The Open Compute Project (OCP) is urging the industry to develop standards for AI data center infrastructure.

In an open letter shared ahead of the group’s Global Summit in San Jose, OCP members contend that for all the investment in networking gear, GPUs, and liquid cooling systems, an overall lack of standards around data center design and deployment is negatively impacting buildout.

“To keep pace with the exponential growth in demand, we need to move toward common infrastructure standards within the data center industry that encompass an interoperable model for data center infrastructure,” the letter reads.

OCP members, including AMD, Google, Nvidia, and Meta, all signed onto the letter, which saw the association launch working groups looking into potential specifications around data center power, cooling, and telemetry.

A dedicated working group will also examine potential mechanisms related to mechanical operations and networking designs.

“The sheer size and immense weight of highly interconnected AI systems has surpassed traditional spatial design of general-purpose compute and storage,” OCP’s letter contends. This creates significant spatial challenges, making many current data halls incompatible.

The association’s networking and mechanical-focused group will look into the potential for standardizing physical parameters, including aisle widths, rack dimensions, and weight-bearing capacity, to ensure AI data center hardware can be deployed “without costly retrofits.”

“The long-term vision is a common physical structure that enables the seamless, often with robotic-assisted operations, placement of any compliant rack, achieving true fungibility at the facility level,” the letter reads.

“Alignment on these key interfaces creates a win-win scenario. Data center providers can support a wider range of customers with common, reusable infrastructure, protecting their large capital investments. At the same time, customers benefit because standard interfaces – rather than rigid, prescriptive designs – encourage faster innovation across the ecosystem.

“This is an ambitious undertaking, but we believe it is essential for the future of our industry. We have a choice: continue with fragmentation or come together to build a more open and innovative future.”

At the time of writing, there is no standard for the construction and deployment of AI data centers. There are, however, reference designs and blueprints created by vendors outlining potential ways to improve deployments.

Earlier this year, Zayo teamed with Equinix to launch such a blueprint for AI infrastructure, specifically targeting neoclouds and AI providers. Their AI Infrastructure Blueprint offers a playbook for architecting the underlying network infrastructure connecting AI training facilities with distributed inference points and end users.

In early September, Nvidia revealed it was developing data center reference designs it plans to share with partners. The chip giant’s AI Factory Giga-Scale Reference Designs are set to contain an Omniverse Blueprint digital twin that would provide insights into building high-performance, energy-efficient infrastructure optimized for AI hardware.