Nvidia will pay AI coding startup Poolside $6 billion to license its Model Factory — the in-house training system Poolside uses to build the Laguna family of open-weight code models. Bloomberg reported the deal on August 20, citing an investor letter obtained by Newcomer. The license is non-exclusive.
Two other things are happening alongside it. Nvidia is separately investing $1 billion in Poolside at a $12 billion pre-money valuation, and 109 employees who worked on Laguna will receive job offers from Nvidia. Poolside's three co-founders are staying on, and the company keeps operating independently. The $6 billion license fee is expected to be distributed to Poolside's investors by the end of 2027.
A deal shaped to avoid a label
The investor letter draws a deliberate line: this is neither an acquisition nor an acqui-hire. That framing has been used repeatedly over the past two years, designed to keep large transfers of talent and technology out of merger-review territory.
Nvidia has two recent precedents of its own. Its deal absorbing Groq's core team and technology was worth $20 billion; the Enfabrica deal was $900 million. Both followed a similar structure. Taken together, the three deals show Nvidia has turned "license plus investment plus job offers" into a repeatable playbook.
Do the rough math: $6 billion spread across 109 people works out to about $55 million each. That math isn't rigorous — the license fee is paying for a software system, and the people are just part of what comes with it — but it's a useful marker of where Nvidia's price floor sits for this kind of asset.
How Poolside got here
Poolside was founded in 2023 by former GitHub CTO Jason Warner and Eiso Kant. It closed a $500 million Series B in October 2024 at a $3 billion valuation, with Nvidia already on the cap table as an investor.
The real turning point came this April: a planned $2 billion Series C, which would have valued the company at $14 billion, fell through after CoreWeave pulled out of a jointly planned data center project in Texas, cutting off the fundraising momentum. Four months later, Poolside's pre-money valuation for this deal is $12 billion — about 20% below the valuation of the failed round — but the company has converted its training system into $6 billion in cash, money that, under the terms described in the letter, flows directly to existing shareholders.
Laguna itself launched on July 21. Laguna S 2.1 is a mixture-of-experts code model with 118B total parameters and 8B active parameters, released with open weights and built for agentic coding, scoring 70.2% on Terminal-Bench 2.1. Its pitch has always centered on training efficiency: producing a self-hostable, Western open-weight model in under nine weeks. That speed is exactly what Nvidia is buying.
What Nvidia wants it for
Nvidia already has its Nemotron model line and keeps releasing open weights and training data of its own. Building models directly puts it in competition with some of its own customers, which makes an outright acquisition of a model company politically sensitive. Licensing a training system, hiring away the team that runs it, and leaving the licensor operating independently in the market is a way around that awkwardness.
For the open-source code model space, it's now unclear who will build future versions of Laguna, or whether they'll keep the weights open, now that Poolside's core team has moved to Nvidia. Poolside keeps the company, the founders, and the right to use the licensed software — but it just sent away the people who know it best.
Sources: Newcomer investor letter, Bloomberg, CocoLoop, The Decoder; the $6 billion license fee, $1 billion investment, $12 billion pre-money valuation, and 109-employee figure are cross-verified across multiple reports, while the Laguna S 2.1 parameters and benchmark score come from Poolside's official release.