Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
Enter your email address below and subscribe to our newsletter

Nvidia CEO Jensen Huang took the stage at GTC Taipei 2026 on Monday to formally showcase the Vera CPU, the company’s first custom-designed processor, built from the ground up for the agentic AI workloads now straining traditional server architectures. OpenAI, Anthropic, and SpaceXAI are among the first organizations to receive the chip, which Nvidia says delivers a 1.5x performance advantage over the latest 128-core x86 processors in agentic workloads.nvidia
The Vera processor packs 88 custom Olympus cores based on the Armv9.2 instruction set, designed entirely in-house by Nvidia rather than using licensed Arm microarchitecture. Each core features a 10-wide instruction front-end and a neural branch predictor, yielding what Nvidia claims is 50 percent faster per-core performance than comparable x86 designs in compilation, scripting, and compression tasks within agentic sandbox containers.inkl
Memory bandwidth is a headline specification: 1.2 TB/s via LPDDR5X across a 1,024-bit interface, giving each core roughly 14 GB/s โ about three times what traditional data center CPUs provide per core. The chip also connects to Nvidia’s Rubin GPUs via NVLink-C2C at 1.8 TB/s of coherent bandwidth. Nvidia says Vera consumes less than 30 watts for memory power compared with more than 100 watts for conventional DDR5 systems.byteiota
Nvidia vice president Ian Buck personally delivered the first Vera CPU systems to Anthropic, OpenAI, SpaceXAI, and Oracle Cloud Infrastructure in mid-May, according to multiple reports. CoreWeave is confirmed as the first cloud provider offering standalone Vera CPU access, and ByteDance is among additional hyperscalers planning adoption.itbrief
Huang told the GTC Taipei audience that the Vera CPU’s early adopters also include SpaceX, and that systems from Dell, HPE, Lenovo, and Supermicro are expected to ship in the second half of 2026.mexc
The Vera CPU is part of a wider Vera Rubin platform that Nvidia says is now ramping into mass production, promising up to 10 times lower cost per token compared with the prior Blackwell architecture. The GTC Taipei keynote also introduced the RTX Spark PC chip, developed with MediaTek and manufactured by TSMC, marking Nvidia’s entry into the Windows PC market in competition with Intel and Apple.cryptobriefing