Summary

• DG Matrix and Skeleton Technologies are developing 800V DC power systems for AI datacentres.
• The partnership combines solid state transformer technology with fast response energy storage for GPU heavy workloads.
• AI infrastructure constraints are moving into grid connection, power stability, backup design, and electrical engineering.

Skeleton Technologies and DG Matrix are working together on 800V DC power infrastructure for AI datacentres, as dense GPU systems force operators to look beyond conventional backup power and grid connection assumptions.

The partnership combines Skeleton’s high power storage systems with DG Matrix’s Interport solid state transformer technology. The companies say the combined architecture is designed to respond to rapid GPU load changes, support peak load shaving, provide backup power, and improve resilience in high density AI deployments.

Datacentre power has usually been discussed in aggregate terms: how much energy a site needs, whether the grid connection is available, and how quickly capacity can be built. AI workloads create a more granular engineering problem. GPU clusters can generate sharp and fast changes in power demand as workloads move between idle and compute intensive states, and those changes can put pressure on systems that were not designed for such dynamic loads.

The 800V DC approach is part of a wider shift in datacentre electrical design. Direct current distribution can reduce conversion losses, while solid state transformer systems offer faster switching and more granular control than traditional transformer arrangements. When combined with high power storage, those systems can help smooth load profiles, protect against interruptions, and reduce the need to overbuild infrastructure simply to handle peaks.

Skeleton gives the development a strong European industrial angle. The Estonia founded company has manufacturing and engineering activity across Europe and has built its business around high power energy storage rather than long duration grid batteries. That distinction is important because AI datacentres often need power that responds in milliseconds or microseconds, not only reserve capacity measured over hours.

The commercial pressure behind the partnership is straightforward. AI demand is encouraging larger and denser facilities, but many projects are already delayed or redesigned because of grid queues, electrical constraints, cooling requirements, and local infrastructure limits. As rack densities rise, the power chain becomes part of the compute strategy rather than a facilities issue left to the end of the project.

Operators will still need evidence from deployment. Novel power architectures have to prove reliability, safety, maintainability, cost effectiveness, and compatibility with existing datacentre designs. Customers buying AI infrastructure will also want assurance that any new electrical layer reduces operational risk rather than creating a harder system to troubleshoot during a fault.

Even with those caveats, the direction of travel is clear. AI capacity is not created simply by buying accelerators and finding floor space. It depends on power conversion, storage response, cooling, software orchestration, grid behaviour, and workload patterns working together. DG Matrix and Skeleton are targeting the layer where a growing share of AI infrastructure friction is likely to sit.

That makes the partnership a useful marker of the next stage in datacentre competition. The winners will not only be those with access to chips or capital; they will be the operators and suppliers that can make dense compute stable, efficient, and buildable under real utility constraints.