Andreas Fornwald

Great to be back, Scott. This is a topic I'm really passionate about because it sits at the intersection of power electronics, grid infrastructure, and the AI compute buildout — and most people in the energy industry haven't fully grasped what's happening here.

Traditional data centers take utility AC power — typically at medium voltage — step it down through transformers, convert it to DC for the servers, and distribute it through an elaborate chain of UPS systems, PDUs, and rectifiers. Every conversion step loses energy. A modern hyperscale data center running traditional AC architecture might have an end-to-end power conversion efficiency of 85 to 90 percent. That sounds good until you realize you're wasting 10 to 15 percent of every megawatt you buy. At 500 megawatts, that's 50 to 75 megawatts of pure waste — enough to power a small city.

Exactly. With a high-voltage DC distribution architecture, you convert AC to DC once at the point of grid interconnection — at high voltage, which minimizes transmission losses — and then distribute DC directly to the server racks. The GPU and CPU chips in AI servers are fundamentally DC devices. They don't want AC. Every AC-to-DC conversion in the traditional chain is an unnecessary step that was only there because the grid delivers AC. If you can eliminate those intermediate conversions, you can push end-to-end efficiency above 97 percent. That's a massive improvement.

800V is emerging as the sweet spot for several reasons. Higher voltage means lower current for the same power level — and lower current means smaller conductors, less resistive loss, and lower infrastructure cost. But you also have to manage insulation requirements, arc flash hazards, and component availability. The EV industry has been a major driver here — Tesla, Porsche, Hyundai, and others moved to 800V architectures for exactly the same reasons. That created a supply chain for 800V power electronics — inverters, converters, busbars, connectors — that data center operators can now leverage. Star Charge has deep expertise in 800V systems from our EV charging work, and we're seeing direct technology transfer into the data center market.

It changes the internal architecture significantly, but the grid interconnection point is still AC — utilities deliver AC and that's not changing anytime soon. What changes is the size and sophistication of the AC-to-DC conversion system at the point of interconnection. You're essentially building a large rectifier station — similar to what you see in HVDC transmission systems — right at the data center. That rectifier station becomes a critical piece of infrastructure. It also creates interesting opportunities for grid services: a large, controllable DC load can provide demand response, frequency regulation, and voltage support to the grid in ways that traditional data center loads cannot.

This is where it gets really interesting. Battery cells are DC devices — lithium-ion chemistry operates at DC voltages. In a traditional AC data center, integrating battery storage requires additional conversion equipment to interface the DC battery with the AC distribution system. In an 800V DC architecture, you can connect battery storage directly to the DC bus. No additional conversion. The battery charges and discharges at native DC voltage, which improves round-trip efficiency and reduces capital cost. You also get seamless, millisecond-level switching between grid power and battery power — no transfer time, no UPS bypass. For AI training workloads where even a brief interruption can corrupt a multi-day training run, that seamless resilience is enormously valuable.

The leading hyperscalers are actively piloting and deploying 800V DC architectures in new builds right now. This isn't a five-year roadmap — it's happening today. The economics are compelling: lower operating costs from efficiency gains, lower capital costs from simplified power distribution, and better resilience. The constraint is the supply chain for high-voltage DC components at data center scale, and the engineering expertise to design and operate these systems safely. That's exactly the gap that Star Charge is positioned to fill — we've been building 800V systems for EV charging at scale, and that expertise translates directly.

Thank you, Scott. The power architecture of AI data centers is going to be one of the defining infrastructure stories of this decade. I'm glad we're getting this conversation out there.