NVIDIA rolls out revenue-sharing for GB300 clouds

NVIDIA on July 1, 2026 unveiled a new “revenue‑sharing and credit‑support” program that lets AI cloud operators bring Grace Blackwell GB300 infrastructure online without paying the full capital cost up front. The company says it will collect its usual hardware revenue and also take a share of the cloud revenue generated on the covered capacity.

The blog post, co‑authored by CFO Colette Kress and Raj Mirpuri, named Sharon AI and Firmus as the first partners and described a DSX AI factory approach built for large, multi‑tenant regional clouds. NVIDIA framed the move as a way to open compute access for startups, enterprises and research groups that otherwise struggle to finance big GPU builds.

NVIDIA’s announcement gives the company two cash streams from the same silicon: immediate product revenue when it ships racks, plus a recurring, usage‑linked royalty as the GPUs run customer workloads. Company materials describe that structure as aligning incentives between NVIDIA and cloud operators. The exact percentage splits and many credit mechanics, however, have not been publicly disclosed.

The first deployments are large. NVIDIA’s blog says Sharon AI plans to deploy up to 40,000 GB300 GPUs; Firmus’s DSX campus in Batam is described as capable of scaling to 360 MW and as many as 170,000 GPUs. Taken together, those named partners represent as many as 210,000 GB300‑class devices under the new model.

For cloud operators and their customers, the headline benefit is speed: DSX‑style factories can be brought online faster because the vendor helps relieve the upfront financing burden for power, facility and racks. That can shorten the time between site selection and usable capacity — a meaningful advantage for regional AI players chasing low‑latency, sovereign or specialized compute markets.

The arrangement also signals a strategic shift for NVIDIA away from pure one‑time hardware sales and toward recurring, platform‑tied income. Analysts and tech press note the move pushes NVIDIA toward a more vertically aligned, usage‑based revenue model — one that shares both upside from high utilization and downside if capacity sits idle.

That downside is real. Commentators warn the royalty ties NVIDIA’s margin growth to partners’ ability to keep racks busy, while lenders must price collateral whose residual value depends on future chip refresh cycles. If utilization falls short, the usage‑linked stream shrinks and the effective return on NVIDIA’s support declines.

Smaller and mid‑sized “neoclouds” say the model can unblock projects that banks have been unwilling to finance, because NVIDIA’s credit support reduces lender uncertainty about demand and vendor risk. But startups and customers will be trading a share of future revenue for that access, a commercial decision that may limit margin and independence down the road.

The program also echoes recent financing patterns around AI infrastructure. Over the past year NVIDIA has both sold hardware at scale and taken stakes or provided financing tied to large AI customers; commentators have framed the new revenue‑share option as the company formalizing that role as a capital partner rather than a pure supplier. Critics have flagged potential circular‑finance concerns when a dominant vendor also becomes a revenue participant in its customers’ businesses.

For regional policy and competition watchers, the model could accelerate local AI cloud builds in places where capital is scarce or grid and permitting timelines are lengthy. NVIDIA explicitly pitched DSX AI factories as a template for regional scale and sovereign AI needs — a point the company uses to justify offering credit‑support tied to regional operators’ revenue. That could change how national and regional cloud ecosystems develop.

Practical questions remain the most important near‑term watch items: how large the revenue share will be, whether NVIDIA takes equity in some deals, how long contracts last, and how lenders treat GPU‑backed facilities as collateral. NVIDIA’s blog post sets out the framework but leaves these economic levers opaque — which is why market participants and reporters are pressing for the underlying deal terms.

NVIDIA’s revenue‑sharing and credit‑support program is a potential accelerant for regional AI clouds, but it also reshuffles risks and rewards across the supply chain. Over the coming months observers will watch contract disclosures, early utilization rates from Sharon and Firmus, and whether other vendors follow suit or regulators take interest. The announcement on July 1, 2026 represents a clear pivot in how one of the industry’s most powerful suppliers chooses to monetize large‑scale AI compute.