On August 24, 2026, NVIDIA released the Vera Rubin and Blackwell chips, setting a new standard for artificial intelligence performance per watt. As artificial intelligence agents expanded from single-turn interactions to multi-step workflows involving reasoning, tool invocation, and sub-agent coordination, the average number of prompt tokens per request increased approximately four times in real-world applications. On August 27, 2026, Ian Buck, vice president of NVIDIA, announced that the Vera CPU system had been officially put into mass production and delivered on a large scale to the AI ecosystem, marking the scaled deployment of processors designed specifically for agents.
NVIDIA Vera Rubin and Blackwell chipsets have set new standards for artificial intelligence performance per watt. As artificial intelligence agents expand reasoning from single-turn interactions to multi-step workflows, involving reasoning, tool invocation, sub-agent coordination, and context continuation, this shift is already evident in raw consumption: according to the OpenRouter “Artificial Intelligence State” report, the average number of tokens required per request has increased by approximately four times with the use of 100 trillion real-world tokens.
Ian Buck, vice president of NVIDIA, has begun delivering the Vera CPU system on a large scale to the AI ecosystem. Vera is NVIDIA’s first processor designed specifically for agents, and it has now been officially put into mass production. This development marks Vera’s large-scale deployment in the AI ecosystem.