Nvidia has reported record quarterly revenue of $81.6 billion for the three months ending April 26, up twenty percent from the previous quarter, driven by $75.2 billion in data centre revenue. The company forecast $91 billion for the coming quarter and authorised $80 billion in share repurchases. CFO Colette Kress said the company’s Blackwell architecture had been adopted by every major hyperscaler, cloud provider, and model maker.
Beyond the headline numbers, the quarter revealed a dramatic shift in Nvidia’s investment strategy. Its privately held stakes nearly doubled from $22 billion to $43 billion, driven by $18.5 billion in purchases — a sharp acceleration from $649 million the prior quarter. The company has also committed $30 billion to OpenAI, and CEO Jensen Huang told investors that compute capacity being built for Anthropic this year and next would be substantial, describing previous coverage of Anthropic as effectively zero.
Huang used the earnings call to introduce what he called a major new growth driver: Vera, Nvidia’s purpose-built agentic AI CPU. He argued that while AI model training relies on GPUs, the agents running on top of those models operate predominantly on CPUs — and that Vera, optimised for processing tokens at maximum speed, is purpose-built for that workload. Nvidia has already recorded $20 billion in standalone Vera CPU sales this year.
Huang said he anticipated a world of billions of AI agents, each requiring dedicated compute infrastructure, and positioned Vera as opening a $200 billion addressable market the company had never previously targeted.
