Nvidia is putting up $12 billion to reverse-acquihire Poolside, the code-generation model startup, folding its research team into Nvidia's own AI lab rather than letting the model ship as an independent product. A reverse acquihire means Nvidia isn't buying Poolside's software business, it's buying the engineers who built it and dissolving the standalone company: Poolside's founders and researchers move onto Nvidia's payroll, and the code-generation model they built stops being sold to outside customers. The deal follows the same shape as Nvidia's compute strategy generally: it already sets the price of the H100 and B200 chips (the GPUs that train and run these models) that every AI lab depends on, and now it's paying to control the software layer that decides how well those chips get used, rather than leaving that layer to a startup Microsoft or Google could buy instead.

The number that matters here is $12 billion for a company that, as an independent vendor, would have sold API access to its coding model by the token, the way OpenAI or Anthropic do. Nvidia doesn't need that revenue. It needs Poolside's model to make Nvidia's own chips look better at the one task, writing and debugging code, that determines whether an enterprise buys Nvidia's hardware stack over a rival's. For engineering teams at cloud providers like Google Cloud or Microsoft Azure who compete with Nvidia on AI infrastructure, this closes off a supplier: Poolside's technology is no longer available to license, it now exists only to make Nvidia's own products better. The next signal to watch is whether Poolside's model shows up bundled into Nvidia's CUDA developer tools before the end of 2026, the point at which this stops being an acquisition and starts being a product line.