Data centres have become a point of interest for wire and cable demand over the last couple of years. Hyperscale and cloud facilities continue to capture headlines on historic capex investments and strong buildout efforts, while material requirements for power distribution, computing processes, cooling and heat management, and data transmission grow. As AI data centres expect larger power draw demand at the server rack level, facilities are expected to undergo a very fundamental change in power supply architecture – a gradual shift from traditional AC systems to 800V DC. At its core, it will impact AI computing costs, data centre design and an entire power equipment supply chain.
The core rationale is straightforward. As upcoming next-generation GPU clusters push rack power densities towards 600 kW to 660 kW and beyond (such as NVIDIA Kyber Ultra and Feynman architectures), traditional 48V–54V DC rack-level distribution following 415-480V AC distribution at facility level breaks down due to extreme copper requirements and resistive losses.
Historically, server racks operated at relatively modest power levels, and 48V–54V DC distribution remained workable in that environment. That is no longer the case for next-generation AI racks. As rack power densities move toward 600 kW and beyond, low-voltage architectures require extremely high current to deliver power. The higher the current, the thicker the copper wires and busbars need to be, the more heat is generated, the greater the power losses, and the less space remains in the rack for GPUs, networking, and cooling systems.
When voltage is increased, the current required to deliver the same power decreases proportionally. For example, delivering 400 kW of power requires over 8,300 amps at 48V, but only 500 amps at 800V DC. With lower current flowing through the conductors, oversized copper busbars and thick cables are no longer required. According to NVIDIA’s white paper on the topic released last year, the same cross-sectional wire gauge can carry up to about 157% more power, requiring up to approximately 45% less total copper. Other claimed benefits include up to 5% efficiency improvement, up to 30% lower total cost of ownership, and up to 70% lower maintenance costs.
800V DC adoption hinges on platform and hyperscale readiness
The adoption pathway for 800V DC is closely tied to NVIDIA's next-generation AI rack roadmap, particularly NVIDIA's upcoming Rubin Ultra and Feynman platforms and related next-generation high-density rack designs. These systems are expected to push rack power requirements to levels at which legacy 48V–54V rack-level distribution becomes increasingly impractical due to current, conductor sizing and thermal losses.
Market expectations currently point to a 2027 launch window, with broader deployment potential building from the second half of that year, especially as Kyber Ultra rack deployments begin to shape hyperscale power design requirements. It is worth noting that the chip generations ramping in late 2026 and 2027, such as Vera Rubin NVL72, top out at rack densities of 180–220 kW, a range that three-phase AC can still deliver without hitting the physical limits of conductor sizing or distribution losses. As such, Vera Rubin-class platforms are not expected to be primary drivers of 800V DC adoption – that transition is more closely tied to the higher-density Rubin Ultra and Feynman generations.
The commercial significance of 800V DC is likely to emerge from late 2027–early 2028 onwards, aligned with the expected launch window for industry leader NVIDIA's Rubin Ultra racks and broader deployment of very high-density AI racks such as the Feynman platform. In the near term, industry attention may stay elevated, but true volume adoption will depend on platform readiness, standards development, hyperscale operator acceptance and evidence that system-level economics are compelling versus alternatives such as 400V DC, ±400V bipolar DC, and enhanced low-voltage architectures.
Early implementation is likely to remain concentrated in newer hyperscale greenfield developments in the late 2020s, where operators have greater scope to redesign facility-level electrical architecture from the outset. Wider adoption is more likely to build from the early 2030s as standards mature, the supply chain broadens and high-density AI power designs become more established.
At the same time, the 800V DC discussion sits within a larger constraint – power availability is becoming the critical bottleneck in AI infrastructure deployment. The competitive frontier is no longer just GPUs and servers. It is increasingly moving to grid access, interconnection, substation capacity, transformers, backup systems and integrated campus power design. 800V DC should, therefore, be viewed as one layer of a broader electrification and power-distribution buildout across the AI data centre value chain.
Cable and non-cable metallic intensities expected to lower
In the near term, wire and cable demand should continue to benefit from the expansion of data centre campuses and rising power requirements. Large-scale facilities still need extensive electrical networks spanning utility interconnection, internal distribution, backup systems and cooling infrastructure. At the campus and facility levels, conductor demand remains substantial and is not under threat from architectural change.
The more significant shift occurs within rack-adjacent and internal power distribution. The heavy-gauge copper cables, busbars and busway assemblies required to distribute power at 54V across a 600 kW or 1 MW rack environment are replaced by materially smaller, lighter conductors under an 800V DC architecture. Across an entire high-density AI campus, that reduction compounds across thousands of metres of internal cabling, row-level distribution and rack-adjacent power delivery.
The same physics apply with equal force to non-cable conductor products. A busbar assembly rated to distribute 600 kW at 54V must handle currents in excess of 11,000 amps. The same assembly rated for 800V carries less than 750 amps for the same power. The cross-sectional area of copper required is a fraction of that needed at low voltage, resulting in a system that is physically smaller, lighter, and materially less copper-intensive, even if more technically sophisticated and higher in unit value.
Additionally, the traditional AC/DC distribution chain – typically involving four to five conversion steps across transformers, switchgear, UPS units and PDUs – will progressively be retired in favour of more compact, streamlined architectures. Equipment such as solid-state transformers (SSTs) consolidates what were previously discrete conversion stages into a single unit, delivering medium-voltage AC directly as high-voltage DC. This not only reduces the physical footprint of power infrastructure but also lowers the metallic intensity per MW, as fewer transformer windings, busbars and enclosures are required across the distribution chain.
Taken together, in CRU's base case, the result is strong absolute growth in data centre-related metallic demand over the next few years. However, copper intensity per unit of incremental capacity declines over time, with an increasing divergence after 2030 between the capacity-led growth trajectory and the lower conductor tonnage growth path. The broader analytical shift is from "more power means proportionally more metal" toward "more power, but more efficient metal use" – and that reframing is likely to become an increasingly important feature of long-run market views.
800V DC adoption is credible, but not guaranteed
The case for 800V DC is strong where rack power density is highest, but adoption is unlikely to be uniform or immediate as several risks remain. The first risk is standards – high-voltage DC deployment inside data centres still requires greater harmonisation around safety, grounding, connectors and facility-wide protection.
The second is ecosystem readiness – 800V DC depends on coordinated deployment across central rectification, DC busways, solid-state breakers, energy buffering, step-down DC/DC conversions (eg. 800 V → 48 V→ 12 V, 800 V → 12 V, etc.) and cooling integration. The third is market fragmentation – some operators may adopt 800V DC, while others pursue alternative high-voltage or hybrid pathways.
These uncertainties affect the pace of the thrifting story. A slower rollout delays intensity reduction, while a faster move by hyperscalers and AI-focused operators would bring that effect forward. Even so, the direction of travel is increasingly clear – as rack power climbs, the legacy 48V–54V model becomes progressively harder to defend on efficiency, thermal and space grounds.
800V DC delivers more capacity with less copper per unit installed
For metals markets, the key takeaway is not that 800V DC reduces demand outright but rather changes its intensity . Higher-voltage distribution lowers current, which in turn reduces the copper required in internal power delivery systems such as cables, busbars and busways. Data centre capacity can continue to expand rapidly, while metallic demand per unit of incremental capacity begins to fall.
The late 2020s and early 2030s are the critical inflection point regarding 800V DC architecture. Up to then, data centre metallic demand is still largely supported by rapid buildout and rising power density. Beyond that point, newer capacity additions should increasingly embed more efficient power architectures, moderating copper and total metallic demand growth across both wire and cable and non-cable conductor categories.
