You can’t hide from 800VDC. It’s coming, whether via ±400 VDC or 0-800 VDC. The efficiency gains, infrastructure benefits, and the potential for global standardization make it hard to ignore. My long-term view is that it will likely become the dominant standard. Yet while voltage standards may eventually converge, the architecture won’t – not yet. We’re going to see multiple approaches before the industry settles on a common design.

Why AC has hit its limits

For years, Moore’s Law kept us comfortable. We could scale performance without exceeding traditional power constraints. That ended around 2015. Today, the power envelope is climbing fast. Compute hardware is so densely packed it seems like it has been vacuum-sealed into the rack. And that density drives power demand into territory where AC doesn’t scale. Beyond 400 kW per rack, AC distribution requires prohibitively costly and impractical amounts of copper cabling, oversized busbars, and incredibly high current levels. That’s why DC is on everyone’s roadmap.

Two camps, two strategies delivered via power sidecars

Right now, I see two camps forming. One is pushing straight to 800VDC – with Nvidia taking the lead – because they need megawatt racks for AI workloads and want to strip out conversion stages. The other camp, led by several hyperscalers, is leaning toward ±400VDC as a transitional step.

By complying with existing standards and leveraging more common components, ±400VDC is seen as less risky. That means easier sourcing, lower upfront design risk, and a power envelope that’s more familiar. It also reduces the arc flash hazard compared to 800VDC, simplifying training and compliance. For brownfield sites, ±400VDC can often be integrated without a full redesign of the facility’s electrical infrastructure, which is a major advantage in time to deployment.

Both approaches make sense; ±400VDC is a bridge for brownfield sites while 800VDC is the endgame for efficiency and scale.

Challenges ahead

At these voltages, we’re talking about potentially lethal power levels. That means new standards, new risk management plans, and a workforce trained to handle HVDC safely. Safety could also slow adoption because operators will need labs, training programs, and validated procedures before they flip the switch.

Then there’s the technology gap. Solid state transformers (SSTs) are critical for making 800VDC practical. They allow us to go from medium-voltage AC to HVDC in a compact, efficient way. We’ve seen prototypes at OCP, but cost and maturity are still hurdles. Until SSTs scale, sidecar solutions – accessory power racks that centralize AC-to-DC conversion – will fill the gap, especially for brownfield deployments that can manage power supply and cooling infrastructure requirements. Greenfield sites may wait for SSTs to avoid unnecessary complexity.

Hybrid architectures will also play a role. I expect to see 800VDC delivered to the cabinet, then stepped down to 48 to 50 volts inside the rack. That lets us leverage existing point-of-load ecosystems while gaining the distribution benefits of HVDC.

Beyond distribution: Power sourcing

Power distribution is only half the story. Many new multi-megawatt sites are asked to “bring their own power” by deploying natural gas turbines or renewable microgrids to avoid long utility interconnection delays. These sources often produce DC power natively, which means operators can feed HVDC directly into the data center, skipping conversion stages for higher efficiency and lower losses.

Will everyone adopt 800VDC? It depends on the workload. AI factories will move first because, at that scale, efficiency is a competitive advantage. Enterprise workloads may stick with ±400VDC or even 48V for a longer period. We may even see Nvidia’s competitors design chips that meet customer needs at lower voltages.

What’s still undecided

Three things remain to be determined:

• How fast SSTs mature and hit cost targets.
• How quickly standards bodies define protocols for HVDC.
• Whether operators take a direct leap to 800VDC or move through ±400VDC first.

The bottom line: physics is pushing us toward HVDC, but the journey won’t be one-size-fits-all. The advice I can offer is to design for migration, with standardized interfaces, modular power systems, and components that can adapt to either AC or DC topologies.

And just as important – stay engaged with the standards work happening now. IEEE, ASHRAE, and OCP are actively shaping the safety envelopes, connector designs, and interoperability guidelines that will define this transition. If you have the expertise, join the conversation. The industry needs collaboration to make this shift safe, efficient, and sustainable.

The transition to HVDC will require ongoing collaboration between operators, standards bodies, and technology providers. For those exploring practical solutions to address the evolving challenges of deploying resilient, agile AI data centres, further resources are available here.