Nvidia Launches Revenue-Share Model With Sharon AI, Firmus

Nvidia moved a piece of its cloud strategy into the open this week, publishing details of a "revenue-sharing and credit-support" arrangement that lets younger cloud providers put its GPUs on the floor without carrying the whole capex load themselves. On the Nvidia blog, the company said AI clouds will sell Nvidia-powered services and Nvidia will earn "both standard product revenue and a share of the cloud revenue on the supported capacity." The named first partners are Sharon AI and Firmus, alongside a broader shout-out to Baseten, Fireworks AI and Together AI as inference-side customers.

The scale numbers are worth pausing on. Sharon AI is deploying up to 40,000 Grace Blackwell GB300 GPUs; Firmus is building a DSX-aligned AI factory campus in Batam, Indonesia that Nvidia says will scale to 360 megawatts and up to 170,000 GPUs. Those are hyperscaler-class site sizes coming from names most readers outside the AI infrastructure beat had not heard of a year ago, which is the point. James Manning, cofounder and CEO of Sharon AI, called it a "pivotal moment" for delivering "sovereign, large-scale AI compute infrastructure," and Tim Rosenfield, co-CEO of Firmus Technologies, framed it as helping "AI-native companies" get "scalable, energy- and cost-efficient compute."

Why it matters is that the bottleneck for a lot of would-be AI cloud operators has been the balance sheet. Buying tens of thousands of GB300s outright is not a thing a small startup can just do, and lenders have been cautious about hardware whose value curve is not yet well understood. A revenue-share plus credit-support template pushes some of that risk back onto Nvidia in exchange for a piece of the cloud revenue, which is a different shape of relationship than "we sell the chips, you figure out the rest." It aligns Nvidia with utilization rather than only with the initial sale.

The honest caveat is that the Nvidia post is thin on mechanics. It does not spell out what percentage of cloud revenue Nvidia takes, how "credit support" is actually structured, or how long the deals run. Take the framing as reported, not as settled terms. What the reporting also does not give you is a read on what happens if a partner underdelivers on utilization, or how this interacts with existing purchase commitments from hyperscaler customers.

If it works as advertised, the near-term winners are the emerging inference-focused clouds that can now credibly commit to multi-hundred-megawatt sites, and enterprises who get more suppliers to choose from than the top three hyperscalers. The one to watch is whether the same template shows up in Nvidia's next set of partnership announcements, because if it does, this stops being a novelty and starts being how a lot of AI compute gets financed.