Nvidia rolls out a revenue-sharing “AI factory” model to help cloud partners launch large-scale inference faster

Nvidia announced a new way for artificial intelligence cloud providers to deploy large-scale “AI factories” for inference, aiming to shorten the time from AI demand to live compute. Under the framework, cloud partners can sell Nvidia-powered cloud services while Nvidia receives both its standard product revenue and a share of cloud revenue tied to the supported capacity, creating what Nvidia described as a recurring, usage-linked earnings stream.

The company’s premise is that AI workloads are shifting from one-time model training to continuously running inference services that generate output tokens at scale. Nvidia said this trend increases the need for accelerated computing that can come online quickly, remain highly utilized, and support the unit economics of token-scale services.

To address capital constraints, Nvidia said many emerging AI companies have historically had limited access to the finance required to build or expand compute infrastructure. Nvidia’s new model is designed to give AI-native companies, enterprises, research organizations, and regional AI players faster access to fully integrated accelerated systems, without waiting through site selection, power procurement, construction, and hardware bring-up.

A central element is Nvidia’s DSX AI factories, described as multi-tenant accelerated computing designed to serve customer workloads across regions. Nvidia said the agreements are structured to align incentives through revenue sharing plus credit support, which it said can be more capital-efficient for partners scaling infrastructure.

Nvidia did not provide the specific contract mechanics, pricing formulas, or the duration of these revenue-sharing arrangements in its announcement. It also did not disclose any quantified financial targets, such as expected incremental revenue or margin impact, tied to the initiative.

Still, Nvidia said the program is already taking shape with AI cloud partners building DSX AI factories. Among the first participants, Nvidia highlighted Sharon AI and Firmus, with Sharon AI stating a plan to deploy up to 40,000 Nvidia Grace Blackwell GB300 GPUs under the program to support large-scale AI workloads.

The move reflects Nvidia’s broader push to stay at the center of the infrastructure layer for the AI boom, not only through chip sales but also through the operational economics of running models. By tying Nvidia’s returns to utilization and customer usage, the company is effectively trying to convert part of the value of AI compute into a recurring business tied to ongoing inference demand.

For now, what remains unclear is how widely the model will scale, how performance risk and credit risk are allocated between Nvidia and its partners, and what portion of future datacenter buildouts will adopt the approach versus traditional procurement. Investors and customers alike will likely look next for more named cloud partners, clearer implementation details, and any disclosure of how usage-linked economics translate into Nvidia’s reported results.