🔍 Read the full analysis: What Is Driving The Move To 800-Volt DC In AI Data Centers? on ThorstenMeyerAI.com
Get smart everyday buys delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
AI data centers are exploring 800-volt direct-current power as rack demand rises and conventional power distribution becomes harder to scale. Vantage reserved more than a gigawatt of factory capacity for Heron Power transformers, while NiSource and Microsoft agreed to a 20-year power deal with generation and battery capacity. Efficiency gains cited by equipment vendors are not independent measurements, and grid access remains a major constraint.
AI data-center power infrastructure is drawing new investment as operators contend with rising rack demand: Vantage reserved more than one gigawatt of factory capacity for Heron Power solid-state transformers designed to convert grid electricity to 800-volt direct current, and NiSource signed a 20-year power agreement with Microsoft that includes about 1,100 megawatts of generation and a 600-megawatt battery. The deals show companies planning around power supply and conversion alongside computing capacity, though they do not establish that 800-volt systems will deliver the efficiency gains vendors project.
The Vantage reservation, reported as made on October 21, is for factory capacity for Heron Power equipment, not a statement that a gigawatt of transformers has already been installed or is operating. The solid-state transformer is intended to convert grid power directly into 800-volt DC. The source material does not specify the deployment schedule or how much of Vantage’s planned capacity the equipment will serve.
Five days earlier, NiSource and Microsoft signed a 20-year power deal. The agreement adds about 1,100 megawatts of generation and a 600-megawatt battery, according to the source report. It is a separate development from Vantage’s equipment reservation, but both reflect the scale of electricity infrastructure now being planned around data-center demand.
The engineering case for higher-voltage DC rests partly on reducing the number of power conversions and the amount of copper needed to deliver electricity at very high rack loads. NVIDIA’s reference design would convert 13.8-kilovolt grid power to 800-volt DC near the edge of a facility, then distribute DC to racks. NVIDIA says its design could use 45% less copper than 415-volt AC distribution and improve end-to-end efficiency by up to 5% compared with current 54-volt systems. Those are NVIDIA’s vendor estimates; the material does not provide independent measurements validating them.
Infographic 01 · From grid to rack
AI racks near a megawatt are outgrowing the AC power chain.
Why the industry is moving to 800-volt direct current: fewer conversions, less copper, batteries on the main path. The numbers are vendor figures with different baselines, so they cannot be added up.
Today: many conversions
Typical AC chain from the grid to the rack
- Medium-voltage AC from the grid
- Transformer steps it down to low-voltage AC
- UPS converts to DC for the batteries and back to AC
- Distribution units feed the racks
- Every server power supply converts AC to DC again
Each stage loses energy, and almost all of the loss becomes heat the cooling system must remove.
800 V DC: convert once, outside
NVIDIA reference design, full-scale production planned with Kyber racks in 2027
- 13.8 kV AC from the grid
- Converted once to 800 V DC at the edge of the site, batteries on the same DC bus
- 800 V DC busway to the racks
- Stepped down inside the rack, for 1 MW racks and beyond
Bridge for existing halls: a “sidecar” cabinet next to the rack converts AC to 800 V DC, while the old chain stays in place.
Power lost per MW delivered to the rack
kW, SMA white-paper model vendor model
What NVIDIA says 800 V DC changes
NVIDIA’s engineering claims claim
Sources: NVIDIA technical blog (20 May 2025, updated 31 Jul 2025); SMA America, sponsored content in Utility Dive (26 Oct 2026). Rack density: IEA, Key Questions on Energy and AI (Apr 2026) (AI server power density 11× from 2020 to 2025, a further 4× by 2027).
Powered by Thorsten Meyer AI
Power Limits Are Reshaping AI Campuses
The shift matters because the power demand of AI racks is growing faster than the infrastructure designs that served conventional server halls. The International Energy Agency (IEA) says AI-server power density rose 11-fold between 2020 and 2025 and is expected to quadruple again by 2027. It estimates a rack could then reach peak demand comparable to 65 households. That is a projection about peak load, not a measure of a rack’s continuous electricity use.
Higher-voltage distribution could reduce the burden of moving large amounts of power through a building. A DC architecture may also make it easier to connect batteries, which store electricity as DC, to the main power path. But the change does not generate electricity or create new grid connections. The IEA says transformers can take two to three years to arrive, while grid connections take five to ten years in many places. For operators, timely access to power may be as decisive as the choice of chips or cooling equipment.
Costs also extend beyond data-center operators. The source material cites a U.S. average household electricity price of 18.21 cents per kilowatt-hour this year, compared with 17.30 cents in 2025, and a 41% rise in wholesale prices in the PJM region. Those figures do not establish that data centers caused the increases. Who pays for new generation and transmission depends on contracts, regulation and local decisions, and can vary from one project to another.
Popular right now
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
Samsung 42-Inch Class OLED S90F 4K Smart TV (2025)
As an affiliate, we earn on qualifying purchases.
From Server Conversions to 800 Volts
Traditional data centers receive alternating current, step it down through transformers and power equipment, and convert it to low-voltage direct current inside servers. Each conversion loses some energy as heat, which cooling systems must then remove. That architecture developed when rack power was much lower; designs intended for racks drawing tens of kilowatts face different constraints as AI systems approach much higher loads.
NVIDIA’s published reference design describes a route from 13.8-kilovolt AC at the grid connection to 800-volt DC distribution. The company says a one-megawatt rack supplied at 54 volts could require as much as 200 kilograms of copper busbar, while power shelves could occupy substantial rack space. NVIDIA plans full-scale production with its Kyber systems in 2027. These statements describe the company’s design and projections, not a broad record of deployed systems.
A Utility Dive article published October 26 was sponsored by inverter maker SMA America, so its modeled savings should be read as a vendor’s case rather than an independent field result. SMA’s white-paper model estimates losses of about 80 kilowatts when delivering one megawatt through a traditional AC chain, versus about 20 kilowatts with an 800-volt DC arrangement. SMA also describes a “sidecar” conversion cabinet that could let new DC racks operate in existing AC halls while much of the older infrastructure remains.
Batteries are part of the same discussion because AI workloads can change their power draw quickly. The IEA estimates data centers could have 20 to 25 gigawatts of batteries by 2030 and says storage could serve the grid if incentives support that role. The estimate is a projection, not current installed capacity.
“Its 800-volt design could use 45% less copper than 415-volt AC distribution and provide up to 5% better end-to-end efficiency than current 54-volt systems.”
— NVIDIA
Deployment and Savings Remain Unproven
The Vantage announcement concerns reserved factory capacity; the source material does not confirm installation dates, operating sites or how many transformers will be delivered. It also does not explain the commercial terms or the specific allocation of equipment. For the NiSource-Microsoft agreement, further detail is not provided on the generation sources, project schedule, battery duration, or how the contract’s costs and benefits are distributed.
Performance figures need similar care. NVIDIA’s copper, efficiency and maintenance estimates, and SMA’s modeled power-loss comparison, are vendor claims or model results. Independent, comparable measurements from operating 800-volt facilities are not provided. It is also unclear how quickly operators will adopt full DC distribution rather than transitional sidecar systems, or how much existing equipment can be reused.
Broader electricity-price figures do not show how much data-center growth contributes to household bills in any particular area. The effect depends on local grid capacity, new generation, transmission investment, utility rules and customer contracts.
Watch Grid Deals and 2027 Systems
Near-term indicators include whether Heron Power’s reserved factory capacity turns into equipment deliveries and when Vantage expects to deploy it. NVIDIA’s stated 2027 Kyber production plan offers another milestone for whether 800-volt distribution moves from reference design toward commercial deployment. The source material gives no confirmed timetable for a wider industry rollout.
For the NiSource-Microsoft deal, subsequent project disclosures should clarify the generation buildout, battery operation and connection to the grid. Utilities and regulators will also shape how costs for new plants and wires are allocated. Until more operating data and project details are available, the central questions are whether 800-volt systems perform as vendors claim and whether new power infrastructure can be delivered in time to meet data-center demand.
Key Questions
Why are AI data centers considering 800-volt DC?
Higher-voltage distribution can move large amounts of power with less current than low-voltage systems, potentially reducing copper needs and conversion losses. The claimed savings are based on vendor designs and models, not independent results supplied here.
What did Vantage reserve from Heron Power?
Vantage reserved more than one gigawatt of factory capacity for Heron Power solid-state transformers intended to convert grid power directly to 800-volt DC. A factory-capacity reservation is not confirmation that this amount of equipment has been installed.
What is included in the NiSource-Microsoft agreement?
The 20-year agreement includes about 1,100 megawatts of generation and a 600-megawatt battery, according to the source material. Details such as the generation sources and project schedule are not specified there.
Are the projected efficiency gains independently verified?
No independent validation is cited for the figures in the source material. NVIDIA’s estimates and SMA’s white-paper model should be treated as vendor claims or modeled results rather than measurements from a broad fleet of operating facilities.
Will 800-volt DC solve data centers’ power shortages?
No. It may change how electricity is distributed and how batteries are integrated, but it does not create generation or grid capacity. Transformer lead times and access to grid connections remain constraints, and the source material says many connections can take five to ten years.
Source: ThorstenMeyerAI.com
