The 800V Data Center Shift: Why the DG Matrix–Skeleton Technologies Partnership Matters for AI Infrastructure

Skeleton Technologies provides fast-response energy storage systems based on ultracapacitors. These handle peak shaving and backup power, designed specifically to smooth the millisecond-scale power pulses created by GPU clusters, preventing grid disturbances and enabling tighter power provisioning .

Together, Skeleton integrates its storage across DG Matrix's SST platform, creating a combined power-delivery and energy-storage solution that responds to load changes far faster than traditional battery UPS systems .

Why the Industry Is Shifting from 48V to 800V

The core driver is pure physics: resistive power loss (I²R) and copper constraints at ever-higher rack power levels. The numbers tell the story clearly:

The exponential current problem

A 40 kW H100 rack draws roughly 830 A at 48V; a 132 kW GB200 NVL72 draws about 2,800 A; a ~600 kW Kyber/Vera Rubin Ultra-class rack on the 2027 roadmap would draw over 12,000 A if kept at 48V . That much current is physically impractical—it requires enormous copper busbars, creates severe heat dissipation, and makes connector reliability a fire hazard.

Ohm's law in action

Because power loss equals I²R, doubling current quadruples resistive losses. By raising voltage 16.7 times (from 48V to 800V), current drops by the same factor for the same power, slashing distribution losses by approximately 97% .

Conversion-stage elimination

800V DC allows feeding power directly to higher-voltage GPU power stages, eliminating multiple AC-to-DC and DC-to-DC conversion steps. SemiAnalysis estimates this cuts facility-level power consumption by about 5%—at 1 GW of IT load, that is over 50 MW of continuous savings .

Contact physics and rack space

At roughly 2,500 A on a 48V bus, a mere 0.1 mΩ rise in a connector's contact resistance dissipates about 625 W of localized heat—a fire and reliability risk hidden in every bolt . Powering a Kyber-class rack on 54V would consume up to 64 rack units of power shelves alone, leaving almost no room for compute. At 800V, that collapses to a few rack units .

The 800V number is not arbitrary: it is high enough to tame current but stays within safe touch-voltage limits and leverages existing EV and industrial-grade 800V silicon carbide (SiC) and IGBT components .

Projected 800V Adoption by 2030

Forecasts vary depending on whether they measure new capacity or total installed base:

• Citi Research (July 2026) forecasts 79.1% of new global data center capacity will use 800VDC architecture by 2030, driving a structural shift from AC to solid-state transformers .
• SemiAnalysis (May 2026) expects roughly 39 GW of incremental data center capacity to be powered by 800VDC by 2030, creating two new markets: approximately $11 billion in power-side racks and around $13 billion in solid-state transformers .
• Leanrs / industry estimates (June 2026) offer a more conservative view: only 15–25% of all data centers will run on 800VDC by 2030, with most early deployments in hyperscaler greenfield sites .

The gap is not contradictory. Most new buildout is hyperscale AI capacity, so 800V could dominate new builds while the legacy installed base remains at lower voltages.

Analysts at SemiAnalysis project a phased rollout: Phase 1 (late 2026/early 2027) involves early hyperscaler retrofits using sidecar power racks; Phase 2 covers rack-level 800VDC distribution; Phase 3 moves to facility-wide 800VDC; and Phase 4 reaches full native 800VDC from grid to chip .

Competitive Landscape: How This Partnership Fits In

The 800V DC data center race is now crowded, with NVIDIA acting as the central orchestrator. In May 2025, NVIDIA launched the 800V HVDC Supplier Alliance at COMPUTEX with 29 members—including DG Matrix as a founding member .

Key competitors and parallel moves

• Siemens + Fluence + nVent (June 2026): Siemens and Fluence completed a reference design for AI data centers based on NVIDIA's AI factory blueprint. Fluence shares soared more than 40% on the news. The design incorporates Fluence battery energy storage, Siemens power skids, and nVent liquid cooling . This team addresses the same pain point (power + storage integration) but uses conventional power skids rather than solid-state transformers.

• LG Uplus + LS Electric (July 22, 2026—the same week as the DG Matrix–Skeleton announcement): Signed an MOU to jointly develop 800V DC power infrastructure for AI data centers targeting NVIDIA's Vera Rubin platform. LG Uplus will provide telecom and operator expertise; LS Electric will supply power equipment .

• LG Energy Solution + LG Uplus (One LG strategy): LG Uplus aims for cumulative AI data center orders of 5 trillion Korean won (roughly $3.8 billion) by 2030, leveraging LG Group-wide capabilities including LG Energy Solution's battery systems .

DG Matrix's strategic position

DG Matrix is currently the only company with a commercially available multi-port solid-state transformer purpose-built for NVIDIA's MGX and 800V guidelines . Skeleton Technologies complements this with ultra-fast ultracapacitor-based storage, differentiating from Fluence's conventional lithium-ion BESS approach and from LS Electric's more traditional power equipment. The partnership was announced the same week as LG Uplus–LS Electric, signaling an intense race to lock in reference designs before the Vera Rubin generation ships.