Nvidia’s 800-volt “revolution”: Global data centers face the largest infrastructure overhaul in history

As the AI arms race enters a new stage, Nvidia is leading a historic “revolution” in global data center power architecture: shifting the voltage standard from traditional AC to 800-volt DC.

Nvidia recently announced more than ten partners including CoreWeave and Oracle, aiming to prepare for an 800V DC power architecture and ultra-high-density computing environments with single-cabinet power density reaching 1 megawatt (MW). This transformation is to support its next-generation “Vera Rubin” architecture and “Kyber” system, expected to debut in 2027, with each cabinet integrating 576 GPUs, whose power and cooling requirements far exceed the limits of the current 415V AC architecture.

Goldman Sachs stated in its latest research report that this technological leap means the focus of capital expenditure for data centers will shift significantly. Investors have begun to re-evaluate the winners and losers in the capital goods industry, as not only will the infrastructure funding gap widen further, but the entire supply chain—from transformers, circuit breakers, cables to cooling systems—will also face mandatory technical upgrades and replacement cycles.

Despite Nvidia predicting that this architecture can reduce total cost of ownership (TCO) by 30% in the long term, in the short term, it undoubtedly presents a huge capital expenditure hurdle. This transformation forces operators to purchase millions of new devices, triggering the industry’s first round of large-scale hardware upgrade cycles.

Breaking Physical Limits: Leaping from Tens of Kilowatts to Megawatt Level

The core driver for data centers transitioning to the 800VDC architecture is the exponentially growing power density requirements of modern AI cabinets. Current cabinet power is rapidly rising from tens of kilowatts to over 1 megawatt, which already exceeds the physical capabilities of traditional 54V or 415/480VAC systems.

Nvidia points out that compared to traditional AC systems, the 800VDC architecture can transmit over 150% more electricity on the same copper conductor, greatly improving energy efficiency. This architecture can reduce copper usage by up to 45%, and even eliminate the need for heavy copper busbars weighing 200 kg required to power a single cabinet.

To accommodate this extreme power density, Nvidia’s next-generation Vera Rubin NVL144 rack design adopts 45°C liquid cooling technology and new liquid-cooled busbars, and increases energy storage capacity by 20 times to maintain power stability. The subsequent Kyber system will contain 18 vertically rotating compute blades, arranged like “books on a shelf,” to support the ever-growing inference demands.

Infrastructure Restructuring: DC and Liquid Cooling Take Over

Goldman Sachs analyst Daniela Costa elaborated on the specific infrastructure impacts of this revolution in her report. The most notable change is that traditional AC power distribution units (PDU) and AC uninterruptible power supply (UPS) systems will become unnecessary. The 800VDC architecture requires a streamlined power path, with centralized integration of battery storage systems at the facility level, replacing dispersed UPS units. These large-scale facility-level battery systems can manage power fluctuations and ensure grid stability, reducing the need for AC PDU cabinets by up to 75%.

For existing data centers, before a complete rebuild, the “Sidecar” model will become the key transitional solution between 2025 and 2027. These modules can be installed on the sides of computing racks, convert incoming AC power to 800VDC, and provide integrated short-term energy storage to smooth GPU load spikes. Schneider Electric, a key supplier of such equipment, is clearly targeting the market for racks up to 1.2MW.

Additionally, as rack power moves toward 1.2MW, traditional air cooling is no longer sufficient—liquid cooling technology will become absolutely mainstream. Schneider Electric has significant exposure in this area through its Motivair assets, and Vertiv has released an 800VDC MGX reference architecture combining power and cooling infrastructure.

Supply Chain Shakeup: Who Benefits from Capital Expenditure?

This technological paradigm shift is redefining market share in the capital goods sector. According to Goldman Sachs, Legrand estimates that the shift to higher voltages will raise per-megawatt revenue potential from €2 million in traditional data centers to a possible €3 million. Although currently three-quarters of rack power remains below 10kW, the industry expects the 800VDC architecture to become the mainstream choice for 80-90% of new data centers in the future.

In the power semiconductor sector, the transition to 800VDC requires more advanced chips, particularly silicon carbide (SiC) and gallium nitride (GaN), to handle higher voltages and frequencies. Suppliers including Analog Devices, Infineon, STMicroelectronics, and Texas Instruments are actively investing in this space.

For power protection and switching devices, mechanical circuit breakers are being replaced by solid-state protection devices. ABB’s current SACE Infinitus is considered the world’s first IEC-certified solid-state circuit breaker, designed for DC distribution, which gives ABB a lead in MV DC UPS systems and other areas. Meanwhile, cable giants Prysmian and Nexans are developing high-end cable solutions tailored to DC and liquid cooling demands.

Timeline and Costs: The 2027 Threshold

Although the outlook for this transformation is promising, full-scale commercialization will take some time. Nvidia expects the transition to 800VDC data centers to synchronize with the deployment of its Kyber rack architecture, with a target date of 2027. Goldman Sachs estimates that the commercial application of related technologies will begin to show scale effects around 2028.

Schneider Electric Data Center CTO Jim Simonelli stated that the move to 800VDC is the “natural evolution” as compute density increases. Although this can cut operating costs and reduce maintenance in the long run, for data center operators who must foot the bill, this means that in the next five years, in addition to the known $5 trillion AI funding gap, they will need to invest enormously for this largest-ever infrastructure transformation.

Risk Warning and DisclaimerThe market has risks; investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users’ special investment objectives, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions herein fit their particular circumstances. Investments made according to this article are at one’s own risk.