Nvidia already ran this exact move once this year. Now it's running it again.
The chip giant agreed to pay $6 billion to license AI model-building software from Poolside, a three-year-old coding AI startup, and will hire 109 of its engineers directly. Poolside's own investor letter is explicit that this isn't an acquisition.
What Nvidia Actually Bought
The license covers Poolside's Model Factory, the system behind its Laguna family of open-weight coding models. Nvidia is separately investing $1 billion in Poolside at a $12 billion pre-money valuation, roughly six times what the company was worth in an October 2024 round.
Poolside's three co-founders, including former GitHub CTO Jason Warner and Eiso Kant, are staying to run what's left of the company independently. The $6 billion licensing fee gets distributed to existing backers, including Bain Capital Ventures, eBay, Citi Ventures and Redpoint, by the end of 2027.
The team behind the actual technology was small. Kant said on the Latent Space podcast last month that fewer than 70 people built Poolside's model, with fewer than 115 total across engineering and research.
This Is the Third Time Nvidia Has Done This
Nvidia ran nearly the identical structure on Groq in December, a deal MAIN has covered already: a non-exclusive technology license, a large payout, key staff hired directly, and the original company left technically independent and still operating.
The scale was even larger that time. Nvidia paid roughly $20 billion for Groq's inference technology license and brought over founder Jonathan Ross along with much of the company's engineering team, after which Groq pivoted from building its own AI chips to running a Nvidia-powered neocloud instead.
The pattern lets Nvidia absorb critical AI talent and technology without triggering the antitrust scrutiny a formal acquisition would draw. Poolside's own letter reportedly said the company needed more Nvidia hardware access than it could otherwise secure to keep competing in open-weight model development on its own, a dependency Nvidia's licensing structure conveniently resolves at the same time it resolves Nvidia's interest in owning more of the AI software stack.
Why Nvidia Is Now Competing With Its Own Partners
Nvidia plans to use the Poolside deal to build one of the most capable open-weight AI models available, aimed squarely at Chinese labs including DeepSeek and Moonshot's Kimi K3, models MAIN has covered gaining real traction with cost-conscious enterprise buyers.
That's also a direct challenge to OpenAI and Anthropic, two of Nvidia's closest customers and partners. Open-weight models are generally cheaper to run and easier to customize than the closed frontier models both companies sell.
Poolside's founders framed the underlying ambition in their investor letter, writing that artificial general intelligence "would not be a closed technology controlled by few but one built by many out in the open."
Jensen Huang's logic seems to be that Nvidia doesn't need to pick a side in the model wars. It can sell the chips to everyone while building a horse of its own.
What This Means for Miami
This deepens a pattern MAIN has tracked all year: Nvidia isn't content supplying picks and shovels to the AI gold rush. It's increasingly financing, staffing and now directly competing in the businesses built on top of its own hardware, the same dynamic already visible in its Groq deal, its Marvell-Google chip financing, and its broader equity stakes across the AI infrastructure buildout.
For Miami's AI investors and enterprise buyers, that consolidation is worth tracking regardless of which specific model wins. A single company controlling more of the chips, the financing and now a leading open-weight model narrows the number of genuinely independent choices available, even as the total number of AI vendors in the market keeps growing on paper.