NVIDIA commits up to $6B to Poolside deal for open-weight AI push

NVIDIA's up-to-$6 billion commitment to Poolside marks its most explicit move yet to become a model provider rather than solely the picks-and-shovels supplier of the AI boom. The deal is expected to fold more than 100 Poolside employees into NVIDIA's team, who will focus on advancing the Nemotron family of open-weight models.

Strategically, this is about controlling the access layer. NVIDIA already dominates the compute substrate; owning competitive open-weight models lets it shape the software stack that runs on its silicon and hedge against a future where model commoditization erodes the premium buyers pay for GPUs. The named rivals — DeepSeek and Kimi K3 — underline that the open-weight frontier is increasingly a Chinese-lab-led arena, per Hugging Face's own 'State of Open Models' data showing Chinese labs releasing larger, more performant open models than U.S. counterparts.

The move dovetails with NVIDIA's broader capital strategy this month, including its role guaranteeing up to $105B in leases for an OpenAI data center and mobilizing a $500B third-party financing platform. Together they sketch a company whose moat is shifting, as CNBC put it, 'from chips to capital' — and now to models.

Skeptics will note the circularity: NVIDIA is simultaneously financing the buildout of AI infrastructure, supplying the chips, and now building the models that consume them. That vertical concentration invites both antitrust attention and the vendor-financed-bubble worries already circulating among developers. Whether Poolside's talent can close the gap to DeepSeek's V4 line — a 1.6-trillion-parameter MoE — is the open technical question.