NVIDIA has won central-processor business with Modal, a fast-growing cloud-computing startup that rents infrastructure to AI developers. Modal said it plans to begin using NVIDIA’s new Vera processor to host AI agents, the systems designed to carry out long-running tasks for human users. Until now, Modal had run that work on Intel and AMD chips. The deal gives NVIDIA an early foothold in a processor market historically controlled by those two rivals, even as Modal keeps its existing suppliers in service.
Modal said early tests showed NVIDIA’s Vera chip doubled the amount of work its AI agents could complete. The company’s adoption will require deep software changes, and executives emphasized that the result does not amount to abandoning Intel or AMD. Modal plans a mixed fleet in which each job can be assigned to the hardware that performs it best. That approach reflects the practical reality of AI infrastructure: performance, software compatibility, availability and cost all matter alongside headline chip specifications.
The announcement arrives as Modal expands quickly. Reuters reported that the company raised $750 million in a funding round led by Accel at a $15.75 billion valuation, and says it is on track for $1 billion in sales over the next year. Modal also reported that service usage has tripled and revenue has grown more than fifteenfold year over year. Those figures help explain why processor suppliers are competing for cloud platforms that serve AI developers rather than only traditional enterprise workloads.
NVIDIA expects at least $20 billion in Vera chip sales during its fiscal year and values the broader central-processor opportunity at roughly $200 billion. OpenAI, Anthropic and Oracle have also been named among planned users, alongside AI companies such as Perplexity. Modal’s decision is an important customer signal, but it is not a final verdict on the chip market. Intel and AMD are developing newer processors, and Modal’s mixed-fleet strategy leaves room for rivals to compete across future AI workloads.
