A Trillion Tokens, Priced at Zero

TypeSafe AI announced on 9 October that it had raised $870 million in a Series A at a $7.5 billion valuation. The company’s post is cheerful and short. Its numeric content is three items: the money, the valuation, and the claim that a third of the Fortune 500 are using Jev, its decision model. It also says it has “saved customers millions of dollars in production already,” with no customer, task, or baseline attached.

I have written about Jev twice, both times about measurement: how the “193.6 times faster” headline depends on which rival sits in the denominator, and how two other decision models used the same benchmark. This piece is about a different layer. A valuation is a bet on future revenue, so the question is what the announcement gives you to hold it against. I looked in four places: the company’s post, the launch post, the lead investor’s announcement, and the wire coverage. None of them reports revenue. That is not unusual for a company that came out of stealth 24 days earlier, and I am not claiming the valuation is wrong. I am claiming that what is offered in its place can be read, and some of it does not survive being read closely.

The seed and the jump

On 15 September, TypeSafe came out of stealth with a $40 million seed round led by DCVC. On 9 October it announced $870 million, led by Andreessen Horowitz with Sequoia and DCVC participating. That is about 21.8 times the seed in 24 days (my arithmetic). Neither the company nor a16z says whether $7.5 billion is a pre-money or post-money figure. If it is post-money, the new money is about 11.6 percent of the company (also my arithmetic). Neither fact is suspicious by itself. The open question is what the round is said to rest on.

One company, two Fortune 500 shares

Two sources published on the same day give two shares. TypeSafe’s post says a third of the Fortune 500 are using Jev. a16z’s post says “25% of the fortune 500 enterprises have integrated Jev.” Bloomberg, as relayed by Investing.com, repeats the third and adds that TypeSafe declined to name any of the companies.

The Fortune 500 has 500 members, so a quarter is 125 companies and a third is about 167. The gap is roughly 42 companies, between a company and the investor who just led its round. The verbs differ too. “Using” can mean a developer tried the API on a free tier; “integrated” suggests something in a deployment. Neither source defines a unit: a signed contract, an active key, one employee’s account? I found no list, no method, and no named customer in the company post or the investor’s note.

Two Fortune 500 adoption figures published on 9 October 2026 A strip from 0 to 500 companies. a16z's note says 25 percent of the Fortune 500, which is 125 companies, shown as a circle. TypeSafe's announcement says about a third, roughly 167 companies, shown as a diamond. A line joins them and marks a gap of about 42 companies. FORTUNE 500 COMPANIES SAID TO USE JEV · TWO 9 OCTOBER 2026 CLAIMS 0 100 200 300 400 500 125 companies (25 percent) — a16z, 9 October 2026 about 167 companies (a third) — TypeSafe AI, 9 October 2026 a16z: 25% = 125 TypeSafe: a third = about 167 gap: about 42 companies Counts are the writer's arithmetic on a 500-company list. Neither source names a customer or defines "use" or "integrated".
Two statements about the same set of 500 companies, made the same day, about 42 companies apart. Neither comes with a method.

A trillion tokens and a price list

The investor’s note offers one volume number: Jev reached “1 trillion tokens generated in merely 3 days post launch.” It calls Jev “the fastest growing model we’ve ever seen.” The note also says the output “is so small that they don’t even bother charging for it.”

Read those together with the launch post’s stated prices. Input tokens are $0.042 per million, “$42 per billion tokens,” and output tokens are free. A trillion is a thousand billion. If “generated” means what it says, those are output tokens and the list price on them is zero. If the investor meant tokens processed, billed at list input price and with no volume discount, the ceiling is 1,000 × $42 = $42,000. Both readings are my arithmetic, and enterprise contracts can differ from a public price list.

At that list price, $1 million of revenue is about 24 trillion input tokens. The first three days, at most, are about 4 percent of a million dollars. It is a fine adoption signal and a weak revenue one. The launch post itself says the company “can’t prove it isn’t subsidized,” which is the honest caveat on any figure built from that price.

One trillion tokens at list price against the Series A and valuation, logarithmic scale A logarithmic dollar axis from 1 thousand to 10 billion. The ceiling for one trillion tokens billed at the list input price of 0.042 dollars per million tokens, which is 42,000 dollars, is a ring near 42 thousand. The 870 million dollar Series A is a circle. The 7.5 billion dollar valuation is a diamond. DOLLARS · LOG SCALE · LIST PRICE $0.042 PER MILLION INPUT TOKENS $1K $10K $100K $1M $10M $100M $1B $10B 1T tokens at list input Series A raised Valuation $42,000 — ceiling if one trillion tokens were billed at list input price (writer's arithmetic) $870 million — TypeSafe AI Series A, 9 October 2026 $7.5 billion — reported valuation, 9 October 2026 $42,000 at most $870 million $7.5 billion Ring is the ceiling if every one of the trillion tokens were input tokens at list price. If they are output tokens, the list price is $0.
The usage number is real as far as anyone can tell, but at the public price list it sits about 20,000 times below the round (870 million divided by 42,000 is about 20,700).

What would settle it

TypeSafe’s launch post says it named Jev after the economist William Stanley Jevons. His 1865 book The Coal Question observed that more efficient steam engines raised coal consumption instead of lowering it. a16z leans on the pun (“the Jev-ons paradox”). The paradox is about volume. It is silent on price: if the metered unit is a tenth of a cent and the output is free, a thousand-fold rise in calls can still be a small business. Whether this one is small is unknown. The company has not said, and neither has anyone else I could find.

Here is what I would want before treating the valuation as anything other than a conviction. The unit behind the Fortune 500 share, and the name of one company in it. The split of the trillion tokens into input and output. Paid usage versus free-tier usage. Anything on revenue, even a range. The company declined to name customers, which is its right; the evidence then rests on the investor’s word and the company’s own. Both are parties to the deal, and the investor note carries a standard disclaimer that it is not investment advice and should not be relied on as such.

The independent evidence on Jev’s performance is the thinner half of this story, and I covered it in the earlier pieces. This piece’s narrower point is that a $7.5 billion price was agreed on a model with a free output, a public input price of $42 per billion tokens, and an adoption claim that its own backers do not state the same way.

References

  1. TypeSafe AI (2026). TypeSafe A Raises Series AI. Company blog, announced 9 October 2026.
  2. Li, J., Wang, S., Casado, M., Andreessen, M., and Horowitz, B. (2026). Investing in TypeSafe AI. Andreessen Horowitz, 9 October 2026.
  3. Almeida, D. (2026). Introducing System One Models & Jev. TypeSafe AI blog. Stated prices and subsidy caveat; the model launched 15 September 2026, though the page itself now shows a later timestamp.
  4. DCVC (2026). TypeSafe emerges from stealth with a new way of doing AI. September 2026. $40 million Series Seed.
  5. Bloomberg, relayed by Investing.com (2026). TypeSafe AI raises $870 million led by Andreessen Horowitz. 9 October 2026.
  6. Jevons, W. S. (1865). The Coal Question. Macmillan, London.