// BLOG / AI COMMERCIALIZATION
What Investors Look For in an AI Startup at Seed and Series A
Short answer. At seed, investors are paying for a founder and a wedge worth a bet. At Series A, they want real revenue, gross margin that is moving the right direction, retention that survives past the free trial, and a team small enough to prove the unit economics before you scale them. In 2026 the AI premium is real, and for most founders building an application on top of someone else's model, it is mostly not for you.
More venture money is going into AI right now than at any point on record. And almost none of the headline numbers describe the company you are actually building. The gap between the AI money that gets written about and the AI money you can actually raise is where founders get their expectations wrong.
The two-tier market: why AI money looks easy and mostly is not
In Q1 2026, more than 60 cents of every venture dollar tracked on Carta's platform went to AI companies, the highest share on record. That headline sounds like a flood. It is not evenly distributed.
An AI foundational model startup at Series A can raise at a median valuation around $300M, while a non-AI startup at the same stage sits closer to $55M.
That $300M figure belongs to a narrow cohort: foundational model and infrastructure teams with the kind of technical defense that a normal application company, built on top of someone else's model, does not have and does not need.
If you are building a product on top of an existing model, foundational-model comps are the wrong benchmark for your raise, your valuation, and your revenue bar. The number that matters for you is the one below.
What do investors look for in an AI startup at seed?
The seed stage got bigger and more selective at the same time. Median seed post-money valuation hit a record $24M in Q4 2025, up from $18M a year earlier and $16M two years before that, according to Carta. The median seed round raised $4M at that $20M post-money mark, with dilution running around 19.5%.
AI is where seed money concentrates. AI startups captured roughly 42% of all seed capital in 2025, and at pre-seed, AI crossed 50% of dollars in Q1 2026, up from roughly 30% a few years earlier, per Carta's data. There is a real exception at the very top: Peter Walker, Carta's Head of Insights, has described elite AI infrastructure and foundational model teams raising seed rounds at $160M to $200M post-money, a number that exists to distort your benchmark, not to inform it.
Two numbers matter more for a normal founder. Median seed headcount is down to four employees, not counting founders, from six or seven a few years ago, Walker reported. And the one-year seed-to-Series-A graduation rate for the 2025 cohort rose to roughly 10 to 11%, up from 4 to 5% in 2022 and 2023. That is still a small minority. A seed round is not a bridge to Series A on a fixed timeline. It is runway to prove something specific, and most companies need more than a year to prove it.
What do investors look for in an AI startup at Series A?
Series A got more expensive and more selective too. Median post-money valuation reached $78.7M in Q4 2025, up 37% year over year from $57.5M, and the median raise moved to roughly $13M to $15M, up from $8M to $10M a few years earlier, per Carta and Walker's commentary.
The revenue bar is the one figure worth getting exactly right, because it is also the one most commonly misquoted.
Silicon Valley Bank's own State of the Markets report puts median Series A revenue at $2.5M in 2025.
That is total revenue, recurring plus one time sales, not ARR, and it is not a Carta figure.
The version circulating as "$3.5M median ARR" conflates a different chart in the same report and mislabels the metric. Get the label wrong in a pitch and a sharp investor will notice before you finish the sentence.
Time is also working against founders. The median wait between seed and Series A reached 616 days by the middle of 2025, the longest on record at the time, and multiple industry trackers since have described it stretching further. Plan your seed runway assuming Series A takes longer than you think, because the data says it now usually does. Median Series A headcount has fallen to roughly 15 to 17 employees, down from about 45 a few years ago, so a lean team at Series A now reads as discipline, not as a warning sign.
| Benchmark | Seed | Series A |
|---|---|---|
| Median post-money | $24M (Q4 2025 record) | $78.7M (Q4 2025, up 37% YoY) |
| Median round size | $4M | $13M to $15M |
| Revenue bar | Early signal, not revenue | $2.5M median total revenue |
| Median team size | 4 employees | 15 to 17 employees |
| AI share of dollars | ~42% of seed, 50% of pre-seed | 60%+ of all VC dollars, Q1 2026 |
Sources: Carta, "State of Private Markets: Q1 2026" and "Record-setting early-stage valuations"; Carta, "State of Seed 2025" and "State of Pre-Seed: Q1 2026"; Silicon Valley Bank, "State of the Markets Report H1 2026"; Peter Walker (Carta), the Product Market Fit Show, Q1 2026 episode.
The four things investors actually look for
Strip away the stage-specific numbers and the same four things get evaluated at every raise. Get these four right and the round gets easier. Get one wrong and no valuation fixes it.
1. Real revenue, correctly labeled. Total revenue and ARR are different numbers, and conflating them in a data room is the fastest way to lose an investor's trust. Builder.ai is the cautionary tale. The company raised more than $445M from backers including Microsoft, the Qatar Investment Authority, and Insight Partners, and hit a valuation near $1.5B. Then, in May 2025, it entered insolvency after restating its revenue. The 2024 figure dropped from a reported $220M to roughly $55M. The 2023 figure dropped from $180M to roughly $45M, tied to alleged reciprocal invoicing with a partner that has publicly denied it. Nobody in that story had a technology problem. They had a revenue quality problem, and it ended the company.
2. Gross margin moving in the right direction. ICONIQ's research puts average AI product gross margin at 41% in 2024, 45% in 2025, and a projected 52% in 2026, well below the 75 to 85 percent range mature SaaS companies run. The reason is inference. Model inference alone runs about 20% of total product cost pre-launch and climbs to roughly 23% at scale, the opposite of classic SaaS, where cost typically shrinks as a share of revenue over time. Investors now expect that curve to be engineered, priced, and routed on purpose, not assumed the way software margin used to be.
3. Retention that holds past the free trial. ChartMogul's study of 3,500 companies found AI native products have a median gross revenue retention of roughly 40%, and the number splits hard by price. AI products under $50 a month retain only about 23% of revenue and 32% of customers a year out. Products over $250 a month retain closer to 70% and 85%. The good news: AI native retention climbed from about 27% in January 2025 to 40% by September 2025, as early tire kickers get replaced by customers who actually depend on the product. Deal size and workflow depth predict which side of that split you land on.
4. Capital efficiency, not headcount. Bessemer's research on roughly twenty high growth AI companies found two credible paths. "Supernovas" hit about $40M ARR in year one and $125M by year two, running around $1.13M in ARR per employee, but often at only 25% gross margin, sometimes negative, trading margin for speed. "Shooting Stars" grow more slowly in year one, around $3M ARR, then compound on what Bessemer calls a Q2T3 path (quadruple, quadruple, triple, triple, triple) to roughly $100M ARR by year four, while holding close to 60% gross margins. Separate research from Kyle Poyar and High Alpha, covering 800-plus companies, puts best in class ARR per employee at $350K in the $20M to $50M ARR band and $400K above $50M, and finds AI native companies at the $1M to $5M ARR stage growing a median of 110%, against 40% for traditional SaaS at the same stage. ICONIQ's research cites Cursor reaching roughly $100M ARR with about 19 employees as the extreme version of that efficiency. A lean team with real revenue per head is now a positive signal at Series A, not a red flag. At that size, nobody else owns whether the AI turns into money. You are your own Chief AI Officer.
Investors are not funding what your product can do. They are funding what your revenue, margin, and retention already prove it is worth.
Why the bar rose
None of this is investors turning cautious out of fear. Down rounds fell to 11.4% in Q1 2026, back in line with 2019 and 2020 levels, so the broader market has actually healed. What changed is the evidence bar, because the buyers on the other side of your revenue have not delivered yet. BCG found 74% of companies have not shown tangible value from AI, with only 26% past proof of concept. McKinsey's most recent survey found 88% of organizations use AI somewhere, but only 39% report any enterprise-level profit impact at all, most of that under 5%, and only about 6% qualify as genuine high performers. A widely cited, preliminary MIT study put the share of enterprise generative AI pilots with no measurable financial return near 95%. That backdrop is about enterprise buyers struggling with adoption, not about your startup, but it is exactly why the investors writing your check now demand proof your revenue is real before they believe your buyer will keep paying.
What actually de-risks an AI startup
Given all of that, here is what you actually control. Build on proprietary data the frontier labs cannot reach, not on a thin wrapper around a general model, because that data is the thing a bigger company cannot copy by matching your prompt. Wire the product deep enough into a customer's workflow that ripping you out costs more than renewing you, since workflow depth is what turns a tool into a system of record. Keep the team lean and the revenue per head high, the way Bessemer's Shooting Stars and Poyar's efficiency data both reward. And treat retention, not growth, as the headline metric you report, because a chart that spikes and churns reads as a demo now, not as a business.
Frequently asked questions
What do investors look for in an AI startup?
At seed, investors look for a founder and a wedge worth a bet, shown through a small, efficient team and early signal, not a mature business. At Series A, they look for four things: real revenue that is total revenue, not a rounded up ARR estimate, gross margin that is moving up as the product scales, retention that survives past the first free trial, and a team small enough to prove the unit economics before you scale them.
How much revenue do you need to raise a Series A in 2026?
SVB's own benchmark puts median Series A revenue at $2.5M in 2025, and that figure is total revenue, recurring plus one time sales, not ARR. The middle of the market clusters between roughly $1M and $6.5M. Investors read this number alongside gross margin and retention, so a company with $2.5M in revenue and strong retention reads better than one with $4M and none.
What is a good gross margin for an AI startup?
ICONIQ's research puts average AI product gross margin at about 45 percent in 2025, projected to climb to roughly 52 percent in 2026, well below the 75 to 85 percent range mature SaaS companies run. Model inference alone runs 20 to 23 percent of total product cost and rises as the product scales. Investors expect that margin to be engineered on purpose, not assumed.
Why does AI have worse retention than SaaS?
Because AI products are easy to try and just as easy to cancel. ChartMogul's research on 3,500 companies found AI native products priced under $50 a month retain only about 23 percent of revenue and 32 percent of customers a year later, while products priced above $250 a month retain closer to 70 and 85 percent. Deeper workflow embedding and higher deal size predict which AI revenue actually sticks.
Sources
Valuation, round size, and headcount data: Carta, "State of Private Markets: Q1 2026" (published May 29, 2026); Carta, "Record-setting early-stage valuations"; Carta, "State of Seed 2025"; Carta, "State of Pre-Seed: Q1 2026"; Carta, "Series A Funding Slides in Q2 2025" (the 616-day figure is dated to that report and has likely lengthened since). Seed and Series A headcount, elite seed valuations, and graduation rate commentary: Peter Walker, Head of Insights at Carta, speaking on the Product Market Fit Show, Q1 2026 episode, a podcast interview rather than a published Carta chart. Revenue benchmark: Silicon Valley Bank, "State of the Markets Report H1 2026." Gross margin and inference cost: ICONIQ Growth, "State of AI: Bi-Annual Snapshot" (January 2026, survey of roughly 300 software executives). Growth archetypes: Bessemer Venture Partners, "The State of AI 2025" (August 2025). Retention: ChartMogul, "The SaaS Retention Report: The AI Churn Wave" (2025, data through September 2025). Capital efficiency benchmarks: Kyle Poyar (Growth Unhinged) and High Alpha, "2025 SaaS Benchmarks Report" (November 2025, 800-plus companies); ICONIQ Growth, "State of Software 2025." Builder.ai collapse: Financial Times and Bloomberg reporting, May 2025. Enterprise adoption backdrop: BCG, "Where's the Value in AI?" (October 2024); McKinsey, "The State of AI in 2025" (November 2025); MIT NANDA, "The GenAI Divide: State of AI in Business 2025" (July 2025, preliminary and non-peer-reviewed).
About the author
Most people writing about AI commercialization have never had to make an AI product pay. Jeff Brokaw has. He is a Certified Chief AI Officer and technical-commercial operator who builds the revenue, margin, and retention investors actually look for, and ships AI in production, not slideware. He runs a multi-model AI system in production, took an AI fintech from zero to $1M ARR in nine months, drove the go-to-market behind a $114M institutional raise that came together in under 30 days, and ran the commercial side of companies tied to more than $850M in associated exits.