NVIDIA offers startups compute-for-revenue-share deals and a 43-state US manufacturing push

NVIDIA is extending its reach from selling chips to taking equity-like positions in the AI economy. Its new partnership program gives fast-growing AI startups compute/token credits up front in exchange for a share of their product and cloud revenue, positioning NVIDIA as a full-stack computing intermediary rather than just a hardware vendor. Named deployments include Sharon AI (40,000 GPUs) and Firmus building toward 170,000.

The structure is strategically potent and controversial. By financing customers' infrastructure in return for ongoing revenue, NVIDIA deepens lock-in and captures upside beyond the one-time GPU sale — but it also concentrates the AI supply chain further around a single company. Critics have long worried about NVIDIA's gravitational pull; revenue-sharing intensifies that concern.

Simultaneously, NVIDIA touted US manufacturing across 43 states with TSMC, Foxconn and Wistron — a politically timed message about reshoring and supply-chain resilience that aligns neatly with the administration's industrial priorities. It also detailed a ~50x Omniverse NuRec speedup via Nsight profiling, keeping its software moat in view.

Competitive context: the revenue-share model competes indirectly with the neo-clouds (CoreWeave, Nebius) that were already rattled by Meta Compute this week — NVIDIA both supplies and now partly finances the compute layer. What to watch: how many startups accept revenue-share terms, whether regulators scrutinize the arrangements as anticompetitive, and the actual output ramp of the 43-state manufacturing claim.