made at the level of the instrument. And structure the holding with specialist care, because the exposure that is documented is the exposure the investor truly owns. For the Gulf-based family office or sovereign-linked investor with long-horizon dollar capital and growing access to United States supply, this discipline converts a structural advantage into a durable one. The opportunity to own the AI cycle before it reaches the public market is real; capturing it well is a matter of process, and the process is the subject of this paper.

[1] Ang, A., Papanikolaou, D. and Westerfield, M. M. (2014). Portfolio Choice with Illiquid Assets. Management Science, 60(11), 2737-2761.

[2] Cumming, D. and Johan, S. (2013). Venture Capital and Private Equity Contracting: An International Perspective. 2nd ed. Amsterdam: Academic Press.

[3] Ewens, M. and Farre-Mensa, J. (2020). The Deregulation of the Private Equity Markets and the Decline in IPOs. Review of Financial Studies, 33(12), 5463-5509.

[4] Gompers, P. A. and Lerner, J. (2001). The Venture Capital Revolution. Journal of Economic Perspectives, 15(2), 145-168.

[5] Gornall, W. and Strebulaev, I. A. (2020). Squaring Venture Capital Valuations with Reality. Journal of Financial Economics, 135(1), 120-143.

[6] Kaplan, S. N. and Stromberg, P. (2003). Financial Contracting Theory Meets the Real World: An Empirical Analysis of Venture Capital Contracts. Review of Economic Studies, 70(2), 281-315.

[7] Metrick, A. and Yasuda, A. (2010). Venture Capital and the Finance of Innovation. 2nd ed. Hoboken: John Wiley & Sons.