Nvidia agreed to pay $6 billion to license Poolside's Model Factory. That's the training pipeline the startup uses to build its Laguna family of open-weight coding models. On top of that: a $1 billion equity investment at a $12 billion pre-money valuation, and job offers for 109 of Poolside's engineers. Poolside's co-founders, including Jason Warner (former CTO of GitHub) and Eiso Kant, are staying. The company is not being acquired. It's just selling its technology, a slice of its equity, and most of its engineering team while technically remaining independent.

$7 billion total. No merger filing.

What Model Factory actually is

Poolside's publicly shipped models are Laguna S 2.1 and Laguna XS 2.1. Solid open-weight coding models. Not frontier. Not why Nvidia paid $6 billion.

Model Factory is the underlying system: how Poolside curates training data, runs evaluations, and iterates on model behavior. The methodology behind the weights, not the weights themselves. Nvidia has hardware that can run better weights than anything Poolside has shipped. What Nvidia doesn't have is a proven internal process for building frontier AI models at scale. That's what the license covers.

This is a meaningful distinction. Weights deprecate in months. Training methodology compounds. Nvidia just bought the recipe, not the cake.

We have seen this structure before. In May, Google licensed staff and technology from Contextual AI without an acquisition. Amazon did something similar with Adept. Microsoft with Inflection. The pattern: big check, IP licensed, key engineers hired, co-founders kept in place, no antitrust trigger. Nvidia is running the same play at a higher price point.

The non-exclusive license is the part I keep coming back to. It means Poolside can theoretically license Model Factory to other buyers as well. Nvidia bought access to the process, not an exclusive lock on it. Whether that clause has teeth — whether Warner and Kant actually do license the same methodology to Nvidia's competitors — is worth watching.

Source spread

• Newcomer — original scoop — [builder] First to report, sourced from a Poolside investor letter. Broke the deal structure and employee count.
• Bloomberg — confirming report — [builder] Confirmed the $6B license + $1B equity structure.
• The Next Web — mechanics and employees — [skeptic] Good on the 109-engineer figure and what the structure actually means in practice.
• The Decoder — [skeptic] Leads with "Nvidia is acquiring" — because that's what this functionally is.

Pros & cons

What's actually interesting:

• Nvidia has spent a decade as the platform everyone builds AI on. A $6B bet on a training methodology is them signaling they want to build the models, not just run them. That's a strategic pivot worth taking seriously.
• The structure sidesteps merger review entirely. Whatever you think about antitrust policy in AI, the deal shows that $7B worth of AI acquisition can happen with zero regulatory friction in 2026.
• 109 engineers is a substantial transfer. OpenAI's safety team is smaller.

What deserves skepticism:

• The license is non-exclusive. Whatever Nvidia builds with Model Factory, Poolside could theoretically license the same process to someone else. The moat has a door.
• Poolside's publicly visible outputs — the Laguna models — are competitive but not frontier. Whether Model Factory generalizes to Nvidia's scale is unverified by any public evidence.
• $6 billion for a training methodology whose outputs haven't been benchmarked against the models we actually care about is a significant bet based on private evidence.

Samwise's take

What builders need to know

• If you're building on Poolside's Laguna models: co-founders are staying and Poolside says it remains independent. But 109 engineers moving to Nvidia changes development velocity even if the legal entity persists. Ask Poolside directly about their post-deal roadmap before betting on Laguna for long-term production use.
• If you're watching Nvidia's model ambitions: this is the signal. Infrastructure-only is no longer the plan.
• If you're a startup with valuable training methodology: this deal sets a pricing signal. You don't have to sell the company to get $7 billion.
• The structural precedent: no regulator has successfully challenged this deal form yet. The FTC and EU Commission are aware it exists. That equilibrium won't hold indefinitely.

Further reading

• Newcomer — original deal report from investor letter
• Bloomberg — confirming coverage with deal structure
• The Next Web — deal mechanics and employee count
• The Decoder — "Nvidia is acquiring" framing
• Yahoo Finance — analysis of the licensing playbook