Agent AI Needs Dedicated CPUs: NVIDIA Releases Vera Processor, Alibaba, ByteDance, Meta and Others Have Planned Deployment

According to CoinWorld, based on monitoring by 1M AI News, NVIDIA officially announced the Vera CPU at GTC, positioning it as the first processor specifically designed for Agent AI and reinforcement learning (previously announced as part of the Vera Rubin platform). As AI expands from generation and inference to autonomous agent actions, workflows such as planning tasks, calling tools, executing code, and verifying results are rapidly increasing the demands on CPUs. Jensen Huang stated, “CPUs are no longer just supporting models; they are driving models.” The Vera CPU features 88 NVIDIA-developed Olympus cores, each capable of executing two tasks simultaneously through Spatial Multithreading technology, making it suitable for large-scale parallel workloads in multi-tenant AI factories. It uses second-generation low-power subsystem LPDDR5X memory with a bandwidth of 1.2 TB/s, doubling that of general-purpose CPUs and halving power consumption. Cloud service providers planning deployment include Alibaba, ByteDance, Cloudflare, CoreWeave, Lambda, Nebius, Oracle Cloud (OCI), Together.AI, Vultr, and others. On the hardware side, 19 manufacturers such as Dell Technologies, HPE, Lenovo, and Supermicro are adapting to Vera. Michael Truell, co-founder and CEO of AI programming tool Cursor, said they will use Vera to improve throughput and response speed for programming agents. After testing Vera running Apache Kafka-compatible workloads, Redpanda’s latency was reduced by up to 5.5 times. The Vera CPU has also been planned for deployment at Los Alamos National Laboratory, Lawrence Berkeley National Laboratory, the Texas Advanced Computing Center, and others. The Vera CPU has entered mass production and will be shipped through partners in the second half of this year.