Field Note · June 2026
How to price an AI product before Series A.
Price for margin before you raise, because margin is exactly what the people writing Series A checks are now reading you for.
Concretely, that is three moves, all of them before the round: price on a hybrid model that ties your revenue to your compute cost, instrument gross margin at the account level so you can prove cost-to-serve, and treat both as a Series A requirement rather than a post-raise cleanup. Do that and the round is winnable on a few million in ARR. Skip it and no amount of revenue reads as fundable, because AI diligence in 2026 underwrites margin, retention, and unit economics before it looks at your growth rate. The reason this blindsides founders is a story problem. The headlines say AI money is free, and for almost everyone building an AI product, that is exactly backwards.
The AI funding market is a barbell, and the averages lie
Start with the number that explains every misleading headline. In Q1 2026, AI startups raised $255.5B globally, more in a single quarter than the entire full-year 2025 total, according to PitchBook. Sounds like a flood of easy money. It is not. Three deals, OpenAI, Anthropic, and xAI, were 67.3% of all of it. Strip out the five largest and the rest of the market raised about $72.2B across roughly 4,595 deals, which PitchBook called broadly consistent with recent quarters. Divide it out yourself and that is roughly $15.7M a deal, my arithmetic on PitchBook's totals, not a figure they published. That smaller, stable market is the one you actually live in.
So when someone quotes you an AI funding average, they are usually quoting a number a few megadeals dragged into the sky. PitchBook's own research director described the market as bifurcated: you are in AI or you are not, you are a giant or you are not. There is no average company in a barbell. There is the tiny heavy end that makes the news, and there is the broad middle that has to earn it. Price for the middle, because that is where you are.
Did the Series A bar drop for AI startups? No. It doubled.
This is the part that cuts directly against the raise-on-vibes narrative. According to Peter Walker, Carta's Head of Insights, the Series A bar is now roughly double what it was a couple of years ago. The median Series A raise moved to $13M to $15M, up from $8M to $10M. Median post-money sits around $75M to $85M. And the number that matters most for this article: SVB puts median Series A revenue at about $2.5M in total revenue, not ARR, sharply higher than a few years ago, and rising. This is a median, not a cutoff. Plenty of AI companies raise with little revenue and a credible path to it. The point is not a magic number. It is that investors now read revenue, margin, and retention together, so weak economics do not clear no matter the top-line. The revenue requirement went up, not down.
The pre-revenue rounds you keep hearing about are a narrow, distorting cohort. Walker's own words on the AI-infrastructure and foundational-model teams raising seed rounds at $160M to $200M post-money: it is "bonkers," and it "mostly affects everyone else's mood and vibe, not their actual market." The data's own author is telling you those rounds are noise in your decision, not a benchmark. If you anchor your plan to them, you will under-build the one thing your real investors are going to test.
And the team writing that revenue is smaller than it used to be. The median Series A company on Carta now runs 15 to 17 people, down from around 45 a few years ago. The median seed company has four. AI compressed headcount, which is precisely why efficiency and margin became the story. There is no VP of Pricing in a fifteen-person company. There is you.
Two-tier valuations: why your comps are probably wrong
If you benchmark your valuation against "AI Series A," you will pick the wrong number, because there is no single AI Series A anymore. Carta put it bluntly in its Q1 2026 report: an AI foundational-model startup at Series A might raise at a $300M median valuation while a non-AI startup at the same stage is at $55M, and "these are not comparable markets." The AI valuation premium at Series A was 38% in 2025 on the median, and it widens to 193% by Series E and beyond.
Here is what that means for you in practice. Slapping "AI" on the deck does not move you into the $300M tier. That tier is foundational models and infrastructure, the heavy end of the barbell. If you are building an AI application on top of someone else's model, your comp is closer to the broad market that has to show ARR and margin, and the premium you do get is one you earn with economics, not with the acronym. Knowing which market you are in is the difference between a clean raise and six months of confused investor conversations.
What "price for margin before Series A" actually means
Pricing for margin is not a number on a page. It is a structure, and I have written about why AI pricing has to respect compute reality in detail: every inference call costs you real, variable money, and your best customers run up the bill the hardest. Seat-based pricing decouples your revenue from that cost, so a power user on a flat plan can look like your healthiest account while quietly going unprofitable. Before a Series A, that is not a rounding error. It is the exact thing diligence is built to find.
The fix most durable AI companies land on is a hybrid: a platform fee for access and the human layer wrapped around the model, plus usage-aligned components for the expensive consumption, structured so that growth in usage shows up as growth in revenue instead of growth in losses. You do not need to hand customers a utility bill. You need cost to appear somewhere in the model, so that when an investor pulls your gross margin and your cost-per-account, the lines bend the right way. That is what makes ARR fundable instead of just impressive.
When I took an AI fintech from zero to $1M ARR in nine months, the number that actually ran the business was cost-to-serve per account, and the part that surprised me was where it concentrated. The accounts the team was proudest of, the power users we put in the deck, were some of the thinnest-margin lines on the board. Love and cost were the same signal. You only catch that if you are metering per account before someone in a diligence room makes you, and by then the pricing decision that would have fixed it is two quarters behind you. The electrical-distribution AI SaaS case shows what per-account metering looks like when it is built before the raise, not cleaned up after.
This is the moment to be honest about the role you are playing. At pre-seed to Series A, with fifteen people and no one else to own it, you are your own Chief AI Officer. The path from AI capability to revenue, the unit economics, the pricing architecture, the build-buy-kill calls: that is the job, and it is yours by default. The founders who clear the new bar are the ones who picked it up early instead of waiting for a hire who is two rounds away.
Fast revenue is not the same as fundable revenue
The market is also getting better at telling the two apart, which is the whole reason pricing discipline pays off now. A demo proves the model works. A business proves someone pays, keeps paying, and the margin survives. Investors learned that the hard way. PitchBook flagged rising AI down rounds in late 2025 as a return to fundamentals, and Builder.ai, a unicorn that raised $445M, slid into insolvency after restating its revenue. Carta named retention fragility in viral AI products as a primary force dragging median valuations down relative to the top names.
None of that means the market is weak. The Q1 2026 down-round rate actually fell to 11.4%, back to 2019 and 2020 levels. The point is narrower and more useful: a viral usage spike with no durable economics underneath it is precisely what is now getting punished, while well-priced, retained, margin-positive revenue is what is getting funded at a premium. If your revenue cannot survive a diligence partner asking "and what does this cost you to deliver," it is the first kind, and pricing is how you make it the second.
What to do before your Series A
If you do nothing else this quarter, do these, in this order:
Pick your tier honestly. Foundational model and infrastructure, or AI application. It decides your comps, your valuation expectations, and how much the word "AI" is actually worth in your raise. Most readers are in the second tier. Plan for that bar, not the headline one.
Connect price to cost. If a customer can 10x their usage with no change to what they pay, you have a leak. Move to a hybrid model where consumption shows up in revenue. Do it before the round, so the trend line in your data room already points up.
Instrument margin per account. Know your gross margin and your cost-to-serve at the account level, not just in aggregate. The first question a sharp Series A investor asks about an AI company is what it costs to deliver. Have the answer on a slide.
Prove retention, not just growth. One clean cohort that stays and expands beats a viral chart that spikes and churns. Durable beats fast, and the data now says investors agree.
Aim the revenue honestly. The median is about $2.5M in total revenue to raise a Series A, per SVB, not ARR. You do not need to hit the median to start, but you need a credible path to it with margins intact, and pricing is the lever that gets you there without burning the round to buy revenue.
Common questions on pricing before a Series A
Do you need revenue to raise a Series A for an AI startup? For almost every founder, yes. The no-revenue rounds belong to a narrow cohort of frontier-lab and AI-infrastructure teams raising at $160M to $200M post-money at seed. For a normal AI application company, SVB's data puts median Series A revenue around $2.5M in total revenue, not ARR, sharply higher than a few years ago, and rising. This is a median, not a cutoff. Plenty of AI companies raise with little revenue and a credible path to it. The point is not a magic number. It is that investors now read revenue, margin, and retention together, so weak economics do not clear no matter the top-line. The bar went up, not down.
How much revenue do you need for an AI Series A in 2026? About $2.5M in total revenue at the median, per SVB, against a median raise of $13M to $15M and a post-money near $75M to $85M, with a team of just 15 to 17 people. You do not have to hit the median to start conversations, but you need a credible path to it with margins intact.
Should AI products use usage-based or seat-based pricing? Neither on its own. Seats hide your compute cost, so your heaviest users quietly go unprofitable while looking like your best accounts. Pure usage pricing scares off the experimentation that drives adoption. What holds up is a blend: charge a base fee for access and the layer you wrap around the model, then attach a usage-aligned component to the consumption that actually costs you, so heavier use lands as more revenue rather than more loss.
What valuation can an AI application company expect at Series A? Far below the headline AI numbers, because those belong to a different tier. Carta's 2025 data puts the median AI Series A valuation about 38% above the non-AI median, a real premium but a modest one. The $300M foundational-model comps you see in the press are the wrong benchmark for an application company built on someone else's model. Your valuation tracks the broad market that has to show ARR and margin, and the AI premium you earn comes from those economics, not from the label.
Your pricing is the one place where the technology of your product meets the economics of your business, and for an AI company those two are tied together by compute in a way they never were for traditional software. That is also why it is the first thing a Series A investor reads you on, even when they are asking about something else. Get it right before the raise, and the raise gets easier. Get it wrong, and you find out at the worst possible time, in a room you cannot leave, with the term sheet on the table.
If you are staring at this gap and the next round is closer than the answer, that is the conversation I have.