NVIDIA shipped a run of developer updates. Robots, data centers and confidential inference are different audiences, but two of them are really about how data moves.
The first is for robotics. In ROS 2, the standard software framework for robots, GPU-resident data can now pass between components without being copied through CPU memory. NVIDIA also showed how to hand the job of updating existing code to an AI coding agent. The second is Topograph, which places workloads on large GPU clusters with the network layout in mind. The third is a walkthrough of Confidential Computing for keeping AI inference on sensitive data private without giving up performance.
A fast GPU still stalls if data is being copied around or routed the long way. The ROS change and Topograph both cut waste in that moving-data layer, a sign that NVIDIA is tending not just to the GPU itself but to how it gets used. Confidential inference answers a different need: regulated industries handling data that cannot leave their control. For the latter two, only the introductions could be confirmed; performance figures are unverified.
Whether the new ROS 2 transport spreads to robot developers' code beyond NVIDIA's own, and whether Topograph's placement gains get backed by concrete numbers.