Nvidia has contributed its Vera Rubin rack and tray designs to the Open Compute Project (OCP), the industry consortium that publishes shared hardware reference architectures for hyperscale data center operators.

The submission locks the chip company's compute roadmap into version 1.2 of OCP's modular reference design standard, MGX. The package spans cable-free, fan-free liquid-cooled compute trays, interoperable rack-scale interconnect infrastructure, and a liquid-cooled 5,000 amp busbar engineered for higher-power systems, Nvidia said during a press briefing.

What's actually in the spec?

• Cold-plate connectors engineered so technicians can swap cold plates inside trays without unwanted coolant spillage
• Native support for 19-inch equipment, alongside higher-density rack layouts
• Rack deployments rated for up to 250 kilowatts
• A separate reference design that pulls MGX beyond GPUs to cover Nvidia's new Vera CPU rack

The 5,000-amp liquid-cooled busbar is the standout number. It signals the kind of current headroom future accelerators need once each socket pulls meaningfully more than today's top-bin parts. Vera Rubin's 250-kW ceiling reflects Nvidia's continued bet on direct-to-chip liquid cooling as the only practical path past air-cooled rack densities. The cable- and fan-free tray design pushes that liquid-first approach down to the component level.

Why open it up?

Nvidia repeated the playbook it first ran with Blackwell. The company previously contributed its Blackwell rack and tray designs to OCP, and the same logic applies here: publishing the reference lets OEMs (original equipment manufacturers) and component vendors start fabricating immediately rather than wait on a single buyer with a proprietary spec.

"The bottleneck for AI factory buildouts is how fast we can manufacture and deploy," said Joe DeLaere, Nvidia's product lead for AI infrastructure solutions. "So when the specification is open, the equipment makers and component developers can start fabricating right away and build a supply chain instead of waiting for a singular customer with one proprietary spec. So overall, that accelerates the entire industry. The ecosystem grows, and we grow with them."

The strategic case for Nvidia is straightforward. Open specs buy goodwill inside the standards community. They keep third-party rack makers aligned with Nvidia's mechanical and electrical interfaces. And they leave Nvidia's GPU silicon as the destination compute inside whatever racks the ecosystem builds. Cumulatively, the move reduces Nvidia's own integration burden while expanding the addressable market for its accelerators.

What does this change for hyperscalers?

Mostly speed. With the spec public, contract manufacturers and ODMs (original design manufacturers) can stamp out trays and racks to a published standard rather than negotiate one-off designs per customer. That shortens the lag between each platform generation and the moment third-party hardware reaches a data center floor.

It also gives procurement teams a shared blueprint for evaluating competing rack vendors. Hyperscalers already invested in OCP-compliant deployments can drop MGX 1.2 trays into existing integration pipelines.

And the Vera CPU?

A second MGX specification covers Nvidia's new Vera CPU rack, which the company confirmed has begun shipping. It marks the first time the MGX reference extends to a CPU-only rack, opening Vera silicon to deployments where the GPU is not the primary workload. That widens the addressable market for Vera parts beyond tightly coupled GPU hosts.