Poolside: Nvidia's $6B Deal Includes 109-Engineer Hire and $1B Stake for Nemotron 4
The AMW Read
Adds new structural detail ($6B license + $1B investment + 109-engineer hire targeting a trillion-parameter Nemotron 4) to a deal already reported at $6-7B, showing Nvidia formalizing a repeatable license-plus-hire structure that reshapes frontier open-source model competition.
Poolside: Nvidia's $6B Deal Includes 109-Engineer Hire and $1B Stake for Nemotron 4
New reporting on Nvidia's agreement with AI coding startup Poolside lays out a three-part structure: a $6 billion license fee for its model-building technology, the hiring of 109 of its core engineers into Nvidia's Nemotron team, and a separate $1 billion investment at a $12 billion pre-money valuation. Poolside, founded in 2023 by Eiso Kant and former GitHub CTO Jason Warner and known for its Laguna coding model, told shareholders it faced a failed emergency fundraise last year for a 40,000-GPU GB300 cluster and would have run out of compute within a year without a rescue. Nvidia's capital and infrastructure closed that gap; Kant, Warner and COO Margarida Garcia remain in place as Poolside stays nominally independent.
This is Nvidia's third license-plus-hire deal, after similar arrangements with Groq and Enfabrica, and it marks a shift from supplying compute to frontier labs toward building a frontier model of its own. The 109 engineers go directly into Nemotron 4, described as targeting trillion-parameter scale, following this month's smaller Nemotron 3.5 Lightning release. Jensen Huang has argued publicly that US AI leadership requires a strong open-weight ecosystem rather than one closed frontier model — a stance that follows Nvidia's Nemotron Alliance with Mistral and Perplexity and its investment in Reflection AI. It also puts Nvidia in tension with OpenAI and Anthropic, its largest GPU customers, as a potential model competitor.
For builders and investors, the structure matters as much as the target: Nvidia is formalizing a repeatable way to buy model talent and IP while avoiding a full M&A review, a structure other compute-rich players may copy. It also resets the compute bar for "leading open-source model" claims — open-weight labs without a hyperscale compute partner may struggle to keep pace with a trillion-parameter, Nvidia-backed effort.