Analysis
Nvidia will pay to non-exclusively license the AI model-development technology built by Poolside, a two-and-a-half-year-old startup that trains large language models to write and refactor code, The Information reported Thursday. Poolside's founders -- Jason Warner, GitHub's former CTO, and Eiso Kant -- are staying on to run the remaining company independently. Pulse previously covered Nvidia's $105 billion backing of OpenAI's Ohio data center, part of the same pattern of Nvidia locking up AI-model and infrastructure partners with capital. The deal terms, confirmed on X by Newcomer's Eric Newcomer, who first broke them:
- Licensing fee -- $6 billion, to non-exclusively license Poolside's model-development technology (The Information)
- Direct investment -- $1 billion in Poolside at a $12 billion valuation
- Employee offers -- roughly 109 Poolside employees have received offers to join Nvidia directly (Eric Newcomer)
Poolside was founded in early 2023 and has a funding and revenue history that puts this week's price tag in context:
“The $12 billion valuation also assumes Poolside's remaining team can still execute against that roadmap with a smaller bench, which is not guaranteed.”
- Prior funding -- roughly $626 million raised across earlier rounds, including a $500 million round led by Bain Capital Ventures in October 2024 that valued the company at $2 billion, with eBay, Citi Ventures, Redpoint, Adams Street and Nvidia itself among the backers
- Products -- Malibu, a large model for multi-file code generation, test writing and refactoring, and Point, a smaller, quantized model built for sub-200-millisecond code completion inside IDEs
- Revenue multiple -- estimated annualized revenue of around $50 million as of 2025, according to Latka, on a roughly 256-person team -- putting the $12 billion mark at something like 240 times revenue, an extraordinarily rich multiple even by 2026 AI standards
A pattern, not a one-off
This is Nvidia's second deal built around technology licensing and staff absorption rather than an outright acquisition. In December 2025, Nvidia reached a similar licensing arrangement with Groq, the inference-chip startup, paying to use Groq's technology without buying the company outright. Google, Microsoft and Amazon have all run comparable plays -- Google's deal with Character.AI, Microsoft's with Inflection, Amazon's with Adept -- structuring payments as licensing and hiring rather than M&A, which lets a well-capitalized buyer absorb a startup's technology and talent without triggering the antitrust review a formal acquisition of this size would likely draw.
The strategic logic for Nvidia is straightforward: CUDA and Nvidia's hardware stack dominate AI training, but the application layer that sits directly on top of coding workflows -- where GitHub Copilot, Anthropic's Claude Code and OpenAI's Codex increasingly compete for developer mindshare -- is not something Nvidia controls. Poolside's training data and model architecture give Nvidia a foothold in that layer, plus a pipeline of engineers who understand how to build coding-specific models, without Nvidia having to build that expertise from scratch or make an acquisition that regulators would scrutinize.
The counterweight
A 109-person hiring offer is not the same as an acquisition, and it's not yet public how many of those 109 accepted or what Poolside looks like with a meaningful chunk of its technical staff gone. Nvidia's own framing -- that Poolside "continues to run independently" -- is doing a lot of work here; a company that just sold its core technology non-exclusively to its own investor, and lost over a hundred employees to that investor in the same week, is not independent in any operational sense that matters to customers evaluating whether to build on Poolside's roadmap going forward. The $12 billion valuation also assumes Poolside's remaining team can still execute against that roadmap with a smaller bench, which is not guaranteed.
There's also a reasonable read that Nvidia is simply securing optionality rather than making a directional bet: $6 billion in licensing fees is real money, but relative to Nvidia's balance sheet it is a rounding error, and paying for non-exclusive access means Poolside can still sell the same technology to Nvidia's competitors. That undercuts the idea that this is primarily a moat-building move rather than a hedge.
