Open Data Center for AI

About This Document

This strategic white paper from the Open Compute Project outlines its Open Data Center for AI program, focused on deploying gigawatt-scale AI data centers. It calls for open reference site designs, standardized grid-interconnection practices, SLAs for third-party operators, integrated telemetry across IT and facility infrastructure, and consistent power-assessment methodologies. The initiative aims to accelerate AI data center construction and grid connection, increase infrastructure reuse, and reduce the risk of supply-chain misalignment.

Key Takeaways

• Rising AI rack densities and the shift toward gigawatt-scale data centers are blurring the boundary between IT systems and facility infrastructure, making it essential to align site design with the compute architecture.
• OCP proposes moving from bespoke AI data center designs to open reference designs built around common power and cooling building blocks, with the aim of reducing time to deployment, commissioning risk, and supply-chain uncertainty.
• Access to power is identified as the primary constraint on site selection and data center expansion; large AI data centers should be viewed not only as passive loads, but also as potentially flexible participants in the power system.
• Grid integration requires standardized technical requirements, including load ride-through during voltage dips (LVRT), as well as practices for deploying BESS, microgrids, and on-site turbines.
• Open SLAs, standardized IT and OT telemetry, and sound power-assessment methodologies are important for avoiding excessive overprovisioning, reducing energy and water consumption, and improving the reliability of third-party facilities.

Key Figures

Target rack power in technology roadmaps

1 MW

The document states that roadmaps anticipate rack densities increasing to this level within the next several years.

Off-peak grid utilization

50 % или менее

According to the document, the grid operates at this level or below for most of the time.

Share of annual hours represented by peak demand

0.25 %

Avoiding peaks that account for only this share of annual hours could free up capacity.

Potentially available capacity in North America

75 GW

Estimate from the cited Duke University study, assuming peak loads representing the stated share of annual hours are avoided.

Document version

0.5

Listed in the version history for the white paper dated March 18, 2026.

Practical Value for Data Center Owners

For data center owners and project teams, the paper provides a checklist for developing a gigawatt-scale AI site: assess compatibility with multiple AI platforms, incorporate liquid cooling and modern power distribution, align grid-connection and LVRT requirements early, and evaluate load controllability, BESS or microgrid options, and local generation. For colocation and neocloud operators, SLA review is especially important: contract terms can artificially increase required capacity, water consumption, cost, and unused reserve capacity.

Where It Applies

Topics