Nvidia delivered results consistent with its track record. Third-quarter revenue more than doubled to $116 billion, including $108 billion from data centers. Vera Rubin, Nvidia’s newest AI computing platform and the successor to Grace Blackwell, generated nearly $20 billion in its first quarter of production shipments. That is the fastest product ramp in the company’s history.
“Vera Rubin is off to an extraordinary start,” Kress said. Sovereign AI revenue more than tripled, while physical AI revenue nearly tripled.
Open Models Drive Demand for Agent Workloads
Open models now carry much of Nvidia’s compute demand. Weekly token volume on OpenRouter rose 35x from a year ago, and agentic workloads account for about 80% of tokens. Nvidia said token prices have fallen nearly 60% since May, which encourages greater AI usage. Its Nemotron model is now the most widely used U.S. open model on OpenRouter. Kress said Nvidia’s acquisition of Hugging Face will make AI more accessible.
Microsoft, Salesforce, Palantir, CrowdStrike, Cadence and Synopsys are building products on open models customized for specific industries. Healthcare company Abridge and legal AI startup Harvey are pursuing similar approaches.
Financial services is another growth market. On-premises revenue there reached nearly $5 billion over the past 12 months, more than double a year ago, as institutions deploy Nvidia systems for quantitative trading. Kress expects enterprise revenue to more than double in fiscal 2028.
Kress said Nvidia wants an ecosystem where “every country, company and industry can build its own intelligence.”
Nvidia also introduced an agent safety platform backed by more than 100 partners. Meta’s Muse agent, running on Nvidia infrastructure, surpassed 10 million setups across the U.S. and Canada in six weeks.
Sovereign and Physical AI Widen Nvidia’s Customer Base
Demand now spans hyperscalers, frontier AI laboratories, enterprises and governments. Frontier laboratories are expected to account for about 15% of data center revenue this year.
Governments are investing in domestic AI infrastructure, including through Nvidia’s September collaboration with cloud partners in Australia.
Physical AI is expanding into commercial deployments. Nvidia introduced a humanoid robotics reference design at GTC Berlin, while Mercedes-Benz is developing robotaxi fleets on its automotive platform.
Asked whether demand could weaken in 2028, Kress said agentic AI growth would continue beyond 2027. She closed the call by saying, “Our demand is accelerating. Our supply is growing.”
What Else Stood Out
- Nvidia expects total CPU revenue to more than double in fiscal 2028. Agentic AI is driving CPU demand that Kress said far outstrips supply, and CoreWeave, Meta, Anyscale and OpenAI have adopted its Vera CPU.
- Edge computing revenue reached $8.2 billion, up 42% from a year ago, on stronger enterprise demand for Blackwell workstations and physical AI edge compute. It rose 14% from the prior quarter.
- Nvidia is relying on training partners such as Scale, Mercor, Prime Intellect and Applied Compute to help enterprises tailor open models with their own data, which feeds the demand for Nvidia compute.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.