🔍 Read the full analysis: Inside The Move To 800-Volt DC And Batteries For AI Data Centers on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
AI data-center power demand is prompting investment in 800-volt direct-current systems, grid generation and batteries. Vantage reserved more than 1 gigawatt of factory capacity for Heron Power solid-state transformers, while NiSource and Microsoft agreed to a 20-year power deal that includes about 1,100 megawatts of generation and a 600-megawatt battery. The deals show growing efforts to address power delivery and grid constraints, but do not establish independent performance results for the new equipment.
Data-center power infrastructure is drawing new investment as AI racks demand more electricity: Vantage reserved more than 1 gigawatt of factory capacity for Heron Power solid-state transformers designed to deliver 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 point to growing efforts to supply and manage electricity for AI facilities, though vendor performance claims and the projects’ precise effects on local bills remain unresolved, amid mounting power bottlenecks.
Vantage made the reservation on October 21, according to the source report. The equipment is intended to convert grid power into 800-volt DC, a design aimed at reducing the number of conversion steps between the grid and high-demand computing racks. The reservation is for factory capacity; the available information does not say that more than a gigawatt of transformers has already been delivered or installed.
Five days earlier, NiSource and Microsoft signed a 20-year agreement tied to roughly 1,100 MW of generation and a 600 MW battery. The source describes the deal as adding generation and storage, but does not provide a project-by-project construction schedule, the generation technology, or a breakdown of how costs and power will be allocated.
The shift is being driven by rising rack density and the difficulty of connecting large new loads to the grid, a challenge explored in more detail in this look at AI data-center power constraints. NVIDIA’s 800-volt reference design calls for converting 13.8-kilovolt grid power to DC at a facility’s perimeter, then distributing it to racks. NVIDIA says the 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 figures are company estimates, not independent measurements; full-scale production with its Kyber systems is planned for 2027.
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 Is Becoming a Constraint
For data-center developers, the issue is no longer only how many chips they can obtain. They also need equipment and grid connections capable of supplying dense, fast-changing loads. The International Energy Agency (IEA) estimates global data-center electricity use could rise from 485 terawatt-hours in 2025 to 950 TWh in 2030. That forecast underscores why operators are pursuing power-system changes alongside new computing hardware.
Higher-voltage DC can reduce current for a given amount of power, potentially cutting the amount of conductor needed and reducing some conversion losses. Batteries could help manage abrupt changes in demand and provide backup or short-term support. But these approaches do not create new electricity or remove the need for grid upgrades. Their value depends on engineering performance, costs, power availability and how facilities are operated.
The stakes extend beyond operators. New generation and transmission infrastructure can affect communities and electricity customers, but the source material does not establish how the NiSource-Microsoft agreement’s costs will be assigned or whether it will raise or lower household bills. Those questions depend on contract terms and regulatory decisions, not just the capacity figures announced.
Products Worth Considering
Apple iPad Air 13-inch (M4), 128GB, Wi-Fi 7, Space Gray
Canon SELPHY CP1500 Wireless Compact Photo Printer with RP-108 Ink and Paper Set
Dell XPS 13 9350 Laptop, 13.4 FHD+ Display, Intel Ultra 7 256V, 16GB DDR5, 1TB SSD
As an affiliate, we earn on qualifying purchases.
From AC Server Rooms to Dense AI Racks
Traditional data centers generally receive alternating current and convert it multiple times before servers use low-voltage DC. Each conversion loses some energy, much of which becomes heat. Designs built for racks drawing tens of kilowatts face different demands from AI systems approaching megawatt-scale rack loads.
The IEA’s April 2026 report said AI-server power density rose 11-fold between 2020 and 2025 and projected another fourfold increase by 2027. It said a rack could then have peak demand comparable to 65 households. These are report figures and projections, not measurements of every facility or rack.
Grid access is another bottleneck. The source cites IEA estimates that transformers can take two to three years to arrive and that grid connections can take five to ten years in many places. Berkeley Lab reported about 2,300 gigawatts of generation and storage capacity in U.S. interconnection queues at the end of 2024, while only 13% of earlier requests were ultimately built. Queue capacity represents proposed projects, not power already available to data centers.
A separate Utility Dive article dated October 26 was sponsored by inverter maker SMA America. Its modeled example estimated losses of about 80 kW when delivering 1 MW through a traditional AC chain, versus about 20 kW in an arrangement converting to 800-volt DC outside the building. Because this is a vendor-sponsored model, it should not be treated as an independent field result.
“Leaving no room for compute”
— NVIDIA, in its 800 VDC technical blog
Performance, Costs and Buildout Remain Open
The reported agreements do not establish when the reserved transformer capacity will become equipment in operation, how many sites will use it, or whether the projects will meet their planned schedules. The source also gives no independently measured efficiency or maintenance results for an operating 800-volt DC data center.
Several widely cited benefits are forecasts or vendor claims. NVIDIA’s copper, efficiency and maintenance estimates, and SMA’s modeled conversion-loss comparison, have not been presented here as results from independent testing. The IEA’s estimate that data centers could have 20–25 gigawatts of batteries by 2030 is a projection, not a confirmed buildout total.
It is also unclear from the reported information how the NiSource-Microsoft deal allocates costs, whether its generation and battery capacity will serve a particular data-center site, or how regulators and local utilities will account for the infrastructure. U.S. household electricity averaged 18.21 cents per kilowatt-hour in 2026 through the source’s reporting date, compared with 17.30 cents in 2025, according to the EIA. That national price movement does not by itself show that data centers caused the increase. Regional impacts and the role of individual projects vary.
Delivery and Deployment Are Next
The next milestones are delivery of the reserved transformer capacity, construction and commissioning of the generation and battery tied to NiSource and Microsoft, and further detail on the projects’ schedules and commercial terms. The source says NVIDIA plans full-scale production of its 800-volt design with Kyber systems in 2027; deployment and real-world performance will show whether its projected savings are achieved.
For operators, utilities and regulators, the practical test will be whether the new systems can connect on time, support fluctuating AI loads and reduce costs without shifting unexamined expenses to other electricity customers. Publicly reported performance data, project filings and utility rate decisions should provide clearer evidence as the buildout proceeds.
Key Questions
What happened in the Vantage deal?
Vantage reserved more than 1 gigawatt of factory capacity for Heron Power solid-state transformers designed to convert grid electricity to 800-volt DC. The report does not say that this capacity has already been delivered or installed.
What is included in the NiSource-Microsoft agreement?
The companies signed a 20-year power agreement that the source says adds about 1,100 MW of generation and a 600 MW battery. The available details do not identify the generation technology or explain how project costs will be allocated.
Why are data centers considering 800-volt DC?
As AI racks draw more power, operators are evaluating higher-voltage distribution to reduce current, conductor needs and some conversion losses. Specific savings cited by NVIDIA are vendor estimates, not independent measurements.
Will these projects lower electricity bills?
The reported deals do not establish whether household bills will rise or fall. The effect depends on project costs, contracts, grid upgrades and regulatory decisions, and a national price change alone does not establish the cause.
When could NVIDIA’s 800-volt design enter production?
NVIDIA says full-scale production with its Kyber systems is planned for 2027. That is a company plan; the source does not report completed deployments or independent operating results.
Source: ThorstenMeyerAI.com
