NVIDIA has listed Anthropic and OpenAI as some of the initial users of its Vera chip.

      During his Computex keynote, Jensen Huang dedicated part of his speech to announcing a guest list. Notable among the initial major users of Vera, Nvidia's latest in-house processor, are Anthropic, OpenAI, SpaceX, and Oracle, as shared with the audience in Taipei on Monday. Nvidia, a company that established its reputation on graphics chips, now seeks to gain recognition for its CPU as well.

      Vera succeeds Grace, Nvidia's earlier data-center processor, but the company is presenting it as a complete redesign instead of just an upgrade. It features 88 of Nvidia's proprietary “Olympus” cores, moving away from the standard Arm Neoverse cores used in Grace, and Nvidia states that it has entered full production. The marketing narrative suggests that Vera is tailored for the era of AI agents, software that organizes and executes tasks rather than merely responding to queries. Nvidia asserts that this chip performs agent-related workloads more quickly than Intel and AMD's x86 processors and offers up to 1.2TB/s of memory bandwidth in conjunction with the cores.

      Initial independent benchmarks indicate that Vera outperforms Intel’s Xeon and AMD’s EPYC in several areas, which aligns with the comparisons Nvidia wishes to emphasize. The identity of the customers is equally significant as the technology itself. Anthropic and OpenAI, in particular, have emerged as key consumers driving current developments, and Nvidia highlighting them as early users reinforces the intended direction for the chip.

      According to Nvidia's account, the initial units were hand-delivered in May prior to the public announcement made in Taipei. For Nvidia, venturing into CPUs represents a strategic move. Their GPUs are already integral to nearly every large AI system, and the data-center processor has traditionally been a component they acquired rather than developed. By designing their own CPU, Nvidia eliminates reliance on others, allowing them to market both the CPU and GPU as a cohesive unit, which forms the basis of the Vera Rubin platform that Huang has referred to as the largest product launch in Taiwan's history.

      Huang has emphasized Taiwan as the “epicentre” of this initiative throughout the year. Oracle Cloud Infrastructure is expected to be the first hyperscaler to implement Vera on a large scale, with wider availability anticipated in other major cloud platforms by the second half of 2026. This timeline is a crucial aspect to monitor. Announcing full production and a lineup of prominent names is straightforward in June; however, successfully shipping substantial quantities to paying customers remains a challenge that lies ahead.

      What Nvidia did not disclose in Taipei was the pricing or the specific number of units that any of the mentioned labs have committed to. The guest list was presented, but the financial details were not revealed.