Agentic AI is the current trend where LLMs are used along with other tools to perform multi-step actions that react to a dynamic environment to achieve a given objective. One of the key highlights of Agentic AI is that it uses both GPU intensive LLMs and also CPU intensive computational tasks such as calling an external system, creating a file or writing python code to run a numerical computation. As NVIDIA has built powerful and expensive GPUs, the rise of Agentic AI has also made CPUs very important. It would be inefficient if an expensive GPU has to wait for a CPU to complete a certain task. Vera is NVIDIA’s answer to this problem where they provide energy efficient CPUs that work seamlessly with their GPUs to ensure that Agentic AI has access to both powerful CPUs and GPUs that can work together optimally.
Based on NVIDIA’s claim Vera can handle parallel software environments and tool calls up to 1.8x faster than traditional x86 server chips. NVIDIA is thus expanding itself from GPUs that power models into the wider AI landscape where Agentic AI is the central focus. For their enterprise customers this might mean a single vendor unified architecture which lowers the per token cost not just for their LLM usage but for their entire Agentic AI pipelines. This architecture also means that their customers might be locked in into this ecosystem.
Geopolitical headwinds mean GPUs would be heavily regulated and exports might come under sanctions. But high end CPUs are generally not sanctioned and hence it is a market NVIDIA still can expand into in other countries such as China through their Vera CPUs.