Two Companies Took 43% of the World's Venture Capital. None of Their Investors Have Seen a Dollar of It.

The headline number is a record, and the record is the least interesting thing about it.

Global startup funding hit $510 billion in the first half of 2026, per Crunchbase — more in six months than all of 2025's $440 billion. Every trade outlet ran it as a boom story: AI is back, capital is flowing, the good times are here. What almost none of them led with is the composition. Two companies — OpenAI and Anthropic — took $217 billion of that $510 billion. Forty-three percent of every venture dollar deployed on the planet, in half a year, went to two pre-IPO firms. In the second quarter, AI took more than 70% of all global startup capital, up from roughly half a year earlier.

That is not a boom. A boom is broad. This is the opposite of broad. This is an entire asset class quietly reorganizing itself into a leveraged bet on two names, and then reporting the leverage as good news.

The number under the number

Here's the part I keep waiting for someone to say plainly, and I'll say it: the gains everyone is celebrating have never become cash.

Menlo Ventures is the cleanest illustration in the market. It put roughly $1 billion into Anthropic over the years. That stake is now marked at around $14 billion, with Anthropic valued north of $900 billion and closing on a trillion. Managing partner Shawn Carolan called it a bet-the-firm moment, and the win powered Menlo's new $3 billion fund — the largest in the firm's 50-year history. A career-defining result, a record fund raised on the strength of it.

And not one dollar of that $14 billion has been distributed to Menlo's limited partners.

It can't be, yet. The only path from a $14 billion paper mark to actual money runs through an IPO or a secondary sale at these historically elevated private valuations. Until one of those happens, the $14 billion is a number in a spreadsheet — a number set not by a public market but by the last private round Anthropic chose to price. Menlo's profile, Menlo's next fund, and Menlo's LP returns all rest on a mark that hasn't converted. That single position is the entire market in miniature: paper stacked on paper, with one narrow bridge to cash.

Now multiply it. Amazon is booking mark-to-market income off its Anthropic stake — a gain large enough to have accounted for more than half of a record quarter's profit. VC portfolios across the industry carry these two labs as their anchor marks. When 43% of global capital flows to two firms, the cross-holdings that prop up fund performance and corporate income statements are disproportionately tied to two valuations, and those valuations were set in private rounds, by the companies raising them, not by anyone who had to buy the stock on an open market.

The trap door

So what converts the paper to cash? IPOs. And this is where the structure gets genuinely precarious, because of who's holding the door.

Anthropic is reportedly targeting a listing around October 2026. Dario Amodei has reportedly said that a 12-month delay would mean bankruptcy for the company — that's how tight the runway is against the cost of scaling a negative-margin business. OpenAI, meanwhile, is reported to be eyeing a delay to 2027 rather than list below a trillion dollars against roughly $24 billion in annual recurring revenue. One company needs the IPO to survive. The other would rather wait a year than let the public price its shares right now. Those are not the postures of firms confident the private marks will survive contact with public price discovery.

And because so much rests on these two, a slip doesn't stay contained. Reprice one lab and you reprice the fund marks tied to it, the corporate income statements booking gains on it, and — this is the part the boom story never mentions — the web of circular deals underneath, the ones structured on the assumption that liquidity arrives on schedule: the 90-day compute clauses, the contingent multibillion-dollar tranches, the warrant packages that vest against milestones. The cascade runs lab → cross-holder marks → funds → LP distributions that simply never show up.

The correction that makes it worse, not better

Here's where I have to correct the lazy version of my own argument, because the obvious take — "the giants are starving the rest of the market" — is wrong, and the data says so cleanly.

The middle is not being drained. Early-stage funding was up more than 100% year over year in the second quarter. Seed stayed elevated. Sixteen companies outside the foundation labs raised billion-dollar rounds — roughly $108 billion across defense, AI infrastructure, robotics, healthcare — and more than 5,000 startups got funded in the quarter. The whole tide came in. The two labs are sitting on top of a rising market, not siphoning it dry.

And that is the uncomfortable part. A two-speed market where the giants starve everyone else would at least be idiosyncratic — a concentration you could quarantine. What actually exists is a large, broadly elevated venture market resting on two uncleared marks. That's not less systemic. It's more. If the anchor valuations reprice, the shock doesn't hit two firms. It hits an entire asset class that inflated underneath them, because the marks holding up the roof are the same two names holding up 43% of the flow.

What I think, and what would change my mind

I think the tell to watch is not the IPO announcements but the secondary market — the prices at which existing Anthropic and OpenAI shares change hands privately before any listing. That's the leading indicator of whether public price discovery will validate the marks or gut them. If secondary prices hold near the private rounds, the bullish path is real: both labs clear near a trillion, the paper converts to cash, Menlo's LPs get paid, Amazon's income is vindicated, the circular deals settle, and the record $510 billion turns out to have been underwriting something that actually cleared.

If secondaries soften before the window opens, that's the sound of the trap door. One slip or down-round resets the mark that anchors the funds and the balance sheets, LP distributions evaporate, and because these two are 43% of the flow, it's a systemic reset, not a bad quarter for two startups.

I'm not calling the top — I don't know when or whether the marks convert, and anyone who tells you they do is selling something. What I'm confident of is narrower and, I think, more useful: a record built this narrowly is not the strength it's being reported as. It's a single point of failure with a champagne label. The number to remember isn't $510 billion. It's the zero dollars that have actually been distributed on the gains everyone is celebrating.