NVIDIA has officially launched its next-generation Vera CPU, designed specifically for agentic AI workloads in AI factories. The processor features 88 self-developed Olympus cores compatible with Armv9.2, delivers memory bandwidth of up to 1.2 TB/s via a second-generation LPDDR5X memory subsystem, and boasts a TDP of just 450 W (with memory power consumption under 30 W), achieving breakthroughs in both core performance and energy efficiency. Real-world testing by Phoronix reveals that Vera excels under agentic workloads. Compared to the previous-generation Grace CPU, it offers a 1.6× improvement in overall performance; versus the latest 128-core x86 processors, it leads with a combined performance advantage of 1.5×, compiling the Linux kernel on a single socket in merely 20 seconds. Its memory performance stands out particularly well: during STREAM TRIAD tests, it consistently maintains 90% of peak bandwidth, delivering per-core throughput four times higher than traditional x86 systems. Concurrent Prime Intellect tests further confirm that Vera sustains high bandwidth and low latency even as multiple agentic AI tasks run in parallel. Phoronix founder Larabe stated plainly that this represents "the most formidable challenge ever posed to the x86_64 architecture." The first batch of Veras has already been delivered to leading AI companies and cloud providers. Commercial deployment through partners is expected later this year across single-socket/dual-socket configurations and air-cooled/liquid-cooled systems, fully supporting standard data center deployments up to dense agentic AI infrastructure requirements.