Photo by AMONWAT DUMKRUT on Unsplash
The Common Belief
$1,785,714.
That is the post-money valuation hiding inside Y Combinator's first check. Divide $125,000 by 7% and you get a company priced, on paper, at under $1.8 million — before a single Demo Day slide is clicked through. As of October 15, 2026, YC's standard deal is reported as $500,000 per company: $125,000 for 7% on a post-money SAFE, plus $375,000 on an uncapped SAFE with MFN (most-favored-nation) terms, meaning the second tranche converts at the best price of the next priced round. Those two sentences explain more about Demo Day economics than any watchlist does.
According to AI Fallback, the original reporting underpinning this topic, the figures above — along with batch sizes of roughly 250–300 companies, two main batches per year, and an AI share reported around 70–80%+ across 2024–2025 cohorts — are flagged as unverified, because live web verification was unavailable at the time of compilation. That caveat is not a footnote to skip past. It is the story. The conventional belief is that Demo Day hands investors and founders a clean shortlist of winners. What it actually hands them is an unverifiable sampling problem with a very precise price tag attached.
Where It Breaks Down
Start with attention, not capital. If a batch runs 300 companies and a diligent reader gives each one three minutes — enough to read a blurb, open a landing page, and form an opinion — that is 900 minutes, or 15 hours, per batch. Two batches a year puts the honest skim cost at roughly 30 hours annually, and that is before a single founder call. No published "top startups to watch" roundup is built on 30 hours of work.
Now layer in concentration. If 70–80% of a 300-company batch is AI-focused, that is somewhere between 210 and 240 AI startups per cohort. A ten-name watchlist drawn from that pool is selecting about 4% of the AI subset (10 ÷ 240 ≈ 4.2%). The list is not a filter. It is a lottery ticket with editorial confidence layered on top — the same headline-credibility gap visible in SpaceX's "earnings beat" coverage, where a confident number traveled further than its sourcing justified.
There is also a structural change the surface reporting tends to mention and then drop. Demo Day shifted to a hybrid and online format after 2020, with investor access routed through YC's own platform. Combine that with the move under president and CEO Garry Tan toward larger and more frequent batches, and the scarce resource flips. Capital is abundant. Signal is not. When every allocator sees the same company list through the same portal at the same hour, the list itself stops being an edge.
The Capital Math Nobody Nets Out
Here is a figure that does not appear in any single source article, because it requires multiplying two of them together. At $500,000 per company, a 250-company batch commits $125 million; a 300-company batch commits $150 million. Run two batches a year and the accelerator is deploying roughly $250–$300 million annually in pre-seed capital. Apply the reported 70–80% AI share to a 300-company batch and between $105 million and $120 million of a single cohort's capital lands on AI agents, developer tooling, and vertical AI applications.
Chart: Capital committed per batch, derived from the reported $500,000 standard deal and 250–300 company batch sizes. AI slices apply the reported 70–80% AI share to a 300-company cohort. Underlying figures are flagged unverified.
A skeptic will push back, and fairly: pre-seed dollars are rounding errors next to late-stage AI rounds, so why does $105 million matter? Because it is not priced like late-stage money. At a sub-$1.8 million implied entry, roughly 240 AI companies per cohort are being bought at a valuation that assumes almost nothing works. That is an index position on the AI-native wedge, not a conviction bet — and index positions only need a handful of Airbnb, Stripe, DoorDash, Coinbase, or Reddit outcomes to carry the whole investment portfolio.
Photo by Stefan Vibes on Unsplash
Who Wins Under Which Condition
The $375,000 uncapped MFN tranche is where founder outcomes genuinely diverge, and it is almost never explained in roundup coverage. Because it converts at the next priced round's terms, the dilution a founder absorbs depends entirely on how strong that round is.
Consider two illustrative scenarios — hypothetical, not reported. A founder who raises a seed at a $15 million post-money price gives up roughly 2.5% for that $375,000 ($375,000 ÷ $15,000,000), landing near 9.5% total accelerator ownership. A founder who raises at $30 million gives up about 1.25%, landing closer to 8.25%. Same program, same check, same 7% first tranche — and a meaningful difference in cap-table outcome determined by momentum the founder creates after Demo Day. The structure rewards the strongest fundraisers twice: better price, less dilution.
So who wins? Investors with proprietary access win when the batch is small and the list is scarce. Investors without it lose as batches grow, because a 300-company online Demo Day converts a relationship advantage into a screening chore. Founders with an obvious ICP-fit and a defensible wedge product win in large batches, because scale makes mediocre pitches invisible and sharp ones stand out faster. Founders still searching for a wedge are the ones large cohorts quietly punish.
A Better Frame: The Founder Move This Quarter
Model your own cap table at two or three plausible next-round prices and see what the uncapped MFN portion actually costs you. Treating accelerator capital as free is the most common financial planning error at pre-seed. Run the arithmetic yourself; it takes ten minutes.
If the AI share of a cohort really is 70–80%, "we use LLMs" is table stakes. The differentiator is a named ICP and a measurable before/after metric — latency cut, hours saved, error rate reduced. ARR trajectory beats architecture in a crowded batch.
Every figure here traces to research explicitly labeled unverified. Anyone allocating real money, or benchmarking their own round against batch norms, should confirm current deal terms and cohort size against YC's primary disclosures rather than any secondary roundup. AI investing tools and AI research assistants are useful for triage, but they inherit whatever sourcing gap sits upstream.
Bottom Line
- YC's reported standard deal — $500,000 total, as $125,000 for 7% plus $375,000 on an uncapped SAFE with MFN — implies a day-one post-money valuation of about $1.79 million.
- At 250–300 companies per batch and two main batches per year, that structure commits roughly $125–$150 million per cohort and $250–$300 million annually.
- With a reported 70–80% AI share, $105–$120 million of a single 300-company cohort flows to AI agents, dev tooling, and vertical AI.
- Hybrid/online Demo Day plus larger, more frequent batches under Garry Tan means investor attention — not capital — is the binding constraint.
- All underlying figures carry an unverified flag; treat watchlists as a starting point for diligence, not a conclusion.
Our read: the interesting question about Demo Day is no longer which ten companies make the roundups. It is what happens to accelerator pricing when a cohort is 80% AI and every allocator is looking at the same portal on the same afternoon — a condition under which a sub-$1.8 million entry price starts to look less like a bargain and more like the cost of buying the whole index. On balance, the more likely outcome is continued separation between founders who can convert Demo Day attention into a high-priced next round and those who cannot, with the MFN tranche quietly amplifying the gap.
Frequently Asked Questions
How much equity does Y Combinator take in 2026?
As of October 15, 2026, the reported standard deal is $500,000 per company: $125,000 for 7% on a post-money SAFE, plus $375,000 on an uncapped SAFE with MFN terms that converts at the next priced round's price. The 7% is fixed; the additional dilution from the $375,000 depends on your next round's valuation. These terms come from research flagged as unverified — confirm against primary disclosures before relying on them for financial planning.
Are most YC startups AI companies now?
Reported figures put AI-focused companies at roughly 70–80%+ of 2024–2025 cohorts, concentrated in AI agents, developer tooling, and vertical AI applications. At a 300-company batch size, that implies 210–240 AI startups per cohort. The practical consequence for founders: AI is no longer a differentiator inside the batch, so the wedge has to be the customer and the measurable outcome.
Is a YC "startups to watch" list useful for investors?
As a triage tool, marginally. As a selection signal, weakly — a ten-name list pulled from roughly 240 AI companies represents about 4% of the pool, chosen by whoever had the fewest hours to spend. Since Demo Day moved to a hybrid/online format with access through YC's platform, the same names reach everyone simultaneously, which erodes any informational edge in the list itself. Nothing here is investment advice; it is an observation about how little a published shortlist narrows the field.
Disclaimer: This article is editorial commentary for informational purposes only and does not constitute financial, investment, or legal advice. It reflects analysis of publicly reported information, not independent product testing or private diligence. Key figures cited above were flagged as unverified in the source research and should be confirmed against primary disclosures. Research based on publicly available sources current as of October 15, 2026.