In August 2026, Poolside sent its investors a letter. Eric Newcomer got hold of it, and the terms in it are worth reading twice.
Nvidia is paying $6 billion to license the software Poolside used to build its AI models. The licence is non-exclusive. Nvidia is separately investing $1 billion at a $12 billion pre-money valuation. And 109 Poolside employees have received Nvidia job offers.
All three founders are staying. The letter says this is "not an acquisition and it is not an acquihire." Poolside intends to distribute the $6 billion to its investors by the end of next year.
Read that again, because the shape of it matters more than the size. A company licensed out its core technology, sent a large share of its people toward the buyer, and expects to return billions to its cap table — without selling itself. On paper, Poolside is still Poolside. Its founders still run it.
The reason to pay attention is not that this happened. It is that Nvidia has now done it three times, and the third time is not even the biggest.
The pattern, and the paperwork it avoids
Nvidia did this with Groq, in a deal reported at around $20 billion — more than three times the Poolside licence. It did it with Enfabrica, reported at over $900 million. Poolside is the third — bigger than Enfabrica, about a third the size of Groq.
The template is consistent across the two we can see in detail — Groq and Poolside. Enfabrica's terms are far less well reported, and the $900m figure is close to all we have on it:
- License the technology, non-exclusively. Not an asset purchase. A licence.
- Move the people who know how to run it. At Groq, founder and CEO Jonathan Ross, president Sunny Madra and senior engineering leadership went to Nvidia. At Poolside, 109 employees have offers.
- Invest in, or pay, what remains, so the cap table gets liquidity.
- Leave the company standing. Groq continues as an independent company — its CFO, Simon Edwards, stepped up to CEO, and GroqCloud kept operating. Poolside keeps all three founders.
Note what varies and what doesn't. Groq lost its founder and got a new CEO; Poolside's founders stayed. Founder retention moves. The licence, the talent transfer and the surviving entity do not — both companies are still standing.
Now the part that is not our inference. When CNBC covered the Groq deal, the anchor asked outright whether it was an acquisition. Reporter Mackenzie Sigalos answered that it is not — the parties, she said, "have been very specific in saying that." The reading across the coverage is consistent and blunt: a full acquisition of a credible competitor by the most valuable company on earth invites prolonged regulatory review, while a non-exclusive licence plus a hiring wave plus nominal independence secures the substance of an acquisition and preserves the optics of competition. Alumni Ventures called the licensing-versus-purchasing distinction "the legal distinction regulators are scrutinizing."
The word "non-exclusive" is doing an enormous amount of work. It is the difference between buying a company's future and renting its capability, and the two do not get reviewed the same way.
What this is worth to a founder who is not Poolside
Most writing about deals this size is spectator sport. This one is not, because the structure scales down and the logic is portable.
You have been taught there are two endings. You raise the next round, or someone buys you. This is a third: someone pays for the machine you built, hires the people who know how to run it, and leaves you holding the company. Your investors get liquidity. Your team gets a landing. You keep the entity and the name.
The precondition is specific, and it is not "have good technology." It is this: you built a capability a much larger company would rather rent than rebuild. Poolside's asset was never really the model. It was the model-building software — the factory, not the output. Nvidia is not short of compute. What it licensed was the thing that turns compute into models.
Eiso Kant, Poolside's CEO, described the effort on the Latent Space podcast: "Less than 70 people built this model… Less than 115 between engineering and researchers, like, together did this effort."
Fewer than 115 people built something a company worth several trillion dollars paid $6 billion to license rather than reproduce. That is the whole lesson, and it is not about AI. Concentrated, hard-won capability held by a small team is the asset. Not the users. Not the ARR. The thing you know how to do that they would have to spend three years learning.
The question the letter does not answer
Here is where we stop being impressed.
What is actually left of Poolside? Its model-building software is licensed to the company best positioned to use it, and 109 of its people are holding Nvidia offers. Nobody has said how many will accept, and an offer is not a departure — but the company has not said what the remaining organisation looks like on the other side of that decision either, and that is the number that matters.
The founders say plenty remains, and they may be right; they have $1 billion of fresh capital, a $12 billion valuation, and no obligation to explain themselves further. But the honest reading is that Nvidia got the capability it wanted, and the entity left standing is part of what makes the deal legible as something other than a purchase — the same role Groq now plays under a new CEO.
That is not a criticism of the founders. Their Series B was $500 million led by Bain Capital Ventures in October 2024, into a funding market that now looks far less willing to write cheques for foundation-model startups without a strategic backer behind them. Given those alternatives, returning $6 billion to your investors while keeping your company is an extraordinarily good outcome, and anyone telling you otherwise has not tried to raise a Series C for a model company lately.
But if you are reading this as a template, be clear about which half you are copying. The valuation is the headline. The structure is the lesson. And the structure works precisely because it lets a buyer take the capability without taking on the company — which means the version you get offered will be shaped by what the buyer needs to avoid, not by what you want to keep.
Ask what happens to the entity after the $6 billion has been distributed. The letter does not say. That is not an oversight.
