Nvidia has agreed to pay Poolside $6 billion for a non-exclusive license to the software the startup used to build its AI models, hire 109 of its employees, and invest a further $1 billion in whatever is left — a structure that delivers Nvidia nearly everything an acquisition would, without the acquisition. The terms, drawn from a letter Poolside sent its investors and first reported Thursday by Eric Newcomer and Tom Dotan at Newcomer, value the continuing company at $12 billion pre-money. Poolside's three co-founders are staying. The letter is emphatic on the point: the deal is "not an acquisition and it is not an acquihire."
What Nvidia is buying is called the Model Factory — the internal platform Poolside built to produce its Laguna family of open-weight coding models at unusual speed. The 109 staff receiving Nvidia job offers are the people who worked on Laguna. Poolside intends to distribute the $6 billion to its investors by the end of 2027. Because the license is non-exclusive, Poolside remains free to sell the same technology to anyone else.
The price is the part nobody has explained. Nvidia has not commented publicly on the deal, and Poolside has not published the letter. The Information's Amir Efrati, reporting the same terms, wrote that it is not clear why Nvidia is paying a licensing fee of this size. Some outlets have described the combined commitment as a $7 billion package, folding the equity investment into the headline; the licensing fee itself is $6 billion.
A price per head that reads like an acquisition
The arithmetic behind the "reverse acquihire" label is blunt. Poolside's co-founder and co-CEO Eiso Kant described the scale of the team on the Latent Space podcast last month, before the deal was public. "Less than 70 people built this model," he said. "Less than 115 between engineering and researchers, like, together did this effort." Nvidia is hiring 109 of them.
That team's output was genuinely competitive for its size. Laguna S 2.1, released July 21, is a 118-billion-parameter mixture-of-experts model that activates only 8 billion parameters per token, supports a 1-million-token context window, and scored 70.2% on Terminal-Bench 2.1 — ahead of DeepSeek-V4-Pro-Max at 64.0 and Nvidia's own 550-billion-parameter Nemotron 3 Ultra at 56.4. It went from the start of pre-training to public release in under nine weeks, trained on 4,096 Nvidia H200 GPUs. "The West needs open-weight models it can trust, run, and build on," Poolside co-CEO Jason Warner said in the launch announcement, framing Laguna explicitly as a Western answer to the Chinese open-weight labs that have dominated the category.
The investor letter also explains, unusually candidly, why Poolside stopped chasing the frontier. "At the end of last year, we had a 6 week window in which to raise $2 billion dollars to pay for a 40,000 GB300 cluster coming online in January," it says. "We didn't close it in time, and we lost the cluster." Next year's frontier, the letter argues, will require a cluster more than an order of magnitude larger still, and the binding constraint "is not only capital, it is physical data center space and contracted compute." Poolside's remaining bet, per the letter, is that human-level capability will be "fully commoditized by open source models" while the real value sits in experiment-bound problems: "AI will become the world's most valuable scientific discovery engine."
Why It Matters
This is the third time Nvidia has run this exact play. In December 2025 it paid roughly $20 billion for a non-exclusive license to Groq's inference technology and hired its founder, president and much of its engineering bench; Groq continued as an independent company and closed a $350 million round at a $3.5 billion valuation this month, with Nvidia participating. In September 2025, CNBC reported Nvidia spent more than $900 million to license Enfabrica's technology and hire its CEO and staff. Across the three transactions, Nvidia has now committed roughly $27 billion without filing a single merger notification.
The structure is not a loophole Nvidia invented — Meta, Google, Microsoft and Amazon have all used variants of it since 2024 — but Nvidia is now its most expensive practitioner. Regulators have noticed. On March 19, Senators Elizabeth Warren and Richard Blumenthal wrote to Jensen Huang arguing the Groq deal appeared structured to evade antitrust premerger review, warning it could further entrench a company that already controls the overwhelming majority of the AI accelerator market. In February, Warren, Blumenthal and Senator Ron Wyden had asked the DOJ and FTC to investigate reverse acqui-hires at Meta, Google and Nvidia more broadly. No enforcement action has followed, and non-exclusive licensing does preserve a real theoretical competitor.
There is a second, subtler tension. Nvidia sells GPUs to every AI lab on earth, and it is now buying the capability to build open models that compete with its own customers' products. Its Nemotron line is already pushing toward a trillion-parameter open model. Absorbing the team behind Laguna — a model that beat Nemotron 3 Ultra on Terminal-Bench while being four times smaller — makes Nvidia a more serious model builder and a more awkward supplier. It struck this deal the same month it halved a $250 billion commitment to OpenAI.
What to Watch
Three things. First, whether the FTC or DOJ treats the Poolside deal as the case that finally tests reverse acquihires — a third instance from the same buyer at a combined $27 billion is harder to wave through than a first. Second, what Poolside actually becomes: the founders say they are "not ready to share the updated vision," while Poolside Infrastructure Company, spun out in January, is building a 1.2GW data center in Texas and hired a CEO and CFO in the past two months. Third, whether the $6 billion license produces anything visible in Nemotron. Nvidia paid an acquisition price for a process, not a product. If the Model Factory does not transfer — if the speed lived in the 109 people rather than the software — the most interesting number in this deal will turn out to be the headcount, not the fee.
“Less than 70 people built this model.”— Eiso Kant, Co-founder and co-CEO, Poolside