AI Venture Funding Hits Records as the Exit Market Stalls

Record money, nowhere to go

By the usual yardsticks, 2026 is the biggest year venture capital has ever had. US startups raised more than $400 billion in the first half alone, which beat every previous full-year total on record and already exceeded all of 2025 2. By the end of the third quarter, US deal value had reached $515.8 billion, about 44% above the full-year record set in 2021 1. Artificial intelligence took an unprecedented 82.7% of that money 1.

The headline totals hide a market under strain. Capital is flowing in heavily, but very little is flowing back out to investors. That imbalance is starting to shape fundraising, valuations, and the prospects of founders who are not building AI.

The exit problem

Only 18 companies went public in the US in Q3, and most of them were healthcare firms rather than the tech names that dominate private funding 1. The US now has 992 active unicorns with a combined value of $5.7 trillion, and none of them has an obvious route to liquidity 1.

This is the sharpest point of disagreement in the year's data. The NVCA's summary of the second quarter was upbeat. It described a recovering market in which improving IPO and M&A activity offered "encouraging signs" that liquidity was beginning to return 2. One quarter later, PitchBook's Q3 read was that liquidity "remains elusive" and that the exit side is "seizing up" 1. Either the Q2 improvement was a false dawn, or it was too small to dent a backlog worth trillions. Both readings point the same way. Private valuations keep rising faster than public markets can absorb them, and unless that changes, many of those unicorn marks will be hard to defend at exit.

Concentration everywhere

The absence of exits has a direct effect on fundraising. Limited partners who have not received distributions are going back to the managers who have returned cash before. Established firms have taken 88.2% of all capital raised so far this year, the highest share in PitchBook's dataset 1. Andreessen Horowitz alone closed $23.8 billion across several funds, more than a fifth of all US venture fundraising for the year 1. The NVCA also noted that both investment and fundraising remain concentrated among a relatively small number of companies and funds 2.

The same pattern appears on the investment side. Crunchbase's Q3 figures put global venture funding at $159 billion. AI accounted for $102 billion of that, or 64% 3. Just 27 companies captured roughly a third of all capital, each through a single round of $1 billion or more 3.

A quarterly dip inside a record year

The Crunchbase numbers also show a slowdown. North American funding fell 35% to $92 billion in Q3, which the data attributes to a pause in the OpenAI- and Anthropic-sized mega-rounds that had inflated earlier quarters 3. This does not contradict PitchBook's year-to-date record 1. PitchBook measures cumulative US totals, while Crunchbase measures one quarter globally. Together they show how much the year's figures depend on a few enormous financings. When those deals take a break, quarterly totals fall sharply even while the annual figure stays at a record.

The squeeze below the mega-rounds

The data on ordinary startups is the most concerning part. The share of seed-stage companies that go on to raise a Series A has dropped from 50% to 38%, and the typical gap between those rounds has grown to 616 days 3. Valuations have also split by sector. The median pre-money valuation for a non-AI Series A is $42.4 million, while AI rounds sit at $78 million, a premium of about 84% 3. While a few companies raise ever-larger rounds, founders outside AI face a widening financing gap after the seed stage 3.

What it adds up to

Taken together, the numbers describe a two-tier market. At the top, a small group of AI companies and the established firms backing them are pulling in capital at record levels. Below them, fewer companies are graduating to later rounds, the wait between rounds is getting longer, and non-AI startups are valued at a steep discount to AI peers. Holding the whole structure up are nearly a thousand unicorns that cannot easily go public or sell.

This is an interpretation rather than a forecast, but the setup looks fragile. Record deployment only produces returns if exits eventually follow. Until IPO and M&A markets reopen at scale, LP money will keep consolidating around the largest managers, and capital will keep favoring the biggest AI names over everyone else. Q2's signs of returning liquidity 2 matter less than whether Q4 brings any of them back.