Google, Microsoft, and Nvidia have joined forces through the Open Compute Project (OCP) to establish 800-volt direct current (800VDC) as an open standard for powering the next generation of high-density AI data centers. The collaboration aims to define common requirements for power distribution, including power conversion, power quality, system interfaces, and safety, as AI workloads drive power consumption to unprecedented levels.
The move represents a significant shift in how data centers are designed and powered. With AI accelerators demanding more electricity than ever per rack, traditional low-voltage alternating current (AC) distribution is becoming increasingly difficult to scale. The three tech giants believe that higher-voltage DC can deliver the same amount of power at lower current, which reduces the amount of copper and conductor infrastructure required. The industry is converging on 800VDC because it can move more power while simultaneously reducing the conductor burden associated with lower-voltage systems.
The Open Compute Project, originally launched by Facebook in 2011 to share data-center hardware designs, has grown into one of the most influential forums for data-center innovation. OCP's open specifications have already transformed server, storage, and rack designs, and now the organization is taking on a more foundational challenge: the electrical architecture of AI factories. By aligning on a common specification, hyperscalers and equipment manufacturers can avoid proprietary, siloed solutions and instead build to a shared standard, much as they do with servers today.
The technical case for 800VDC
Traditional data centers distribute power in alternating current at 480V or 415V, then step it down through several conversion stages to the low-voltage DC used by servers, typically 12V or 48V. Each conversion step adds losses in the form of heat. At the same time, the growth of AI has created racks that consume 100kW or more. Nvidia's GB200 NVL72 architecture, for example, draws enormous amounts of power in a single rack, and future systems from the company, such as Vera Rubin, are expected to push those figures even higher.
With 800VDC, data centers can carry power at a higher voltage, which means lower current for the same power level. Since current is what generates heat through resistive losses (I²R losses), lower current translates directly to cooler operation. This also allows for longer cable runs and more flexible rack layouts. The savings over conventional AC distribution are considerable. With AC there are four wires, while DC requires only two, producing significant wiring savings in an all-DC facility. Fewer wires, lower heat generation, and reduced conversion losses all add up to a more efficient and less complex data center.
According to OCP analysis, adopting 800VDC can reduce copper usage by 50% to 80%. That is not a marginal improvement; for a one-gigawatt data center, it means saving several million pounds of copper wire. In addition to material savings, operators can see an 8% to 12% reduction in annual energy-related operating expenses through lower conversion and distribution losses. For AI-first facilities, the upfront capital expenditure savings can be substantial: anywhere from $4 million to $8 million per 10 MW build by reducing upstream AC infrastructure. These figures are beginning to attract attention from data-center operators beyond the three founding companies.
A collaborative push through OCP
Google, Microsoft, and Nvidia began presenting their 800VDC work through OCP in 2025. What started as a demonstration of technical feasibility has evolved into an effort to establish specifications that can be adopted throughout the data-center supply chain. The companies have initiated an OCP workstream focused on converting medium-voltage AC power directly to 800VDC, eliminating intermediate transformation stages. They are also aligning requirements covering power quality, power smoothing, and end-to-end system interfaces, ensuring that different vendors' equipment can interoperate seamlessly.
The open approach could have significant implications for data-center power equipment manufacturers. Instead of designing separate power systems for each major customer, manufacturers could build equipment using a common specification. This would lower engineering costs, accelerate time to market, and enable economies of scale in production. It would also give smaller data-center operators access to cutting-edge technology that has historically been reserved for hyperscalers, because standardized components are more widely available and easier to deploy.
AWS is notably absent
One major player has not joined the 800VDC effort: Amazon Web Services (AWS). AWS has made no commitment toward 800VDC, nor has it announced its own alternative power distribution standard. However, the company is clearly feeling the pressure of power density increases. AWS has been aggressively redesigning its AI data centers for much higher-density systems through an internal initiative code-named “Titus.” The project reportedly focuses on next-generation infrastructure capable of supporting NVIDIA GB200-class systems and future Vera Rubin hardware, while increasing data center capacity and improving power and cooling efficiency.
Industry observers speculate that AWS's silence on 800VDC is strategic. The company has a history of pursuing its own components and standards, from custom silicon like Graviton to its own networking hardware. It is possible that AWS will eventually endorse the OCP effort, but it may also be developing a proprietary power architecture. With power density increasing and electricity usage becoming untenable under current distribution models, AWS will have to make a move. Whether that move is toward the open standard or a divergent route remains to be seen.
Broader implications for AI infrastructure
The push for 800VDC reflects a fundamental change in data-center power requirements. AI accelerators are being deployed in increasingly dense configurations, driving power consumption per rack higher and making traditional low-voltage AC distribution more difficult to scale. The need for higher-voltage DC is not entirely new; earlier initiatives explored 380VDC, but those efforts did not achieve widespread adoption. The difference now is that AI has created a compelling and urgent use case, and three of the largest players in the industry are aligned on a single target voltage.
The benefits extend beyond the data center's four walls. Reducing copper usage has economic and environmental implications. Copper is expensive, and mining it has a significant carbon footprint. Saving pounds of copper across a gigawatt facility reduces material costs and extraction impacts. Lower energy losses also mean lower power plant emissions, a key consideration as data centers face heightened scrutiny over their environmental footprint. In a reality where AI growth is accelerating, any increase in efficiency is welcome.
For power equipment manufacturers, the move toward 800VDC creates a clear roadmap for product development. Power supplies, bus bars, switchgear, connectors, and cables can all be designed to a common voltage standard, allowing for better integration and reliability. The OCP workstream is also addressing safety standards, as 800VDC carries different risks than traditional AC distribution. Arc flash hazards, contact protection, and grounding all require new considerations. Defining these requirements openly helps the entire industry build safer, more reliable systems.
Google, Microsoft, and Nvidia have framed their effort as a way to create a common electrical foundation for AI factories before increasingly dense systems force every hyperscaler and equipment supplier to develop its own isolated solution. That kind of fragmentation would raise costs, slow innovation, and make it harder for the industry to move forward collectively. By rallying around an open standard, they are setting the stage for a more sustainable and scalable AI infrastructure.
The technical work is already underway, and OCP expects to publish formal specifications that data-center operators and equipment vendors can adopt in the next phase. As AI workloads continue to grow and new accelerator generations arrive, the demand for higher voltage and more efficient power distribution will only intensify. The 800VDC standard, if widely adopted, could become as foundational to AI data centers as the server chassis or the network switch. It remains to be seen whether AWS will eventually join, but the momentum is clear: the industry is moving toward higher-voltage DC power.
Source: Network World News