Nvidia has agreed to pay $6 billion to license the software that startup Poolside used to build its family of AI coding models, while simultaneously investing $1 billion in the company at a $12 billion pre-money valuation. The deal was first reported in a Poolside investor letter by Newcomer on August 20, 2026, and confirmed by Bloomberg and The Information. The investor letter is available on Newcomer and marks the third time Nvidia has structured a major AI transaction this way.
The arrangement is not an acquisition and not an acquihire, according to Poolside's letter to investors. The three co-founders remain at the company, which continues to operate independently. Nvidia is paying for a non-exclusive license to what it calls the Model Factory, the internal platform Poolside used to construct its Laguna family of open-weight coding models. Nvidia also extended job offers to 109 Poolside employees who worked on the Laguna project.
What Nvidia Is Licensing
The Model Factory is the software infrastructure Poolside used to build the Laguna family, which includes the Laguna XS.2 open-weight model for local agentic coding workflows, the 118-billion-parameter Laguna S 2.1, and the 225-billion-parameter Mixture of Experts Laguna M.1 designed for government and regulated enterprise environments. The platform covers data processing, training infrastructure, reinforcement learning tooling, and evaluation frameworks.
A non-exclusive license means Poolside can still sell access to the Model Factory to other companies. Nvidia gets the manufacturing capability without owning the company outright. The $6 billion licensing fee will be distributed to existing investors by the end of 2027, according to Newcomer's reporting on the investor letter.
The Deal Structure
Poolside's pre-money valuation of $12 billion represents a fourfold jump from the $3 billion mark the company hit during its $500 million Series B round in October 2024, which already included Nvidia as an investor. The startup has operated with remarkable efficiency. CEO Eiso Kant told the Latent Space podcast that fewer than 70 people built the Laguna model, and fewer than 115 worked across engineering and research in total.
The structure mirrors two earlier Nvidia deals. In May 2026, Nvidia paid Groq $20 billion for its inference technology and hired its top engineers while Groq remained independent and raised $650 million for what was left. In December, Nvidia executed a similar arrangement with another AI startup. Across all three deals, Nvidia has committed roughly $27 billion, each structured as a license and hiring round rather than a purchase, with every company continuing to operate.
Why Poolside Stopped Building Frontier Models
The investor letter lays out a stark thesis about the economics of frontier AI. Poolside says it was directionally correct in a race where capital requirements went vertical. At the end of last year, the company had a six-week window to raise $2 billion to pay for a 40,000 GB300 cluster coming online in January. They did not close it in time, and they lost the cluster.
The letter argues that next year's frontier will need far more than an order of magnitude larger cluster than what Poolside was planning. The constraint, it says, is not only capital but also physical data center space and contracted compute. Human-level capability will be fully commoditized by open source models, the letter argues, while superintelligence will not.
Poolside did not wait around. The company spun out Poolside Infrastructure Company in January, which is building a 1.2 gigawatt data center in Texas. It appointed a chief executive two months ago and a chief financial officer this week, both appointments predating Friday's licensing announcement.
What This Means for the AI Industry
The deal signals a shift in how Nvidia positions itself in the AI stack. Rather than remaining solely a chip supplier, Nvidia is using licensing agreements, strategic equity stakes, and targeted hiring to move into the software layer that builds AI models. The approach avoids the antitrust scrutiny that would accompany absorbing these companies entirely, while deepening Nvidia's control over the technical talent and proprietary systems that differentiate AI models.
Poolside's Laguna models were pitched as the West's answer to Chinese open-weight coding models like DeepSeek and Qwen. Nvidia builds its own open models in the Nemotron line and has been working toward a trillion-parameter open model, according to The Next Web. Poolside's Laguna family, trained on Nvidia server chips, would compete directly against OpenAI and Anthropic models that currently dominate AI coding assistants.
The deal also reflects the broader reality that Nvidia's Q2 fiscal 2027 revenue of $96.2 billion, reported August 26, was driven by insatiable demand for AI infrastructure. Data center revenue reached $89 billion, up 117 percent from a year earlier. Nvidia's own earnings results, published on the NVIDIA Newsroom, noted that return on invested capital for AI data centers is now less than a year, which speaks to the productivity of the technology being built on Nvidia hardware.
The non-exclusive nature of the license leaves an open question about whether Poolside will become a supplier to multiple AI infrastructure companies, or whether the Model Factory will remain a niche capability. Either way, the $6 billion price tag confirms that the software layer for building AI models is becoming one of the most valuable pieces of the AI supply chain.
See also: Nvidia's Groq deal for agentic AI inference acceleration and Nvidia to acquire Hugging Face for $12.9 billion