Two deals, structured to avoid one word
Nvidia has agreed to pay $6 billion to license Model Factory, the internal platform artificial intelligence (AI) startup Poolside built to train its own models, and will separately invest $1 billion directly into what remains of the company. Together the two moves value the rest of Poolside at roughly $12 billion. The reporting, first broken by Newcomer on August 20, 2026 and since corroborated by other outlets, centers on one specific detail: Poolside told its investors in a letter that the transaction is neither an acquisition nor an acquihire, even as 109 of its employees are being offered jobs at Nvidia.
Those 109 people are specifically the team that worked on Laguna, the family of coding and open-source models Poolside built using Model Factory. Poolside's three co-founders are staying at the company and keep control of it; they are not among those moving to Nvidia.
Poolside had positioned itself as a builder of large language models aimed specifically at software development, competing for attention and compute against a crowded field of coding-model startups. Model Factory was the training pipeline behind that work, the part of the company Nvidia is now paying to use rather than own.
What Nvidia actually gets
The license is non-exclusive, meaning Poolside is still free to license Model Factory to other cloud providers, infrastructure companies, or enterprise customers. Nvidia has framed its use of the platform as a complement to Nemotron, its own line of open-weight models, rather than a replacement for anything Poolside continues to build.
The $6 billion license fee is set to be distributed to Poolside's existing investors by the end of 2027, effectively functioning as a payout for the people who backed the company before this deal, without those investors having sold the company itself.
Reporting from PYMNTS frames the underlying business logic simply: Poolside needed more compute and hardware access than it could secure on its own to keep pace in open-source model development, and this structure gets it there without a formal sale.
A repeat of Nvidia's playbook
This is not the first time Nvidia has used this shape of deal. According to PYMNTS, the company previously arranged a $20 billion non-exclusive intellectual property license with Groq paired with a hiring push, and a separate $900 million deal with Enfabrica that followed a similar pattern. On the Groq deal, a Nvidia spokesperson put the distinction plainly: "We haven't acquired Groq. We've taken a non-exclusive license to Groq's IP and have hired engineering talent."
The repetition matters because straightforward acquisitions of AI startups by dominant chipmakers and cloud providers draw antitrust attention, particularly from regulators in the United States, the European Union, and the United Kingdom who have grown more willing to scrutinize deals that consolidate AI talent and compute access even without a formal change of ownership. A non-exclusive license paired with individual job offers is not structured the same way as an acquisition on paper, even when the practical effect, a chip giant absorbing the team and technology behind a model line, looks similar.
Why this keeps happening
Two forces are pushing deals into this shape. Compute-constrained AI startups need capital and hardware access at a scale only a handful of companies can supply, and Nvidia in particular wants the model-building expertise and engineering talent that startups like Poolside and Groq have accumulated without the delay or scrutiny that comes with a formal takeover filing. Licensing the technology and hiring the team piecemeal gets Nvidia most of what an acquisition would deliver, while leaving the startup's brand, remaining staff, and cap table technically intact.
What it means going forward
For AI startups running short on compute, this deal is a signal of a viable exit path that does not require a full sale: license the platform, let key engineers move to whoever is buying the license, and keep a smaller, independent company running with a fresh valuation. For anyone tracking the concentration of AI infrastructure and talent, the more useful measure is no longer just who acquires whom, but who ends up on Nvidia's payroll and whose technology Nvidia's own products increasingly run on, regardless of what a deal is called on paper.
