Subscribe
18:00Tao calls OpenAI’s Navier–Stokes push “resource extraction”17:10LAPTOP memecoin hits $190.81, then loses 99% inside an hour17:05Hubinger puts the odds of AI killing everyone above 10%; a colleague resigns16:39CancerBench launches; five frontier models tied at zero cancer types cured16:30ElevenLabs preparing 2028 IPO after $11bn round, The Information reports16:30Anthropic retracts its July explanation: Mythos 5 attacked systems knowingly
Hardware3 min read

Nvidia's 800-volt data centre now starts with one rack inside a building you already own

The 2025 pitch was converting grid power to 800 VDC at the perimeter. The 2026 product slots into existing AC infrastructure and changes nothing about the building.

In briefNVIDIA's 800 VDC rollout is phased and operates alongside existing AC distribution, with a power rack first, then a row-level power center, then a hall-level DC power block1NVIDIA's MGX-compatible 800 VDC power rack arrives in the second half of 2026 and slots into existing AC infrastructure with no changes to the building's electrical system2The row power center supports up to 2 megawatts per row over an overhead 800 VDC busway, with availability expected in 20273
Copper busbars inside an electrical cabinet
Photo: ToT89 (CC BY-SA 4.0)

Nvidia's 800 VDC architecture is being rolled out in three steps that run alongside existing alternating-current plant rather than replacing it, with the adoption level decided facility by facility. SemiAnalysis laid out the sequence on September 9th: a power rack that delivers 800 volts DC into the compute racks and standardises the DC interface and backup configuration; then a row-level power center handling centralised rectification and feeding a high-density row over a busway; then, eventually, a hall-level DC power block.

SemiAnalysis@SemiAnalysis_

NVIDIA’s 800 VDC plan includes a phased rollout that operates alongside existing AC distribution systems. The adoption level depends on each facility individually, rather than replacing an entire plant all at once.

In the short term, NVIDIA’s Power Rack delivers 800V DC into the compute racks and standardizes the DC interface and backup configuration so it can fit existing AC infrastructure. Following this, a row-level Power Center manages centralized rectification and supplies a full high-density row via a busway, reducing cabling complexity in white space. Over the longer term, a hall-level DC Power Block becomes the standard power building block, supporting higher-density loads.

NVIDIA will collaborate with industry partners to expand demonstration and pilot sites ahead of large-scale deployment.

on X · 24.4K views · captured Sep 10, 2026

Read that against what Nvidia was saying in May 2025 and the change of register is the story. Back then the architecture was a redesign from the grid inward — 13.8 kV AC converted straight to 800 VDC at the data centre perimeter by industrial rectifiers, eliminating most intermediate conversion steps. Now the first product is a rack that, in Nvidia's own words, is "designed to slot into existing AC infrastructure and deliver 800 VDC to compute racks within the row — no changes to the building's electrical system required."

So it is a smaller claim. It is also, if you operate a building that already exists, a far better product.

The physical argument has never been in doubt, and it is worth stating plainly because it is the reason any of this is happening. At 54 VDC, a 1 MW rack needs up to 200 kg of copper busbar. Scale that to a gigawatt campus and the rack busbars alone reach 200,000 kg of copper. And a Kyber-class rack running MW-scale power shelves at 54 V would eat up to 64 rack units (of 42-odd in a standard cabinet) before you install anything that computes.

What NVIDIA claims 800 VDC delivers against today's distribution (%)
More power through the same conductor85Less copper required45End-to-end efficiency gain5Maintenance cost reduction70Total cost of ownership reduction30

Hold those last three loosely. The 5% efficiency, 70% maintenance and 30% TCO figures were published in 2025 against a fully native 800 VDC facility with single-step conversion at the perimeter. A power rack sitting behind an unchanged AC plant keeps every upstream conversion stage it was supposed to delete, so it captures the in-rack part of the saving and very little of the rest. Nobody has published a number for the hybrid case, and until someone does, treat the 30% as a description of a building almost nobody will operate before 2029 (which is not the same as a wrong number).

So is the retreat bad news? Probably the opposite. Electrical standards get adopted through hybrids or they do not get adopted at all, and the supporting evidence here is unusually solid: Nvidia, Google and Microsoft developed the architecture jointly through the Open Compute Project, published a white paper in March and an LVDC solid-state transformer specification in July, and more than 80 equipment makers are building to it. A specification with 80 suppliers behind it is a supply chain rather than a slide.

What we would watch is the row. The power center — 2 MW per row over an overhead busway — is the step where an operator has to change something structural, and Nvidia dates it to 2027. SemiAnalysis says Nvidia will "collaborate with industry partners to expand demonstration and pilot sites ahead of large-scale deployment," which is the language of a product that does not yet have a reference customer to name.

Our expectation is that essentially every 800 VDC deployment announced through 2027 is a power rack inside an AC building, and that the hall-scale DC power block still has no named operator by the end of 2027. A greenfield gigawatt campus announcing native 800 VDC before then would move us.

Either way, somebody is going to save 200 tonnes of copper per gigawatt, and copper is not getting cheaper.

Sources

01
NVIDIA's 800 VDC rollout is phased and operates alongside existing AC distribution, with a power rack first, then a row-level power center, then a hall-level DC power blockNVIDIA’s 800 VDC plan includes a phased rollout that operates alongside existing AC distribution systems. The adoption level depends on each facility individually, rather than replacing an entire plant all at once. In the short term,…” — x.com · primary · Sep 10
02
NVIDIA's MGX-compatible 800 VDC power rack arrives in the second half of 2026 and slots into existing AC infrastructure with no changes to the building's electrical systemThe NVIDIA MGX-compatible 800 VDC power rack, arriving in the second half of 2026, creates a hybrid architecture that brings next-generation rack-scale compute performance to facilities that are already built and operational. It's…” — blogs.nvidia.com · primary · Sep 10
03
The row power center supports up to 2 megawatts per row over an overhead 800 VDC busway, with availability expected in 2027For operators building out dedicated AI factory environments, the row power center — a centralized power station for a full rack row — uses an overhead 800 VDC busway to scale power distribution across multiple rack rows, supporting up…” — blogs.nvidia.com · primary · Sep 10
Show all 10 sources
04
NVIDIA, Google and Microsoft developed the 800 VDC architecture through OCP, publishing a joint white paper in March 2026 and an LVDC solid-state transformer specification in July 2026, with more than 80 equipment makers building to itNVIDIA, Google and Microsoft have been developing the 800 VDC architecture together through the Open Compute Project (OCP), and published a joint white paper March 2026 and the LVDC Solid-State Transformer Specification v0.3 July 2026.…” — blogs.nvidia.com · primary · Sep 10
05
NVIDIA's 2025 architecture converted 13.8 kV AC grid power directly to 800 VDC at the data centre perimeterBy converting 13.8 kV AC grid power directly to 800 VDC at the data center perimeter using industrial-grade rectifiers, most intermediate conversion steps are eliminated. This streamlined approach minimizes energy losses, which typically…” — developer.nvidia.com · primary · Sep 10
06
A 1 MW rack at 54 VDC requires up to 200 kg of copper busbar, and a 1 GW data centre up to 200,000 kgCopper overload: The physics of using 54 VDC in a single 1 MW rack requires up to 200 kg of copper busbar. The rack busbars alone in a single 1 gigawatt (GW) data center could require up to 200,000 kg of copper.” — developer.nvidia.com · primary · Sep 10
07
At 54 VDC, power shelves for a Kyber rack at MW scale would consume up to 64 rack unitsUsing the same 54 VDC power distribution would mean power shelves would consume up to 64 U of rack space for Kyber at MW scale, leaving no room for compute.” — developer.nvidia.com · primary · Sep 10
08
NVIDIA claims 800 VDC busways carry 85% more power through the same conductor and cut copper by 45% versus 415 VAC, with up to 5% end-to-end efficiency gain, up to 70% lower maintenance cost and up to 30% TCO reductionUsing 800 V busways and switching from 415 VAC to 800 VDC in electrical distribution enables 85% more power to be transmitted through the same conductor size. ... With lower current, thinner conductors can handle the same load, reducing…” — developer.nvidia.com · primary · Sep 10
09
Full-scale production of 800 VDC data centres was planned to coincide with NVIDIA Kyber rack-scale systems in 2027Full-scale production of 800 VDC data centers will coincide with NVIDIA Kyber rack-scale systems in 2027, ensuring seamless scalability for increasingly demanding AI models.” — developer.nvidia.com · primary · Sep 10
10
Vladimir Troy of NVIDIA on the OCP ecosystem"800 VDC unlocks the compute performance and power density required for AI at scale," said Vladimir Troy, vice president of data center infrastructure at NVIDIA. "Through OCP, NVIDIA is working with more than 80 ecosystem companies to…” — blogs.nvidia.com · primary · Sep 10
Up next · Keep readingHardware · 3 min read

A20 Pro is the first 2nm phone chip. The number that matters is 32 Neural Engine cores after three years at 16

Apple doubled the Neural Engine, widened memory bandwidth 50 percent, moved the DRAM off the thermal path and tripled the vapor chamber. Every one of those decisions is about running models on the phone.

Continue ↓