🔍 Read the full analysis: AI Data Centers Are Changing Their Power Systems. Here’s Why 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 center operators are changing how power reaches high-density racks, with a Vantage transformer reservation and a Microsoft-NiSource supply agreement illustrating two parts of that shift. The moves reflect rising rack loads and long waits for grid connections; efficiency and cost figures cited for new designs are largely vendor claims, and household impacts vary.
AI data center operators are changing how electricity is delivered and buffered as server racks demand far more power: Vantage reserved more than one gigawatt of factory capacity for a solid-state transformer, while NiSource and Microsoft signed a long-term deal involving new generation and a large battery. The developments point to a shift toward 800-volt direct current (DC) systems and storage, a topic explored in questions about U.S. data center capacity, though they do not establish that the technology will cut costs or electricity prices.
Vantage reserved the factory capacity on October 21 for a solid-state transformer made by Heron Power. The source says the equipment is designed to convert grid power directly into 800-volt DC. Heron’s system also includes storage intended to buffer changes in data center demand. The reservation concerns manufacturing capacity; the available information does not say how many units will be delivered, where they will be installed or when they will enter service.
On October 16, NiSource signed a 20-year power agreement with Microsoft that adds about 1,100 megawatts of generation and a 600-megawatt battery, according to the source. The agreement is a separate example of efforts to secure power and storage for data center needs amid growing grid constraints. The source does not specify the generation technology, project locations, or how the costs will be divided among the companies and other electricity customers.
The engineering case for DC is tied to the number of conversion steps in conventional data centers. Grid electricity arrives as alternating current (AC), is stepped down and converted through facility equipment, then converted again for servers. Each conversion loses some energy. NVIDIA’s design proposal would convert grid power to 800-volt DC near the facility perimeter and distribute it to racks. The company says this 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 estimates, not independently verified results from operating sites.
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 Delivery Becomes a Capacity Constraint
For data center builders, the issue is no longer only how many chips they can buy. They also need equipment capable of supplying dense, rapidly changing loads, plus a grid connection that can deliver power when a facility is ready. The International Energy Agency (IEA) says AI server power density rose 11-fold from 2020 to 2025 and is projected to quadruple again by 2027. It says a rack could then draw as much power at peak as 65 households.
That projection helps explain interest in higher-voltage distribution: delivering the same amount of power at a higher voltage can reduce current and the amount of conductor needed. NVIDIA says a one-megawatt rack supplied at 54 volts could require up to 200 kilograms of copper busbar. But changing a power system involves more than replacing a component. Operators must account for existing buildings, safety and reliability requirements, equipment availability, and how new systems work alongside installed AC infrastructure.
Storage could help manage short-lived load swings and grid disturbances, as well as support a more controlled return to full demand after an interruption. The IEA estimates data centers could have 20–25 gigawatts of batteries by 2030, and says storage could become a grid asset if incentives are suitable. That is a projection, not a confirmed deployment total, and its benefits depend on how batteries are operated and paid for.
The stakes reach beyond operators. The IEA projects global data center electricity use will rise from 485 terawatt-hours in 2025 to 950 terawatt-hours in 2030. Berkeley Lab estimates U.S. data centers used 4.4% of national electricity in 2023 and could account for 6.7% to 12% by 2028. These estimates show why power projects and cost allocation are receiving attention, but they do not show that a specific data center caused a particular customer’s bill increase.
Products Worth Considering
HP Sprocket Panorama Portable Color Photo and Label Printer
FLEXISPOT 42-Inch Electric Height-Adjustable Desk Converter with Keyboard Tray
Apple iPad Air 13-inch (M4), 128GB, Wi-Fi 7, Space Gray
As an affiliate, we earn on qualifying purchases.
From AC Server Rooms to Dense AI Racks
Conventional data centers have used AC distribution for decades. Electricity passes through transformers, uninterruptible power supplies and distribution equipment before being converted to the lower-voltage DC used by servers. At older rack loads of roughly 10 to 20 kilowatts, that arrangement could serve the equipment without taking up an excessive share of rack space. Much higher rack demand changes the physical and operational requirements.
NVIDIA has proposed an 800-volt DC architecture for its future Kyber systems, with full-scale production planned for 2027. The company says its design would move more power through conductors of a given size and reduce maintenance costs. An October 26 Utility Dive article, sponsored by inverter maker SMA America, describes another possible route: a “sidecar” conversion cabinet that lets 800-volt racks operate in rooms built around AC systems. The article’s model estimates losses of about 80 kilowatts when delivering one megawatt through a traditional AC chain, versus about 20 kilowatts in its proposed DC arrangement. Those figures come from a vendor-sponsored model, not a reported independent field test.
Physical equipment and grid access also affect project schedules. The IEA says transformers can take two to three years to arrive and grid connections can take five to ten years in many locations. Berkeley Lab reported about 2,300 gigawatts of proposed generation and storage capacity in U.S. interconnection queues at the end of 2024; only 13% of earlier requests were ultimately built. Queue capacity is not the same as power that will be constructed or delivered.
““Leaving no room for compute.””
— NVIDIA, describing its proposed 800-volt design
Deployment, Costs and Bill Effects
The announced activity does not establish how quickly 800-volt systems will be installed or how much of the new factory capacity Vantage will use. The source also gives no project-level schedule or detailed cost terms for the NiSource-Microsoft agreement. It remains unclear how much of the planned generation and battery capacity will serve data center load at particular times, or how costs and benefits will be shared.
Efficiency comparisons are not all based on independent operating measurements. NVIDIA’s savings figures are company claims, and SMA’s loss estimates appear in sponsored industry content and are model results. Actual performance will depend on the site, equipment, load profile and system design. The source reports average U.S. household electricity prices of 18.21 cents per kilowatt-hour in 2026, compared with 17.30 cents in 2025, and a 41% increase in wholesale prices in the PJM region. Those figures do not by themselves establish that data centers caused the changes; the source does not provide a causal breakdown or explain how costs vary among customers and regions.
Factory Capacity and Grid Projects
The next indicators will be whether Vantage’s reservation leads to delivered transformers and operating installations, and whether the NiSource-Microsoft agreement moves from a signed contract to built generation and battery capacity. The source gives no construction timetable for those projects. NVIDIA’s stated target is full-scale production of its Kyber systems in 2027; performance in deployed facilities will be needed to test its efficiency and copper-reduction claims.
Operators and utilities will also need to address connection queues, equipment lead times and the terms under which storage can support the wider grid. The figures available so far describe a fast-growing demand problem and proposed responses, not a settled outcome on how much the new systems will save or who will pay for power infrastructure.
Key Questions
Why are AI data centers considering 800-volt DC power?
Higher rack loads make power delivery more demanding. Distributing at 800 volts DC could reduce current and conductor requirements and avoid some conversion steps, though claimed efficiency gains still need to be tested in operating systems.
What did Vantage reserve from Heron Power?
Vantage reserved more than one gigawatt of factory capacity for Heron Power solid-state transformers designed to convert grid electricity to 800-volt DC. The source does not state delivery dates or how many units will be installed.
What is included in the NiSource-Microsoft agreement?
The 20-year agreement adds about 1,100 megawatts of generation and a 600-megawatt battery, according to the source. It does not provide details on generation technology, project locations or cost allocation.
Are the reported efficiency savings independently verified?
No independent operating measurements are provided in the source. NVIDIA’s copper and efficiency figures are vendor claims, while SMA’s AC-versus-DC loss comparison is a vendor-sponsored model.
Does this mean data centers are causing household electricity bills to rise?
The source reports national and regional electricity-price figures but does not establish that data centers caused them. The effect on bills depends on local power markets, infrastructure costs and how projects’ expenses are allocated.
Source: ThorstenMeyerAI.com
