AI data centers generate such extreme and repetitive power surges from GPU clusters that traditional electrical systems can no longer handle them, forcing a fundamental redesign of how power is delivered to compute infrastructure. Two companies announced a strategic collaboration on July 22 to address this crisis: DG Matrix, which builds solid-state transformers, and Skeleton Technologies, which specializes in fast-response energy storage systems.

Why Are GPU Power Surges Becoming a Critical Infrastructure Problem?

Modern artificial intelligence (AI) workloads don't consume power steadily. Instead, graphics processing units (GPUs) create what engineers call "pulse loads," sudden spikes in electricity demand that repeat thousands of times per second. These bursts exceed the dynamic response capabilities of the power grid and conventional backup power systems like uninterruptible power supplies and batteries.

The scale of the problem is staggering. Global data center electricity consumption reached 415 terawatt-hours in 2024, and the International Energy Agency projects that AI data center electricity demand alone will reach 945 terawatt-hours by 2030, exceeding Japan's entire national electricity usage. As nations race to build sovereign AI infrastructure, the power demands are accelerating faster than grid operators can adapt.

How Is the Industry Solving the GPU Power Pulse Problem?

The DG Matrix and Skeleton Technologies partnership integrates three complementary technologies to absorb and manage these extreme load transients:

• Interport CM Medium-Voltage Transformer: Converts standard grid power (12 to 34.5 kilovolts AC) directly to high-voltage direct current, eliminating intermediate conversion steps that waste energy and slow response times.
• Interport CL Sidecar Solution: Positioned directly in the data center white space to mitigate GPU pulse loads before they propagate upstream to the broader electrical infrastructure.
• Interport Flex Outdoor System: A grey-space product that can be skid-mounted and paralleled into the complete Interport 360 solution, providing flexibility for different data center layouts and expansion scenarios.

Skeleton Technologies' energy storage systems respond in microseconds to minutes, absorbing peak loads before they destabilize upstream infrastructure. DG Matrix's Interport platform routes power in real time across alternating current (AC) and direct current (DC) sources, loads, and storage at multiple voltage levels, creating what amounts to an intelligent nervous system for data center power.

"The shift to 800 VDC power architecture represents one of the most important evolutions in the data center industry. As power demands for AI workloads become even greater, fast-acting energy storage is essential to ensuring that GPU bursts don't destabilize electric grids," said Taavi Madiberk, CEO and co-founder of Skeleton Technologies.

Taavi Madiberk, CEO and co-founder of Skeleton Technologies

What Is 800 VDC and Why Does It Matter?

The industry is transitioning to 800-volt direct current (800 VDC) power architecture as the standard for next-generation AI data centers. This higher voltage enables more efficient power delivery over longer distances with less energy loss. Research from SemiAnalysis projects that approximately 39 gigawatts of incremental AI data center capacity will use 800 VDC architecture by 2030, creating a solid-state transformer market worth roughly $13 billion.

Both DG Matrix's Interport platform and Skeleton Technologies' GrapheneCBU800 and GrapheneBBU800 energy storage systems are built natively for 800 VDC, positioning them at the center of this industry transition. This compatibility matters because retrofitting existing infrastructure is far more expensive and time-consuming than building new systems with the right voltage from the start.

"DG Matrix is building the power fabric for the intelligence age. By integrating Skeleton Technologies' fast-response energy storage with our Interport solid-state transformer platform, we can advance a more complete architecture for AI data centers: one designed for dynamic pulse loads, resilient power, and faster deployment," said Haroon Inam, CEO and co-founder of DG Matrix.

Haroon Inam, CEO and co-founder of DG Matrix

How Does This Solve the Broader Data Center Deployment Challenge?

Speed to power is speed to inference, meaning the faster a data center can be connected to reliable electricity, the faster it can begin serving AI workloads. The integrated DG Matrix and Skeleton Technologies solution enables hyperscale operators, neocloud providers, and colocation facilities to bring AI capacity online in months instead of years, at scale, anywhere in the world.

This acceleration matters because nations are committing unprecedented capital to sovereign AI infrastructure. France launched a 109 billion euro sovereign AI initiative in early 2025, with 87 billion euros earmarked specifically for constructing physical data centers on French soil. The United Kingdom allocated 1.1 billion pounds to its AI Hardware Plan, including a 500 million pound Sovereign AI Fund. India initiated procurement of more than 10,000 GPUs under its sovereign mandate, and Japan's domestic capital expenditure on AI infrastructure is forecast to reach 694.6 billion yen in 2025.

Without solutions to the GPU power pulse problem, these massive infrastructure investments would face crippling delays. Grid operators cannot accommodate sudden, extreme load swings without destabilization. The DG Matrix and Skeleton Technologies partnership directly addresses this bottleneck by ensuring that power delivery systems can respond as dynamically as the compute hardware itself demands.

The collaboration signals a broader industry recognition: as AI compute density increases and power demands become more dynamic, static power architectures are obsolete. The future of AI infrastructure depends on intelligent, responsive electrical systems that can absorb transients, support fault ride-through, and deliver backup power with the speed and cycle life that modern GPU workloads require.