AI data centers adopting 800VDC power architecture could cut copper use by 45% per rack while creating new demand for gallium nitride power semiconductors.
At a glance
- At 1 megawatt per rack, conventional 54VDC data center power distribution requires up to 200 kilograms of copper busbar per rack; shifting to 800VDC architecture eliminates that need and cuts overall copper use by an estimated 45%.
- The 800VDC transition relies on gallium nitride (GaN) chips for fast, precise low-voltage power delivery to GPUs, alongside silicon carbide (SiC) for high-voltage conversion -- making both wide-bandgap semiconductor materials direct beneficiaries of the shift.
- The 800VDC hardware ecosystem had grown to 29 named companies by mid-2026, with Texas Instruments and STMicroelectronics unveiling GaN-based power conversion chips in March 2026 claiming up to 97.6% efficiency.
- 2026 is considered a pivotal year for 800VDC stabilizing as the standard architecture for AI data centers, as rack power draw has pushed past 1 megawatt and made the copper and space savings of higher-voltage distribution increasingly necessary rather than optional.
What happened
AI data centers are shifting toward 800-volt direct current (800VDC) power architecture as AI training and inference racks push past 1 megawatt of power draw each, with 2026 emerging as a pivotal year as power semiconductor vendors bring 800VDC-compatible products to market. At 1 MW per rack, the conventional 54VDC distribution system requires up to 200 kilograms of copper busbar per rack just to carry the current at lower voltage; 800VDC eliminates that requirement, reduces conversion losses and cuts overall copper use by an estimated 45%. The architecture leans on two power semiconductor materials: silicon carbide (SiC) for high-voltage conversion in components like solid-state transformers, and gallium nitride (GaN) for delivering precise low-voltage power to GPUs at high speed. By mid-2026, the 800VDC hardware ecosystem had grown to 29 named companies, and in March 2026, Texas Instruments and STMicroelectronics both unveiled GaN-based power conversion chips claiming efficiencies up to 97.6% for 800V-to-low-voltage conversion.
The details
The shift to 800VDC power architecture in AI data centers is, at its core, a response to a physics problem: as AI training racks have pushed past 1 megawatt of power draw each, the copper required to deliver that power at conventional 54-volt distribution has become a genuine constraint, not just a cost line. At 1 MW per rack, a 54VDC system needs up to 200 kilograms of copper busbar just to carry the current -- copper that also consumes rack space increasingly needed for compute hardware itself. Moving to 800 volts lets the same power flow through dramatically less conductor, because higher voltage means lower current for the same wattage, and lower current means less copper is needed to avoid excessive resistive losses. The result, by current estimates, is roughly 45% less copper required per rack.
That copper reduction is the headline number, but it's only half the story -- the other half is what has to replace the old system's power-conversion stages. Two wide-bandgap semiconductor materials split the job. Silicon carbide (SiC) handles the high-voltage end, stabilizing conversion in components like solid-state transformers that step incoming utility power down to the 800V rail. Gallium nitride (GaN), which switches faster and more efficiently than older silicon-based chips, takes over from there, delivering precise, high-speed power conversion from 800V down to the low voltages GPUs actually run on. Industry commentary has taken to describing SiC as the architecture's 'strongman' and GaN as its 'sprinter' -- a division of labor that reflects each material's genuine electrical properties rather than marketing shorthand.
The scale of the emerging GaN opportunity is concrete, not speculative. Texas Instruments and STMicroelectronics both unveiled complete GaN-based power conversion chips in March 2026, claiming efficiencies up to 97.6% for the 800V-to-low-voltage conversion step -- a meaningful efficiency gain that matters directly to data center operators, since every percentage point of conversion loss at gigawatt-scale AI facilities translates into real electricity cost and waste heat that then has to be cooled away. By mid-2026, the broader 800VDC hardware ecosystem, spanning chipmakers, power-supply vendors and data center infrastructure companies, had grown to 29 named participants -- evidence this is now a genuine multi-vendor industry buildout rather than a single company's roadmap.
For the metals behind this shift, the implications cut in two directions at once. Copper, already stretched thin by demand from EVs, grid buildouts and other AI-data-center uses like networking and cooling, gets a genuine demand offset from the architecture shift -- even as overall data center construction accelerates, the copper intensity of each new rack is falling. Gallium, by contrast, gains a new, concrete source of demand growth through GaN chip production, a metal that has historically traded on a much smaller, more specialized market than copper and has gone extended stretches with minimal price-moving news of its own.
Why it matters
AI data center buildout is one of the largest sources of new metals demand this decade, and how that demand is actually structured matters as much as its total size. A 45% cut in copper intensity per rack doesn't mean data centers need less copper overall -- construction is accelerating faster than that efficiency gain -- but it does mean copper demand forecasts built on a fixed copper-per-megawatt assumption could overstate the data-center contribution, while gallium, a metal that rarely gets dedicated market coverage, is gaining a genuine new demand driver through GaN chip production worth tracking in its own right.
Our read
Outlook: neutral. This is a structural, multi-year demand-shift story rather than a near-term price catalyst for either gallium or copper -- the mechanism (higher gallium demand via GaN chips, lower copper intensity per rack) is well-documented, but its price impact depends on the pace of 800VDC adoption relative to overall AI data center construction growth, which remains uncertain.
What to watch
- The pace of 800VDC adoption across major AI data center operators and hyperscalers as a leading indicator of gallium nitride chip demand growth
- New product announcements from the 29-company 800VDC hardware ecosystem, particularly around GaN chip efficiency and cost improvements
- Data center construction growth rates relative to the 45% per-rack copper savings, to gauge the net effect on overall copper demand from AI infrastructure buildout
For information only, not investment advice.
Gallium price in India
metalscost.com India reference price as of 2026-11-01.
Detailed analysis
Timeline
- 2026-03-01: Texas Instruments and STMicroelectronics unveiled complete GaN-based power conversion ICs for 800V-to-low-voltage conversion, claiming efficiencies up to 97.6%.
- 2026-06-01: The 800VDC hardware ecosystem had grown to 29 named companies, reflecting broadening multi-vendor adoption of the architecture.
Demand Drivers
Gallium demand is gaining a new, concrete driver from gallium nitride (GaN) power conversion chips used in 800VDC data center architecture; Texas Instruments and STMicroelectronics both launched GaN-based 800V conversion chips in March 2026, and the broader 800VDC hardware ecosystem had grown to 29 companies by mid-2026.
Global Consumption
The 800VDC architecture shift is estimated to cut copper use by about 45% per data center rack at 1 megawatt of power draw, an offset to overall copper consumption growth even as AI data center construction itself accelerates.
What could lift prices
- Gallium gains a concrete, quantifiable new demand source through GaN power chips, with a 29-company hardware ecosystem and major chipmakers like Texas Instruments and STMicroelectronics already shipping products as of March 2026.
- 2026 is described as a pivotal, stabilizing year for the 800VDC standard, suggesting the architecture shift -- and the gallium demand tied to it -- is moving from early adoption toward broader industry standardization.
What could weigh on prices
- The same architecture shift that boosts gallium demand structurally reduces copper intensity per data center rack by an estimated 45%, a genuine demand offset for copper even as overall AI data center construction accelerates.
Country impact
| Country | Impact | Reason |
|---|---|---|
| United States | Medium | US-based chipmakers Texas Instruments and NVIDIA's ecosystem partner Navitas are among the companies developing 800VDC power conversion technology, positioning US semiconductor firms as key suppliers for the architecture shift underway at AI data centers globally. |
Industry impact
| Industry | Effect | Reason |
|---|---|---|
| Semiconductors | Positive | The 800VDC transition is creating concrete new product demand for gallium nitride and silicon carbide power chips, with a 29-company hardware ecosystem already built out by mid-2026 around this specific architecture shift. |
| Data Centers | Neutral | 800VDC architecture reduces copper costs and conversion losses for data center operators building AI infrastructure, but requires new capital investment in different power hardware -- a genuine tradeoff rather than a clear net positive or negative for the industry. |
Who gains, who loses
- Gallium nitride and silicon carbide chipmakers: Companies like Texas Instruments, STMicroelectronics and Navitas are positioned to capture new revenue from the 800VDC architecture shift, which requires their wide-bandgap semiconductor products for both high- and low-voltage power conversion stages.
- Copper busbar and conventional power-distribution suppliers: The estimated 45% reduction in copper use per rack under 800VDC architecture directly reduces demand for the copper busbar and lower-voltage power-distribution hardware that conventional 54VDC data center racks require.
Other ways this could play out
- If 800VDC adoption accelerates faster than currently expected as AI rack power draw keeps climbing past 1 megawatt, gallium demand from GaN chip production could scale meaningfully faster than the metal's historically thin market has priced in.
- If copper prices remain elevated for reasons unrelated to data centers -- as seen in 2026's broader copper supply squeeze -- the 45% per-rack copper savings from 800VDC could become an even stronger incentive for data center operators to accelerate the architecture switch.
Historical comparison
- Conventional 54VDC architecture: At 1 megawatt per rack, a 54VDC system requires up to 200 kilograms of copper busbar per rack -- the baseline the 800VDC shift is estimated to cut by about 45%.
Technical view
Price is trading below both its 20-period and 50-period moving averages, a bearish alignment.
Computed from metalscost.com's own stored price history.
Related
United States