Nvidia is making a full-scale entry into the data centre central processing unit (CPU) market, taking aim at the server semiconductor market led by Intel and AMD. The move is seen as a strategy to build its own ecosystem beyond graphics processing units (GPUs) and extend its grip across AI infrastructure.

Cryptopolitan, a blockchain media outlet, reported on July 8 that Nvidia has secured AI search startup Perplexity as the first customer for its next-generation data centre CPU platform, Vera.

The move reflects Nvidia's view that CPUs are playing an increasingly important role in AI data centres, not just GPUs. Nvidia says CPUs handle application execution, code processing and data movement when AI models are put into live services, creating bottlenecks. It says this increases the time high-performance GPUs spend waiting and reduces overall system efficiency.

Perplexity plans to adopt Vera in its next-generation AI operating environment. Nate Cupp (네이트 컵), a vice president at Perplexity, said internal tests showed Vera was the best fit for the company's AI service.

In tests conducted jointly by the two companies, Vera processed related tasks about 1.5 times faster than existing server CPUs. The performance gap was confirmed in AI agent tasks that clone code repositories and then run tests in an isolated sandbox environment. Nvidia said Vera performed the task about 1.5 times faster than existing x86-based server processors, and parallel sandbox execution posted up to 1.9 times higher performance.

The results, however, come from tests conducted jointly by Perplexity and Nvidia and have not yet been verified by an independent external organisation.

Nvidia says changes in the structure of AI services are behind its entry into the CPU market. Unlike AI chatbots that simply generate answers, AI agents sequentially handle tasks such as document searches, code execution, calls to external tools and evaluation of results. In this process, CPU single-thread performance and data-processing capability, as well as GPUs, determine overall response speed.

Nvidia expects CPU performance to become a key factor in competitiveness in next-generation data centres, where thousands of AI agents operate at the same time. Jensen Huang (젠슨 황), Nvidia's chief executive, also formalised plans to expand its CPU business at COMPUTEX in Taipei in June. He said building a computer and server CPU business was a "strategy to reinvent the CPU market".

Nvidia estimates the global CPU market at about $200 billion. Reuters forecast that the Vera CPU could generate about $20 billion in revenue in the current fiscal year.

The push is also seen as a head-on challenge to the server CPU market dominated by Intel and AMD. Intel and AMD, both based on the x86 architecture, have long split the server CPU market, but Arm-based processors, which can be custom-designed, have been rapidly gaining share among large cloud operators. Vera also uses an Arm-based design and is positioned around performance optimised for AI workloads.

Market reaction has also emerged. As of Yahoo Finance on July 7, Nvidia shares rose 0.7 percent to $196.93, while AMD and Intel fell 6.5 percent and 9.7 percent, respectively. Some analysis says the declines reflected not only the Vera announcement but also issues related to Samsung Electronics and a broader weakening in investor sentiment across the semiconductor sector.

Nvidia is also continuing follow-on CPU development. Its next-generation server processor, Regel, is being developed based on the Arm v9.2 architecture and is set to be installed on the Rosa platform. The company expects Regel to further improve per-core performance.

The industry is watching whether Vera can prove real competitiveness in the AI data centre CPU market. The outlook is that customer expansion after pilot adoption and third-party performance verification results will determine the success of Nvidia's CPU business.