Analysis
Count the liquidity events of the last two weeks and almost none of them involve a stock exchange. Stripe agreed to acquire OpenRouter for more than $7 billion, roughly five times the AI gateway's May valuation of $1.3 billion. SpaceX closed its acquisition of Cursor. Archer Aviation absorbed Boeing's Wisk Aero along with SkyGrid and Insitu. Joby Aviation paid $500 million for Resonant Sciences. Four exits, zero prospectuses.
The mechanism is not mysterious. Private buyers in this cycle have currency that public markets would discount. Stripe was marked around $106.5 billion in a February tender and has deferred its own listing for years; paying in a mix of cash and private stock lets it move at a price no public acquirer would defend on an earnings call. SpaceX has the same structural advantage. The result is that the most aggressive bidders for AI assets are companies that have themselves avoided going public.
Inverting the Classic Sequence
That inverts the classic sequence. The venture playbook assumed a company either compounded into a listing or sold to a public strategic at a premium anchored to public comps. In 2026 the highest bids are coming from large private companies whose valuations are set by their own primary rounds -- which are, in turn, set by the same growth-stage investors who own the targets. Nobody in that chain has an incentive to mark anything conservatively.
The scale gap is what makes the trade attractive to sellers. Pulse previously covered the warnings investors are raising about OpenAI's eventual offering, and those concerns generalize: a listing means quarterly disclosure, lockups, and a public price that can fall 40% in a quarter. A trade sale converts a mark into distributed proceeds in one transaction. For a fund needing DPI rather than TVPI, that is not a close call.
The risk sitting underneath is concentration. If four or five large private acquirers account for most AI exit value, then exit pricing depends on those specific balance sheets staying aggressive. Stripe, SpaceX, Databricks and a handful of others can each absorb one or two multibillion-dollar deals a year. A pullback from any of them removes a meaningful share of the market's bid. Antitrust is the second constraint -- the FTC and DOJ have both signaled interest in AI-stack acquisitions, and a $7 billion purchase of the routing layer between applications and models is precisely the kind of vertical deal that draws a second request.
The measurable thing over the next two quarters is the ratio: total disclosed AI M&A value versus total venture-backed tech IPO proceeds. If trades keep outrunning listings by the margin they did this month, the 2026 exit market is structurally different from the one every LP model still assumes.
The historical comparison worth making is 2013 to 2016, when Facebook and Google absorbed most of the mobile-era startup value through acquisition -- Instagram at $1B, WhatsApp at $19B, Waze at $1.1B -- and a generation of venture returns came from trades rather than listings. That period ended not because acquirers stopped buying but because regulators started scrutinizing, and the companies that were forced to stand alone produced the larger outcomes.
For LPs the practical consequence is timing. Trade sales close in three to nine months and distribute quickly; IPOs deliver over a two-year lockup and post-lockup selling window. A fund built on acquisition exits looks better on DPI at year seven and can look worse on total multiple at year twelve, because the buyer captures the compounding. Which of those a GP optimizes for is a choice worth stating explicitly in an annual letter rather than discovering at the exit committee.
