Different hardware for server and personal agents
Nvidia outlined an expansion of its AI-agent hardware strategy on May 31, 2026, describing Vera server processors while Microsoft detailed Windows PCs based on Nvidia RTX Spark. The announcements addressed both large-scale computing infrastructure and local workloads on personal machines.
Nvidia said Vera was in full production and designed for CPU-heavy work around agents, including code execution, data processing and orchestration. The company identified an 88-core design and plans for both standalone servers and integration with its accelerated systems.
Customer interest did not mean every named organization had already deployed the processor. Nvidia described companies evaluating Vera or planning adoption, and said partner systems would begin arriving in the fall. Those distinctions separate a production milestone from actual availability and measured customer results.
Windows prepares for a local-agent PC platform
Microsoft's announcement described RTX Spark as a platform for Windows laptops and compact desktops, rather than the Vera server product. It listed configurations with as many as 20 Arm-based CPU cores and up to 128GB of shared memory.
The Windows work extended beyond adding a chip. Microsoft described changes to scheduling, power management and memory handling, plus optimization of its Prism compatibility layer for x86 applications. It also outlined GPU access for local AI software through Windows ML and TensorRT.
Microsoft said devices from Surface and other PC manufacturers would begin arriving in the fall. Its performance and battery-life descriptions were vendor claims, with battery testing based on prerelease units, not independently established results for every shipping configuration.
Read the announcements as separate deployment plans
The common theme was running more of an agent's work across specialized computing resources. But a server CPU announcement and a Windows PC announcement answer different deployment questions. Neither, on its own, proves customer adoption, application compatibility across every workload or the eventual economics of running agents.