NuAxi
NuAxiSemiconductor trade intelligence

Are DC data centers the solution to AI scale?

What happened

This is an industry analysis of data center power architecture (authored by Peak Nano's CMO) arguing that migration to DC data centers is one of the fastest paths to meeting AI power demand. Per Gartner, data centers consumed an estimated 448 TWh globally in 2025, projected to more than double to 980 TWh by 2030; Goldman Sachs Research expects global data center demand to grow about 50% to 92 GW by 2027; Deloitte estimates U.S. AI data center power demand will grow more than thirtyfold to 123 GW by 2035, up from 4 GW in 2024; McKinsey projects $5.2 trillion in global capex through 2030. The article notes Infineon estimates about 17% of an average AC data center's energy is lost to power conversion, and that specialized power semiconductors, central to 800 V designs, could nearly halve those losses to about 9% — lifting a 1-GW facility's IT load from about 830 MW to 910 MW, recovering 80 MW; Nvidia's own estimate is more conservative, up to 5% end-to-end efficiency improvement (~50 MW). Nvidia announced the 800 VDC architecture in May 2025 for 1 MW racks beginning in 2027, released its Vera Rubin DSX AI Factory reference design in March 2026, and expects full-scale production of 800 VDC data centers in 2027 alongside its Kyber rack systems; its MGX-compatible 800 VDC power rack for existing AC facilities arrives in the second half of 2026. Delta, Eaton, ABB, Schneider Electric, and Vertiv have announced 800 VDC power products or architectures; the Mt. Diablo Initiative among Meta, Microsoft, and the Open Compute Project is testing 400 VDC rack distribution. Per Nvidia's figures, 800 VDC enables 85% more power through the same conductor versus 415 VAC, cuts copper requirements by 45%, lowers total cost of ownership by up to 30% at gigawatt scale, and could cut maintenance costs by up to 70%; DSX Max-Q enables 30% more AI infrastructure in a fixed-power facility. The article attributes the urgency to power and grid bottlenecks delaying AI data center development: Wood Mackenzie's 2025 survey found large power transformers averaged 128 weeks for delivery and generator step-up units 144 weeks; GE Vernova is taking gas turbine reservations for 2031 delivery; Dell'Oro projects the data center physical infrastructure market at $120 billion by 2030. All production and efficiency figures are vendor projections or plans, not achieved results.

Impact and signal

Medium impactImpact score: 30/100

Affected components and suppliers

specialized power semiconductorsadvanced film capacitors
800 VDCdata center powerAI power demandpower conversion lossesDC data centers

Source and time

Source: EDN · 11/18/2026 · View original source

Open the NuAxi live feed →

Real-time pricing, supply signals and AI analysis for semiconductor traders