NVIDIA Brings Modular AI-Factory Systems to OCP Summit

NVIDIA is showcasing modular computing, networking and cooling systems for AI factories at the Open Compute Project Global Summit in San Jose, with the company targeting infrastructure bottlenecks created by increasingly complex AI workloads.

The summit runs through Oct. 15. NVIDIA’s Vera Rubin platform combines five purpose-built racks into one AI supercomputer designed for agentic AI, or systems that complete multistep tasks rather than respond only to individual prompts.

Vera’s central processing unit completes tasks 1.8 times faster and operates at twice the efficiency of traditional x86 CPUs. NVIDIA said May 31 that Vera Rubin was ramping into full production across more than 350 factories in 30 countries, including 150 factories in Taiwan. At scale, Vera Rubin can deliver 10 times the agent throughput of the previous Grace Blackwell platform.

AI factories combine CPUs, GPUs, data-processing units, storage, networking, power and cooling. NVIDIA’s MGX architecture supports hundreds of configurations across those components, giving system builders a reusable modular design for different deployments.

Networking is another part of the buildout. Spectrum-X Ethernet Photonics uses co-packaged optics and 200-gigabit-per-second SerDes technology. The platform delivers five times the power efficiency, five times longer AI uptime and 1.3 times faster deployment than networks using traditional transceivers.

The open-source Falcon networking protocol is being co-engineered by Google and NVIDIA through the Open Compute Project for high-bandwidth, low-latency data-center traffic. Meta’s infrastructure footprint uses NVIDIA’s Spectrum-X Ethernet platform as part of a multiyear AI infrastructure partnership.

“We’re excited to expand our partnership with NVIDIA to build leading-edge clusters using their Vera Rubin platform to deliver personal superintelligence to everyone in the world,” Meta CEO Mark Zuckerberg said.

“AI agents will be the largest users of computing,” NVIDIA founder and CEO Jensen Huang said. Anthropic head of compute James Bradbury said, “Scaling compute is an important accelerant for the growth of models.”

The infrastructure push builds on NVIDIA programs covering power, cooling and digital-twin systems for AI factories. Its DSX Ready program has qualified providers including Tesla for battery energy storage systems and Vertiv for cooling distribution units, extending the infrastructure work covered in NVIDIA’s provider qualification program.