AI Pulse · AI Trends Pulse

Nvidia is drawing a clearer line between two parts of an AI agent workload. GPUs do the heavy model computation. A new CPU, called Vera, is designed to handle the surrounding work, including tool calls, orchestration and data processing. That matters because an agent can spend a lot of time deciding which system to query, moving data and running steps in a workflow. If those jobs tie up expensive GPU capacity, the whole setup becomes slower and more costly.

Vera has 88 Olympus cores and 1.2TB/s of bandwidth. Nvidia says SpaceXAI is already deploying it, and a Vera Rubin system is planned for the Starmind AI satellite in Q4 2027. The important signal is bigger than one chip or one customer. Nvidia is now treating agent orchestration as its own hardware category, not just a side task for general servers or GPUs, according to Nvidia’s announcement.

For business owners, this points to where AI systems are heading. The challenge will not only be picking the best model. It will be managing the work around it, connecting systems, checking data, routing tasks and tracking what the agent did. This is the kind of thing we build into an AI command centre, giving teams a practical way to oversee agents and the business processes they touch.