Nvidia Strikes $6 Billion Deal to License Poolside's Model Tech and Acquire 100+ Engineers

Nvidia holds a near-monopoly on the silicon that powers the world's leading artificial intelligence labs. For years, the company has been content to sell the picks and shovels to frontier model builders. But a new $7 billion transaction signals a fundamental shift: the hardware giant is now aggressively moving to compete with its own largest customers by building its own highly capable models.[1][6]

In a deal first disclosed to investors in late August, Nvidia agreed to pay $6 billion for a non-exclusive license to "Model Factory," the internal software infrastructure developed by AI startup Poolside. Alongside the licensing fee, Nvidia is injecting $1 billion in equity into Poolside at a $12 billion pre-money valuation.[4][5]

Crucially, the agreement is not a traditional acquisition. While Poolside's three co-founders will remain at the helm of an independent company, Nvidia is extending job offers to 109 of Poolside's core engineers and researchers. These employees are being brought in specifically to supercharge Nemotron, Nvidia's in-house family of open-weight AI models.[5][8]

Poolside originally built Model Factory as an automated pipeline for training models that write software code. The system orchestrates the complex data preparation, routing, and reinforcement learning required to turn raw compute into capable intelligence. By licensing this underlying machinery rather than just the finished models, Nvidia gains the architectural blueprint to accelerate its own training runs.[1][8]

The transaction exposes the harsh reality of the current AI ecosystem: compute access dictates survival. Earlier this year, Poolside attempted to raise $2 billion to lease a massive cluster of 40,000 Nvidia GPUs. When that funding round failed, the $12 billion startup was effectively stranded, possessing the talent and the software but lacking the hardware required to stay in the frontier race.[1][5][6]

The structure of the deal—licensing intellectual property and hiring staff without buying the company outright—mirrors a growing trend in the tech sector. Similar to recent maneuvers involving Character.AI and Google, or Nvidia's own previous deals with Groq and Enfabrica, this "acquihire-in-disguise" framework is designed to bypass the lengthy and increasingly hostile antitrust reviews that accompany formal mergers.[4][5][8]

For Nvidia, the strategic prize is the open-weight model ecosystem. The company has increasingly championed open-source AI, with CEO Jensen Huang recently organizing an open letter against premature restrictions on the technology. By releasing highly capable Nemotron models for free, Nvidia lowers the barrier to entry for enterprise AI adoption, which in turn drives sustained demand for its data center GPUs.[6][8]

This open-weight push places Nvidia in a complex geopolitical and commercial battleground. The company is racing to counter the rapid ascent of highly efficient, low-cost Chinese open-weight models like DeepSeek's V4 Flash and Kimi K3, which have begun capturing significant global market share.[2][7]

At the same time, Nvidia's expanding model ambitions put pressure on American frontier labs that rely on proprietary, closed-source approaches. As the cost of intelligence drops and open-weight models become more capable, the premium pricing power of OpenAI and Anthropic faces a severe test. Nvidia is betting that a world of ubiquitous, cheap intelligence will ultimately require more silicon than a world dominated by a few closed ecosystems.[3][6][7]