Scaling to 1 MW Racks: Power and Cooling for AI Infrastructure

One of the most important announcements at the summit focused on increasing rack-level power delivery. The Mount Diablo initiative, developed by Google, Meta and Microsoft, proposes a new power architecture delivering up to 1 megawatt per rack using a 400 volt direct current system. This represents a significant increase from the 48 volt systems currently in use. The design separates power from compute, placing conversion in a dedicated rack to improve cooling and hardware density.

To support this step forward, liquid-cooled busbars are being introduced. These are designed to handle high loads while controlling temperature. Some of the first demonstrations of this technology were on display at the summit, showing clear potential to support 140 kilowatts and beyond in future deployments.

Google also introduced the fifth generation of its coolant distribution unit, called Project Deschutes. The unit is engineered for redundancy and scale and is scheduled to be contributed to the OCP community later this year.

At Circle B, we are already working with customers who are preparing for this transition. We help organizations plan for higher density, more efficient rack-level power and cooling systems using a combination of OCP-based and traditional infrastructure.

Watch: Data Center Evolution: Powering the AI Revolution Sustainably – Google

Open Hardware for Hyperscale AI

Meta presented its experience with building infrastructure for Llama, their large language model with over a billion downloads. Their approach is based on modular, open server designs that allow fast deployment and scale. These same designs are available to the broader industry through OCP, and their flexibility makes them ideal for mid-size AI infrastructure as well.

Ethernet is becoming the preferred interconnect for modern AI clusters. Many in the industry, including OCP members, are moving toward Ultra Ethernet standards that support high performance with vendor-neutral implementations.

Microsoft emphasized that open infrastructure must also be secure. Their keynote addressed the need for verifiable firmware, trusted boot processes, and secure hardware modules. These are not limited to hyperscalers. They are equally important for enterprises building scalable AI platforms.

Europe continues to play a growing role in shaping OCP’s direction. With more than one quarter of OCP solution providers now located in the region, innovation from Europe is helping define how open infrastructure will be built and supported.

Watch: Innovations in Open-Source AI and Hardware – Meta

Sustainability in Practice

At the summit, it was clear that sustainability is no longer an optional goal. It is a requirement. One session introduced a tool that allows operators to measure Scope 1, 2 and 3 emissions, which are increasingly part of procurement criteria. Transparent reporting is becoming a basic expectation for infrastructure providers.

The community also shared examples of design strategies that reduce waste and extend hardware life. These include memory reuse using the CXL standard, as well as liquid-cooled systems that reduce total energy use over time.

At Circle B, we support sustainability goals not just through system efficiency, but by designing for long-term adaptability. That means helping customers align infrastructure with policy, compliance and performance targets across Europe.

Watch: Cloud Services Environmental Impact Assessment Tools – OVHcloud

Further Resources

More information and recordings from the summit are available on the Open Compute Project’s website and YouTube channel:

• Open Compute Project

• OCP YouTube – EMEA Summit 2025 Playlist

Circle B at the Summit

At Booth C5, we showcased a Rittal Open Rack V3 solution featuring hardware from Bel Fuse and GIGABYTE. The system was a working example of how trusted suppliers and open standards come together in fully integrated, deployment-ready platforms.

We also took time to join Techarena for a recorded interview on how Circle B connects innovation from the OCP community with operational needs in European data centers.

Watch the interview: Circle B Delivers End-to-End OCP Innovation in Europe

Looking Ahead

Many of the advances on display at the summit originated with hyperscale platforms, but their relevance extends much further. Power systems, cooling methods, and modular hardware designs developed for large data centers are beginning to shape the next generation of infrastructure for enterprise and colocation providers alike.

At Circle B, we help our customers prepare for that future. Our focus is on solutions that are practical, efficient and sustainable. Whether based on open hardware or hybrid designs, we deliver infrastructure that is built for the long term.

Circle B.
Innovative Infrastructure. Simplified.