Nvidia Pays Poolside $6 Billion to License Its 'Model Factory' — the Third Mega-Deal in Its New Playbook

Nvidia has agreed to pay $6 billion to license the AI model-development software of startup Poolside, while separately investing $1 billion in what remains of the company at a $12 billion pre-money valuation. The terms come from a letter Poolside sent to its investors, first obtained and reported by Newcomer on August 20, with The Information, TNW, and PYMNTS confirming the details the following day.

The software Nvidia is licensing is called the Model Factory — the internal system Poolside used to train and build its models, including its open-source model, Laguna, which the company released running on Nvidia server silicon. Nvidia will also extend job offers to 109 Poolside employees who worked on Laguna. The three co-founders, including CEO Eiso Kant, are staying.

Not an acquisition, not an acquihire

Poolside’s investor letter goes out of its way to define what the deal is not: “It is not an acquisition and it is not an acquihire.” The license is explicitly non-exclusive, meaning Poolside remains free to license the same Model Factory software to other customers. And the $6 billion licensing payment won’t sit on the startup’s balance sheet — it will be distributed to Poolside’s investors by the end of 2027, at $76.20 per share according to Newcomer’s reporting.

For a company that was valued at roughly $2 billion in October 2024, the $12 billion pre-money valuation marks a 6x jump in under two years — a remarkable outcome for a startup that, by its own admission, just stepped off the frontier-model treadmill.

The third deal from the same template

What makes this story bigger than a single licensing check is the pattern. This is now the third time Nvidia has structured a deal in exactly this shape: buy a non-exclusive license to the technology, hire the engineers, take an equity stake, and let the company continue operating independently.

• Enfabrica (September 2025): roughly $900 million to license the interconnect startup’s technology and hire its CEO and key staff. Enfabrica’s tech can link more than 100,000 GPUs into a single effective computer.
• Groq (December 2025): $20 billion for a non-exclusive license to Groq’s inference technology, plus the hire of its founder and top engineers. Groq stayed independent, raised $650 million for what remained, and by this month had closed a further $350 million round at a $3.5 billion valuation — with Nvidia participating.
• Poolside (August 2026): $6 billion plus 109 hires plus a $1 billion investment.

The strategic logic is twofold. First, these deals let Nvidia absorb critical talent and know-how — in chip interconnects, inference acceleration, and now model training infrastructure — without the friction of a full acquisition. Second, and perhaps more importantly, they sidestep the antitrust review that a formal purchase of a competitor-adjacent startup would trigger. In a period when regulators are scrutinizing every move by the dominant AI compute provider, a licensing deal is structurally harder to challenge than a merger.

Why Poolside stepped off the frontier race

The most revealing part of the investor letter is the explanation of why Poolside stopped building frontier models at all. “For the last 3 1/2 years we’ve been directionally correct in a race where capital requirements went vertical,” the letter reads.

Then the specific, brutal detail: at the end of last year, Poolside had a six-week window to raise $2 billion to pay for a 40,000-GPU GB300 cluster coming online in January. It didn’t close the round in time — and it lost the cluster.

The letter argues Poolside could have built a frontier-rivalling model with 10,000 to 20,000 of those chips, but that next year’s frontier will require a cluster “far more than an order of magnitude larger.” The binding constraint, it says, “is not only capital, it is physical data center space and contracted compute.” In other words: even a well-capitalized startup with the right technical team can no longer compete at the frontier, because the compute simply doesn’t exist for them to buy.

That admission is becoming a genre of its own. Poolside is one of the first prominent startups to say it plainly in writing to its own investors.

A remarkably small team

Another striking datapoint: CEO Eiso Kant said on the Latent Space podcast last month that fewer than 70 people built the model, with fewer than 115 total across engineering and research. The fact that Nvidia is hiring 109 employees from a company of that size means it is effectively absorbing the core technical team — which is why observers have dubbed the structure a “reverse-execuhire”: the founders stay behind, and the builders leave.

What’s in it for Nvidia

Nvidia has become a major developer of open models in its own right, and the Model Factory gives it battle-tested training infrastructure from a team that operated with unusual capital efficiency. Nvidia was already an investor in Poolside, having committed up to $1 billion last October as part of a planned $2 billion raise — a commitment that has now materialized as the equity piece of this deal.

Per PYMNTS’s reporting on the letter, Poolside also told investors that on its own it would have needed more access to Nvidia hardware than was realistically possible to keep competing in open-source model development. The subtext is hard to miss: when the dominant chip supplier is also your landlord, your potential competitor, and your licensing partner, the independent path narrows.

For Nvidia, the deal fits a broader strategy of building an integrated chip-plus-software ecosystem — owning not just the GPUs, but the training stacks, inference engines, and interconnects that make clusters act like single computers.

Analysis: liquidity without exits

The deeper significance of the Poolside deal is what it does to the startup funding ecosystem. Traditional venture returns come from acquisitions or IPOs. Nvidia’s licensing template creates a third path: early investors get cash liquidity ($76.20 per share by end-2027), employees get prestigious jobs at Nvidia, the startup survives with a fresh $1 billion and a $12 billion headline valuation, and no regulator gets a merger filing to review.

Expect more of these. If the Groq and Poolside templates become standard, the market for “partial exits” could reshape how AI startups negotiate with the compute oligopoly — and how much leverage the chip supplier accumulates over the entire model-building stack, all without ever writing the word “acquisition.”

Neither Nvidia nor Poolside immediately responded to press requests for comment on the deal.