Nvidia is aggressively expanding its role in AI data centers, promoting its new Vera Rubin chip system as a complete platform that includes not only GPUs but also its own CPUs. The move signals a strategic shift for the company, which has long dominated the GPU market but now seeks to own every chip inside the rack. The announcements come ahead of rival AMD's annual product event in San Francisco on Thursday, according to WIRED.
.jpg)
Nvidia's New CPU Strategy
During a technical workshop at its Santa Clara, California headquarters, Nvidia executives briefed journalists on the Vera Rubin system, which pairs one Vera CPU with every two Rubin GPUs. In a single Vera Rubin NVL72 super chip system, there are 36 Vera CPUs for every 72 Rubin GPUs. The company is also selling the Vera CPU as a standalone product and has told Chinese customers these could be ready as soon as August, according to WIRED.
Ian Buck, Nvidia's vice president of accelerated computing and architect of the CUDA software, led the briefings. "We're on a roadmap to crank out new architectures, not just GPUs but CPUs," Buck told reporters. "We're going to keep innovating, because it's do this or die in Silicon Valley." Nvidia CEO Jensen Huang was absent, announcing partnerships in Japan for AI robotics.
Performance Benchmarks and Claims
Nvidia claims the Vera Rubin NVL72 system will process ten times as many tokens per watt as its predecessor, the Grace Blackwell super chip. The company also says its Vera CPU is faster at processing agentic AI tasks compared to rival CPUs from AMD and Intel, though the benchmarks used slightly older generations of competitors' CPUs. Additionally, localized memory subsystems on the new chips offer nearly three times as much memory bandwidth as Blackwell, a key advantage amid the ongoing high-bandwidth memory shortage.
| Metric | Vera Rubin NVL72 vs. Grace Blackwell |
|---|---|
| Tokens per watt | 10x improvement |
| Memory bandwidth | Nearly 3x improvement |
Operational Improvements: Liquid Cooling and Cable-Free Design
Nvidia has made significant design changes to reduce data center complexity. The Vera Rubin NVL72 racks are 100% liquid-cooled, which requires less energy than air cooling. The company also touts a "cable-free compute" approach, dramatically reducing cable counts. Andrew Bell, Nvidia's senior vice president of hardware engineering, noted that installation time per rack can drop from a couple of hours to a few minutes because the system is "hot-swappable." This "plug-and-play" capability, as executives described it, should appeal to enterprise customers seeking faster deployment.
Customer Adoption and Roadmap
Nvidia executives shared that OpenAI already has one Vera Rubin rack in use, according to a brief tour of Nvidia's data center lab. The chip system was first unveiled in spring 2025, and Nvidia has been gradually releasing details. The company's push into CPUs reflects the industry's shift toward more complex agentic AI systems, which demand CPUs for orchestrating data flows, networking, and software tasks alongside GPU computation.
With these developments, Nvidia is positioning itself as a supplier of complete AI systems, not just chips. For enterprise technology leaders evaluating data center infrastructure, the Vera Rubin platform offers a tightly integrated solution that promises significant performance and operational gains, though independent validation of the benchmarks will be important.
