The race to build the next generation of AI “factories” is moving beyond the silicon. While the industry has spent years obsessing over chip architectures and GPU clusters, a quieter but equally critical transformation is happening in the electrical veins of the data center. Hyperscale facilities are beginning to transition from traditional alternating current (AC) to direct current (DC) power distribution to keep up with the staggering energy demands of artificial intelligence.
This shift gained significant visibility during the NVIDIA GTC conference held from March 16–19, 2026, in San Jose, California Eaton at NVIDIA GTC 2026. The event highlighted a coordinated push by infrastructure giants—including Vertiv, Eaton, Delta Electronics, and Schneider Electric—to move away from the complex AC-to-DC conversion cycles that have defined data center design for decades. As AI racks scale from kilowatts to megawatts, the inefficiency of traditional power delivery has become a physical and financial liability.
At the heart of this evolution is the move toward high-voltage DC architectures, specifically the 800 VDC standard. By simplifying how electricity travels from the grid to the GPU, operators can drastically reduce energy waste, lower the amount of raw materials required for construction, and shrink the physical footprint of power equipment. For the global tech industry, this represents a fundamental pivot in how the physical layer of AI is engineered.
The Efficiency Gap: Why AC Distribution is Failing AI
For years, data centers have relied on AC utility power since This proves the standard for long-distance transmission. However, the path from the grid to a computing chip is currently a convoluted series of conversions. Power typically enters a facility as medium-voltage AC, is stepped down to low-voltage AC via transformers, converted to DC within an uninterruptible power supply (UPS) for battery storage, converted back to AC for distribution, and finally converted again to low-voltage DC (typically 54 V DC) at the server level to power the chips.
This “double conversion” process was sufficient when traditional computational racks drew roughly 10 kW each. However, AI workloads are pushing rack power requirements toward 1 megawatt (MW). At this scale, every conversion step incurs energy loss and generates heat, creating a compounding efficiency problem. The physical requirements also become untenable; according to an NVIDIA blog, a single 1-MW rack could require as much as 200 kilograms of copper busbar. For a gigawatt-scale data center, this would equate to 200,000 kg of copper.
The transition to DC architectures aims to eliminate these intermediate steps. By converting grid power directly to 800 V DC at the perimeter of the data center, operators can bypass multiple conversion stages. This reduction in hardware leads to fewer fans, fewer power-supply units, and lower heat dissipation, which in turn improves overall system reliability.
The Advantages of 800 VDC Architecture
The move to high-voltage DC is not just about removing steps; it is about the physics of power transmission. Switching from 415-V AC to 800-V DC allows for 85 percent more power to be transmitted through the same conductor size. Because higher voltage reduces the current demand, resistive losses are lowered, making the entire transfer process more efficient.
The practical implications for gigawatt-scale facilities are substantial. This architecture can reduce copper requirements by 45 percent and provide a 5 percent improvement in overall energy efficiency. These gains contribute to a projected 30 percent lower total cost of ownership (TCO) for massive AI factories. In this model, power is converted from medium-voltage AC to roughly 800-V DC and distributed via a DC bus, with compact DC-to-DC converters at the rack level stepping the voltage down for CPUs and GPUs.
This trend is already manifesting in global markets. Reports from technology advisory group Omdia indicate that higher voltage DC data centers have already appeared in China. In the Americas, the Mt. Diablo Initiative—a collaboration between Meta, Microsoft, and the Open Compute Project—has been experimenting with 400-V DC rack power distribution.
Industry Response and the Road to 2026
Several major infrastructure vendors are now racing to commercialize these systems. Delta Electronics has showcased power, cooling, and microgrid solutions designed to bolster the 800 VDC architecture of next-gen AI GTC 2026 reports. Specifically, Delta has released 800-V DC in-row 660-kW power racks featuring 480 kW of embedded battery backup units.
Vertiv is developing an 800-V DC ecosystem that integrates with NVIDIA Vera Rubin Ultra Kyber platforms, with commercial availability expected in the second half of 2026. Similarly, Eaton is advancing 800-V DC innovation through the use of medium-voltage solid-state transformers (SST) to serve as the core of DC power distribution. SolarEdge is also developing a 99%-efficient SST paired with a native DC UPS.
Despite these advancements, a full industry transition remains a steep climb. Patrick Hughes, senior vice president of strategy, technical, and industry affairs for the National Electrical Manufacturers Association, notes that much of the current innovation is still centered at the 400-V DC level. He emphasizes that the industry requires a coordinated ecosystem—including standardized connectors, sensing, and protection components—before widespread adoption can occur. This will require significant retooling of manufacturing capacity and long-term demand commitments to justify the capital investment.
Key Technical Transitions at a Glance
| Feature | Traditional AC Architecture | Next-Gen 800 VDC Architecture |
|---|---|---|
| Conversion Steps | Multiple (AC $rightarrow$ DC $rightarrow$ AC $rightarrow$ DC) | Simplified (AC $rightarrow$ DC) |
| Copper Requirement | High (up to 200kg per 1MW rack) | Reduced by approximately 45% |
| Power Density | Optimized for ~10kW racks | Scalable to 1MW+ racks |
| Energy Loss | Higher due to multiple conversions | Lower; ~5% efficiency improvement |
The transition to DC power is a necessary response to the “five-layer cake” of AI described by NVIDIA CEO Jensen Huang, which spans from the physical land and power shell up to the applications. As the industry moves from adding capacity in the teens of gigawatts per year to potentially over 100GW in a decade, the physical infrastructure must evolve or risk becoming the primary bottleneck for AI growth.
The next major milestone for this technology will be the commercial rollout of integrated 800-V DC ecosystems in the second half of 2026. We will continue to monitor the stabilization of safety frameworks and industry standards that will determine how quickly these “AI factories” can be deployed globally.
Do you believe the shift to DC power will accelerate the deployment of AI infrastructure, or will the lack of standardized safety frameworks slow the transition? Share your thoughts in the comments below.