Where the Money Went

Everyone in the business knew it. The fund counts say it out loud.

Funds closed fell from 1,838 in 2022 to 1,094 in 2025. First-time funds fell from 479 to 195, and only 81 have closed in the first nine months of 2026, on pace for the fewest since 2016. The dollars came back this year, but look where they went: funds of $500M or more took 78% of the money while making up 6% of the funds, established firms took a record 88%, and Andreessen Horowitz alone raised $23.8B, more than a fifth of everything. Call that a consolidation, not a recovery.

• 1.6×Capital called vs. distributed by U.S. venture funds since 2022
• −$200BNet cash flow to LPs since the slowdown began in 2022
• 88%Share of 2026 fund capital raised by established firms, a record

Underneath is the part that matters. Since the start of 2022, managers have called 1.6× more capital from LPs than they have given back; in the ten years before, they returned 1.3× what they called. No exits, no distributions. No distributions, no re-ups. No re-ups, and only the biggest names survive the next raise.

Lock money up for ten years, pay two and twenty, and you deserve a premium over an index fund. You didn't get one.

The usual hurdle is three to five points a year over public stocks. The Kauffman Foundation ran that test on its own portfolio in 2012: 20 of 100 funds cleared three points, and 62 trailed the public markets outright after fees. They titled the report “We Have Met the Enemy… and He Is Us.” Nobody listened. By Cambridge Associates' index, venture returned 13.1% a year over the ten years to June 2025 against 13.7% for the S&P 500 on the same cash flows, and 0.1% against 19.8% over three. Only the 15- and 20-year windows show a premium, and the larger is about one point, not three.

Fund by fund it's worse, because paper isn't cash. Carta tracks about 2,900 funds raised since 2017. The median 2017 fund says it is worth 1.89× what LPs paid in and has handed back 0.37×, nine years in. The median 2019 fund has returned 0.04×. The median 2021 and 2022 funds have returned nothing, and fewer than a quarter of 2022 funds have made a single distribution. The rest is marks, set by the people whose fees depend on them.

Nobody publishes a standard deviation for venture funds, because the returns aren't a bell curve. They're a power law with a long right tail, and a standard deviation would only tell you the tail exists. What gets published is percentiles. In the 2019 vintage the bottom quartile sits at 1.02×, the median at 1.33×, the top quartile at 1.9× and the top decile at 3.01×: the gap from the 75th to the 90th percentile is wider than the gap from the 25th to the 75th. Three times paid-in capital is the mark of a fund that earned its risk, and Carta itself calls it the threshold for exemplary performance. One fund in ten is on track. In the 2017 vintage the 75th-percentile fund sits at 2.53×, still under the bar, and by IRR no vintage since 2017 has a 75th percentile above 15.5%. The winners win enormously. Everyone else should have bought the index.

Averages hide the only funds worth being in. Here is what the good ones made, and what the famous ones have told their pension LPs.

Line up thirteen vintages and the pattern never breaks. The top-quartile fund beat the S&P 500 on its own cash flows in all thirteen: by 2 to 5 points a year in the 2005 to 2009 vintages, and by 7 to 19 points in every vintage from 2010 on. The median fund trailed the index in every vintage from 2005 to 2009 and again in 2017, and beat it by one to six points in 2010 through 2016. The bottom-quartile fund lost to the index in all thirteen, by 1 to 14 points. That spread is the whole business. Cambridge's standard deviation of fund IRRs runs 12 to 29 points per vintage, and 49 in 2013, which is why nobody quotes it: the number is dominated by a handful of funds that found one company. For scale, the ten years to mid-2025 paid 13.7% a year in the S&P 500, 16.3% in the Nasdaq and 7.1% in small caps, and a 10-year Treasury yielded about 4%. The top quartile of every vintage since 2007 cleared all of it. The median cleared the bonds.

So are the names different? Yes, in the way you'd expect. Andreessen Horowitz's 2012 fund is marked at 11.3×, 9.1× with its parallel fund. Menlo's 2011 fund has returned Washington State 4.3× and a 29% IRR, most of it in cash. Khosla's 2011 and 2014 funds sit at 4.1× and 3.4×. Benchmark, which reports to no public pension, is said to have returned 7.5× net across its first eight funds. The premier firms earned the premium. Now read the dates. Every fund above 3× was raised in 2015 or earlier, with two exceptions: Sequoia's 2018 venture fund at 4.55× and a16z's 2019 late-stage fund at 3.3×, while Sequoia's 2020 growth fund sits at 1.1×. Menlo's 2015 fund is at 1.8× after a decade. The 2021 funds of a16z, TCV and Insight are at 1.1× to 1.4×, which is what everyone else's 2021 fund looks like. A brand gets you into the deal. It does not repeal the vintage.

Flood a market with buyers who can't price what they're buying and prices go up. 2021 ran the experiment.

Public software is the clean read, because it reprices every day. The median public cloud company sold for 20× next year's revenue in February 2021, 4.9× by November 2022, and 4.1× in January 2026, a ten-year low. An 80% cut in the price of a dollar of software revenue. The private market mostly pretended not to notice.

Who were the buyers in 2021? Mostly not venture firms. PitchBook counts deals with “nontraditional” investors in them: hedge funds, mutual funds, crossover asset managers, sovereign wealth funds, corporates. They were in $127B of U.S. venture deals in 2020 and $255B in 2021. Tiger Global raised a $12.7B fund and made 315 investments in twelve months, nearly one a day, and by mid-2024 that fund sat in the bottom tenth of its vintage. PitchBook's verdict in its latest report: nontraditional activity, “especially by more specific groups like hedge funds, was an opportunistic bet in 2021 that did not pan out.” When rates rose the tourists left and took the bids with them. That is the link between oversupply and inflated prices: a mob with a checkbook and no discount rate.

• $127B → $255BU.S. venture deals with hedge funds, crossovers, corporates or sovereign funds in them, 2020 to 2021
• 315Investments Tiger Global's $12.7B 2021 fund made in twelve months
• Bottom 10%Where that fund ranked among 2021 venture funds by mid-2024

The private market is doing it again with a narrower cast. Median pre-money valuations rose at every series but seed this year; Series D+ doubled, from $832M to $1.72B, and AI companies step up 2.2× between rounds against 1.5× for everyone else. Meanwhile the companies that raised at 2021 prices are learning what their shares are worth. Sellers on Forge who last raised in 2021 take a median 59% discount, and Miro and Airtable, once $17.5B and $11.7B, sold at discounts of nearly 90%. Public software at a ten-year low and private AI at records can't both be right.

• −59%Median discount on Forge for companies that last raised in 2021
• −45%Median discount for companies that last raised in 2023
• 2.2× vs. 1.5×Median valuation step-up between rounds, AI vs. everyone else, 2026

The surviving capital isn't spreading out. It's piling into a handful of foundation-model companies.

AI took 64% of U.S. venture dollars in 2025 and 83% in the first nine months of 2026: $427B, more than 2022, 2023 and 2024 combined. Worldwide, OpenAI ($122B), Anthropic ($30B), xAI ($20B) and Waymo ($16B) absorbed nearly 65% of all venture investment in Q1 2026 alone. Four cap tables and a rounding error. Deal counts kept falling while dollar totals set records; in Q1 2026 North America's venture dollars rose 190% year over year while its deal count fell 26%.

This is the squeeze a founder feels, not just an LP.

Seed rounds got bigger: the median U.S. seed is about $3M, three times the 2018 level. The odds of graduating collapsed. Of the companies that raised $1M+ seeds in 2023, 24% have reached a later round; for 2024 it's about 16%, against 55% or more for every cohort through 2020. The 2024 cohort is young, but 2023's has had the time earlier cohorts needed. Seed to Series A now takes more than two years and $2–4M of ARR, up from about $1M. Bigger checks, worse odds: what happens when the seed round becomes a product somebody sells.

The gains land almost entirely in the Bay Area. Being there helps in a way the data can show, and being American doesn't.

In 2025 the Bay Area took 41.3% of U.S. venture dollars, more than the next seven markets combined. In Q1 2026, heavy with AI mega-rounds, San Francisco alone took 65.9% and the wider Bay Area 82.6%, records since at least 2014. New York, long the clear number two, fell to 4% that quarter.

Does being there improve access to capital, or just the odds of a headline? The Bay Area and New York together closed 34% of U.S. deals in Q3 2026 and 53% of the dollars. At early stage, the Series A and B rounds that decide whether a company lives, the two hubs took 40% of the deals and 60% of the dollars. Hub companies raise more per round at every stage. Whether they raise more often, the data can't say; nobody publishes a seed-to-A graduation rate by region. So the honest claim is this: the anecdote fits the data, and the mechanism is proximity. A founder in San Francisco can pitch forty funds in a week without boarding a plane.

And the money does not care where the founder was born. The National Foundation for American Policy traced the founders of America's 775 private unicorns and found birthplaces in 76 countries, India first, then Israel, the UK, China and Canada; a quarter of the companies have a founder who arrived as a student. That isn't a diversity statistic. It's a pragmatism statistic. Whatever else the industry gets wrong, at the moment of writing the check it backs whoever can build the thing. That is the best evidence in this piece that the system can still price something correctly. Policy that makes it harder to come here, or to stay, overrides the market's own judgment, whatever the people selling it say.

Against 2021, only the U.S. is above water, and only because of AI. China is still 28% below but has nearly doubled since 2025. Europe is edging up. The rest of Asia is down by half.

On PitchBook's count, U.S. venture fell 11% from 2021 to 2025. Through September 2026 it is already 44% above 2021's full year, and the annualized pace is 92% above, with two frontier-lab rounds in the first half doing most of it. Strip out AI and it runs 49% below 2021, in line with everyone else. Europe was 36% below 2021 in 2025 and is 14% below on this year's pace, with AI and defence doing the lifting. Latin America's $16B in 2021 became $4.5B in 2024 by LAVCA's count, and Southeast Asia lost 79% from 2022 to 2024.

China is the one I nearly got wrong in both directions. It is not back: on the same count it fell from $149B in 2021 to $55B in 2025, and this year's pace of about $108B is still 28% short of 2021. It is also not dead, which is what the Western numbers through 2025 implied. It is the only region besides the U.S. that is climbing, and climbing fast: $81B in nine months, nearly double 2025 and level with all of Europe, on DeepSeek's $7.4B, Moonshot's $3.5B and Kling's $3B rounds, with robotics and semiconductors behind them. And the Western trackers undercount it badly. CVSource, which counts the yuan-denominated and state-backed money, put investment in the first eleven months of 2025 at 1.19 trillion yuan, up 32%, with a new 1-trillion-yuan national guidance fund behind it and Hong Kong on pace for its biggest IPO year ever. Outside China, Asia was 58% below 2021 last year and is 42% below on this year's pace. There is no global venture recovery. There is one country above 2021, one climbing back toward it, and in the second the biggest LP is the government.

• −49%U.S. venture dollars outside AI, 2026 pace vs. 2021
• $55B → $108BChina venture deal value, 2025 to the 2026 pace, still 28% under 2021 (PitchBook via KPMG)
• ¥1.19TChina VC and PE investment, January to November 2025, up 32% (CVSource)

The region didn't cool. Funding collapsed, and what's left pools in one city-state.

• $10.1B → $2.2BSoutheast Asia tech funding, 2022 to 2024
• 92% → 96%Singapore's share of regional funding, H1 2025 to January 2026
• −85%Indonesia's own startup funding, 2023 to 2025

Regional tech funding fell 79% from 2022 to 2024, and H1 2025 was the weakest half-year in seven years. Singapore took 92% of what was left in H1 2025 and 96% by January 2026, with under 6 million people in a region of nearly 700 million. Indonesia, the region's largest economy, went from 42% of regional funding in 2021 to 8%, and 2025 corruption convictions tied to state-backed venture investments told international LPs exactly what they feared. One analyst's framing: Singapore risks becoming “just an expensive address,” where capital lands and goes nowhere. That's less timidity than a capability gap: regional startups haven't shown they can turn product-market fit into paying customers at the rate investors now demand, and the pan-Asia mega-funds that absorbed about 85% of Asia-Pacific private equity capital in H1 2026 can put the same dollars into China, India, Japan or Korea instead.

Seattle didn't break. It never had a foundation-model company to ride the AI wave, and the wave is now the whole ocean.

• $4.5B → $2.7BSeattle-area venture dollars, H1 2025 to H1 2026
• 5th → 7thSeattle's rank among the top 10 U.S. metros by venture dollars
• 9.9%Washington's new top capital-gains tax rate, cited as a headwind

Seattle-area startups raised $2.7B across 163 deals in H1 2026, down about 40% from a year earlier. Among the top 10 U.S. metros by venture dollars, Seattle slipped from 5th to 7th in a single year and ranked last by deal count. Its biggest deals were in fusion energy, space and cybersecurity: real businesses, outside the one thesis absorbing the pool. Washington's new capital-gains tax and proposed taxes on startup exits add a headwind for anyone deciding whether to stay.

So what should a founder do?

• Move to San Francisco? If you're raising from venture, yes, at least for the raise. The two hubs take 60% of early-stage dollars for 40% of the deals, and the mechanism is proximity, not magic. If you're not raising, your rent is the only thing that cares where you live.
• Build a foundation model? No. Four companies took 65% of the world's venture money in one quarter, and fifth place is a consolation prize you won't be raising. Build on top of them, or build the business that gets cheaper because they exist. AI-adjacent companies step up 2.2× between rounds against 1.5× for everyone else; adjacency is priced, the frontier is a capital game.
• Raise the seed like it's the last money. Sixteen percent of 2024's seeds have reached a later round, the Series A bar is $2–4M of ARR, and the clock now runs past two years. Raise what gets you to $2M of revenue, not what gets you a party.
• Ask the VC for their DPI. The median 2019 fund has returned 0.04×. A fund with no cash back needs your markup more than your exit, and will push you to a price you can't grow into. Sellers of 2021 paper on Forge are taking a 59% haircut. If they laugh at the question, that's your answer.
• Price for the market you'll meet, not the one you're in. Public software sells at 4× revenue. A Series D at double its 2025 price is borrowing from an exit that hasn't happened.
• Not in AI, not in San Francisco? Then you're raising in a 2019 market with 2026 rent. U.S. venture dollars outside AI are running at almost exactly their 2019 level. Plenty of companies got built in 2019. They just didn't get built at 20× revenue.
• Your passport doesn't matter to the check. Your visa does. Sort the visa.