AI takes two-thirds of venture money, and your odds are still one in six

The Study · Economics

AI takes two-thirds of venture money, and your odds are still one in six

Akira Sumi · Jul 10, 2026 · 13 min read · researched

AI companies took 65% of US venture dollars in 2025, and almost none of it landed where a small team stands. A seed check costs about 20% of the company per round and buys entry to a funnel where the Series A bar sits near $3 million in revenue and roughly one seed company in six clears it within two years. Here is the decision math, then the playbook for the builders whose math says raise.

February 2026 was the biggest month venture capital has ever had: $189 billion of global funding, per Crunchbase. Three companies took 83% of it. OpenAI raised $110 billion, Anthropic raised $30 billion, Waymo raised $16 billion, and every other startup on earth split what was left. That month is the AI funding boom in a single frame: the records are real, the dollars are real, and the money is pooling somewhere you are not standing.

Because here is the same year from where a one-or-two-person AI product stands. US seed rounds between $200K and $5 million fell roughly 20% in 2025, in both count and dollars. The share of seed money going into $10M+ mega-seeds hit 51%. And the venture funds themselves raised $118.6 billion globally, the lowest in a decade. The boom moved the ceiling, not the floor. So the question “how do I raise venture money for my AI product” has a step zero that most guides skip: whether the trade on offer is one you should take. This piece runs that math first, with the published odds on every branch, and then the mechanics for the builders whose answer is yes.

The short version

If you are deciding whether to pitch anyone at all:

The boom is not looking for you. AI took 65.4% of US venture deal value in 2025, but the gains concentrated in megadeals; small seed rounds shrank.

Know the price. A median seed sells about 20% of the company; by Series A the median founding team holds 36%.

Know the odds. Since late 2021, 13 to 18% of seed cohorts reach a Series A within two years, and the bar sits near $3 million in revenue.

The other branch has receipts now. Base44 sold for $80 million cash six months after launch, bootstrapped. Gamma got profitable first and raised later, on its own terms.

Build distribution either way. It is the one asset that pays on both branches: customers if you keep the company, term sheets if you sell a piece of it.

The rest is the evidence, and the playbook.

The boom is a barbell

Start with the shape of the money, because the headline number is misleading on its own. AI and machine learning companies took 65.4% of US venture deal value in 2025, $222.1 billion of $339.4 billion, up from 46.4% the year before, per the PitchBook-NVCA Venture Monitor. By the first quarter of 2026 the share hit 88.8%. Two-thirds of the asset class now flows to one theme, which sounds like the easiest fundraising climate an AI founder could ask for.

The distribution says otherwise. Crunchbase’s year-end data shows just 68 companies, each raising $500 million or more, absorbed over a third of all global venture funding in 2025, up from 24% the year before. Meanwhile the small end contracted: the classic $200K-to-$5M seed round fell by a fifth, and more than half of US seed dollars went into rounds of $10 million and up, deals that look like small Series As and mostly go to second-time founders and hot labs. The barbell has a fat end and a thin end, and a new founder with a working product is standing at the thin one. None of this means you cannot raise. It means the boom did not improve your odds, and the numbers that follow are the ones that actually apply to you.

What a check actually costs

The sticker price of a seed round is published, and it is remarkably stable. Across 2025, the median priced seed on Carta was $4.0 million raised at a $20.0 million post-money valuation: founders selling right around 20% of the company. AI carries a premium, $19 million median pre-money against $13 million for everyone else in early 2025, per Carta’s Peter Walker, but the premium changes the valuation, not the fraction sold. Stack the rounds and the arithmetic compounds: on Carta’s data, the median founding team holds about 56% of the company by the seed round and 36% by Series A. Before most companies have found durable revenue, the founders are minority holders as a group.

The second cost is the treadmill. Venture pricing assumes venture growth, and the AI-era benchmark is steeper than the old SaaS one. Bessemer’s State of AI 2025 proposes Q2T3 in place of T2D3: quadruple, quadruple, then triple for three years, with its “Shooting Star” archetype reaching roughly $3 million ARR in its first year of revenue. Take the check and that curve becomes the plan you are graded against.

The third cost is the one founders feel last: the fund’s math is not your math. Across more than 21,000 US venture financings, 65% failed to return even 1x and only about 4% returned 10x or more, per Correlation Ventures data reported by VC Seth Levine (2004 to 2013 vintage; the firm’s later update through mid-2023 shows the same skew slightly softened). A fund built on those odds needs its winners enormous. A $20 million exit would change your life; on a fund’s spreadsheet it rounds to zero, and the fund is in your ear at every fork. That mismatch, not any villainy, is why funded companies get pushed to swing bigger than their founders sometimes want.

Should you even raise: the decision math

Now the funnel itself. On Carta’s cohort data, about 30% of US startups that raised a priced seed in 2018-19 reached a Series A within two years. Every cohort since the second half of 2021 that has completed its two-year window graduated at 13 to 18%, roughly half the old baseline. Precursor’s Charles Hudson called the reset “real and permanent” back in 2023, and the completed cohorts keep proving him right, though for fairness the newest data points the other way at the margin: Walker notes the early-2025 seed cohort is graduating faster in its first year (10 to 11%) than the 2022-23 cohorts did at the same age. Call the odds one in six, maybe improving toward one in five.

The bar you have to clear got taller too. The median Series A company was doing nearly $3 million in ARR in 2024, per SVB data cited by Walker, up from about $1 million in 2021, and by early 2026 it had crept toward $3.5 million. The median gap from seed to A stretched to 2.1 years, from about 1.6. And the downside is not abstract: 966 startups on Carta shut down in 2024, the highest in its published data, and Walker himself says the true count runs higher.

Against that, the branch nobody modeled five years ago now has receipts. Maor Shlomo built Base44 solo, bootstrapped, in public. By his own account it went from zero to $1 million ARR within about three weeks of turning on monetization, and six months after launch, with 250,000 users, Wix bought it for roughly $80 million in cash, with milestone payments that Wix’s own reporting later put on track for $90 million more. Gamma ran the hybrid: profitable since early 2024, $50 million ARR with about 30 people on just $23 million ever raised, and then, past $100 million ARR, chose a $68 million Series B at a $2.1 billion valuation. Raising from strength, after profitability, is a different product than raising from need. Both stories are founder-reported and both are outliers, the same way funded unicorns are outliers. But they price the alternative: an outcome you keep all of, reached on revenue instead of runway.

So the decision test is three questions, not a vibe. One: does your market punish patience, the way Stripe’s data suggests the current window does, with the top 100 AI companies on its platform hitting $1 million in annualized revenue in a median of 11.5 months? If competitors will Q2T3 past you while you compound slowly, capital is a real weapon. Two: does the product itself need money revenue cannot supply, training runs, inventory, compliance, a sales motion, or is it software a small team can ship, the kind AI has made radically cheaper to build? Three: is a one-in-six shot at the venture path, priced at 20% a round, worth more to you than a Base44-shaped outcome you own outright? Two or three yeses, raise. Zero or one, the rest of this piece is cheaper than a board seat.

If you raise: the instrument and the funnel

The paperwork question is settled, so do not relitigate it. On Carta, SAFEs are essentially the whole pre-seed market: 93% of pre-seed rounds in Q1 2026, with convertible notes at a record-low 7%. The standard deal is a post-money SAFE with a valuation cap and no discount, about six in ten SAFEs in Carta’s data. In 2025 the median caps ran around $10 million for rounds raising $250K to $1M and $15 million for $1M to $2.5M rounds; the smallest rounds, under $250K, carried a $7.5 million median cap as of mid-2025. Those three numbers are your sanity check: an AI pre-seed with traction can price above them, but if an angel offers terms far below, you now know what the market clears at.

Know the reference deal too. YC’s standard offer, unchanged since 2022 and still live as of this writing, is $500K total: $125K on a post-money SAFE for a flat 7%, plus $375K on an uncapped SAFE with a most-favored-nation clause. The MFN is the detail founders miss: that second tranche converts at the terms of the lowest-cap SAFE you sign between the start of the batch and your next priced round, so a small cheap SAFE signed casually mid-batch reprices YC’s $375K along with it. YC’s own example puts the all-in at roughly 9.5% for a company that later raises at a $15 million cap.

On effort, plan for a campaign, not a coffee chat. In DocSend’s 2023 seed report, the last edition to publish full funnel averages, founders contacted an average of 66 investors to get 38 meetings, and half of successful raises took 13 to 24 weeks. Its December 2024 report showed the market easing, with most successful seed rounds closing in 12 weeks or less. Split the difference and budget a quarter of full-time work, which for a solo founder is a quarter of no shipping. That, more than the dilution, is the cost that kills small teams: the product stalls while you pitch.

What investors screen now that demos are free

The pitch itself changed when the demo stopped being evidence. Alison Imbert, a partner at Partech, put it flatly in April 2026: “The demo is worth nothing. It’s so easy to build.” What she screens for instead is documented customer research, her first question being how many customers a founder interviewed in the last two months, and she wants the messy notes, plus a real distribution plan. A working AI prototype was a signal in 2023. In 2026 it is table stakes, and the scarce evidence is that specific people with a specific problem talk to you.

The spending screen is just as concrete. Speaking to TechCrunch in late 2024, Everywhere Ventures’ Jenny Fielding put a reasonable pre-seed founder salary at $85,000 to $125,000, and described killing a deal after finding a founder paying himself $300,000; C-suite titles like COO and CFO at that stage read as overhead. And size the raise for the real timeline: with the median seed-to-A gap at 2.1 years, the old 18-months-of-runway plan is undersized before you sign it.

Marketing yourself: the asset that pays on both paths

This part comes last because it is the connective tissue between the branches. “Market yourself” sounds like a soft skill; in this market it is a priced asset. Base44’s entire go-to-market was Shlomo narrating the build on LinkedIn and X, zero paid marketing by his account, and the audience did two jobs at once: it delivered users, and it made the company legible to an acquirer in month six. Investors now underwrite the same thing explicitly. When a16z’s Bryan Kim led Cluely’s $15 million Series A, his stated thesis was that for consumer

[truncated for AI cost control]