Article
Could superconductors power AI data centers?
AI data centers are concentrating more computing power in fewer facilities, putting pressure on the electrical path from utility connection to server. Superconducting DC busbars are a studied option for moving large currents compactly, but a proposed layout is not the same thing as an operating data center.
- Published
- Reading time
- 4 min
- Cited
- 7 sources
Why does AI increase power-delivery pressure?
The International Energy Agency estimates that data centers used around 415 terawatt-hours of electricity in 2024, and its base case projects around 945 terawatt-hours in 2030 [1]. The IEA identifies accelerated servers, used heavily for AI, as a driver of greater power density inside data centers [1]. The IEA figures cover the whole data-center sector rather than a single AI campus.
A facility's bottleneck is not only the amount of electricity available at the grid connection. Power must pass through transformers, switchgear, converters, busways, and connectors before it reaches computing equipment. As more electrical capacity is installed in a fixed building footprint, the conductors and equipment that distribute it must handle high current without excessive heating, voltage drop, or space requirements. A cable technology that can carry substantial current in a compact corridor is therefore worth evaluating, but only as one part of a larger power-delivery design. [1][2]
Why consider a superconducting DC busbar?
For a given power, lowering voltage requires raising current. Resistive heating increases with the square of current, so a high-current copper bus can require a large conductor cross-section and careful thermal management. A superconductor can carry current with very low electrical resistance while it stays within its temperature, field, and current limits. That combination makes a superconducting busbar or cable an engineering candidate where current density and space matter, not a universal replacement for copper. [2]
The DC form is relevant because server power systems convert incoming power through several stages and distribute it to electronics that ultimately use DC. A superconducting DC link could be placed at a high-current point in that chain, such as a facility bus between conversion equipment and downstream distribution. The conversion architecture still matters: a cable does not remove rectifiers, protection, standby systems, or the need to deliver power at the voltage each load requires. The useful comparison is a complete electrical path, not conductor resistance in isolation. [2]
What has been demonstrated, and what is proposed?
An IEEE Transactions on Applied Superconductivity paper documents construction and circulation testing for the Ishikari DC superconducting cable [3]. The Ishikari project consortium's August 2015 release reports a 500-meter field transmission test with a current of 5 kiloamperes and capacity of 100 megavolt-amperes [6]. Sakura Internet's September 2015 release says it began sending direct-current power from its solar plant to the Ishikari data center, where direct-current servers were supplied directly [7]. Together, these sources document an external solar-to-data-center field demonstration, not an AI server-bus installation or proof of continued operation today. [3][6][7]
A separate line of work is design analysis. Chen and colleagues published an Energy paper titled as a design and economic evaluation of superconducting DC busbar networks for a 10-megawatt-class data center [2]. The publication establishes that researchers have evaluated the concept; it is not evidence that such a network has been installed in a commercial data center. Keep the distinction clear when comparing a cable test, a modeled architecture, a vendor proposal, and an operating facility. Each answers a different question about feasibility.
What limits a real installation?
The Ishikari project consortium reported that its new piping configuration reduced heat loss in the transmission route and liquid-nitrogen circulation loss. This illustrates that cryostat and coolant-loop design are part of the electrical system's efficiency, not a secondary detail. [6]
Cost and reliability are equally important. The superconducting tape, cryostat, terminations, cooling plant, protection equipment, installation, and service contracts all contribute to lifecycle cost. A fault can push part of the conductor beyond its operating envelope and create a resistive region, or quench. The grid then needs protection that detects and safely isolates both an electrical fault and a developing thermal problem. IEEE research on HTS cable protection treats quench detection and short-circuit protection as distinct but coordinated functions. [4]
What should a data-center buyer ask a vendor?
Ask for a one-line diagram that identifies the proposed cable's exact place in the power chain, its DC voltage, continuous and transient current, load profile, and the conventional copper alternative. Require conductor performance at the specified temperature and field, including engineering current capacity for the complete cable rather than a tape-only figure. Ask whether ratings include the terminations, joints, coolant loop, and protection system, and what test evidence supports each rating. A headline conductor rating is not a facility-level power rating.
Then examine the cooling and failure case. Request measured heat leak, cooling power at the operating point, expected auxiliary load, redundancy, coolant-flow monitoring, vacuum monitoring, and response to a pump or refrigerator outage. Ask how quench detection coordinates with breakers, how a faulted section can be isolated or bypassed, and what recovery procedure restores service. Require test records for overloads, thermal cycling, joints, and insulation, plus stated availability assumptions and maintenance intervals. IEC's HTS AC cable standard explicitly does not cover cooling-system test requirements, so the system boundary and additional qualification plan should be written down. [5]
Where might superconductors make sense first?
The strongest case is a constrained route where conventional conductors would consume valuable space or require expensive electrical infrastructure, and where high current is concentrated over a defined link. A data center with a planned DC architecture could compare an HTS trunk against copper busways, transformers, switchgear, and cooling as one system. The comparison should use delivered energy, capital cost, serviceability, and outage risk under the site's actual load profile, not an assumed always-full operating point. [2]
For now, the evidence supports a careful conclusion: superconducting power delivery has reached field demonstrations, and superconducting DC busbars have been analyzed for data-center designs. Those are meaningful engineering steps, but they do not amount to broad adoption in AI facilities. A credible project needs independent system-level measurements, a practical protection strategy, a maintainable cryogenic plant, and economics that remain favorable after auxiliary power and lifecycle service are counted.
Sources
- Energy demand from AIInternational Energy Agency, 2025
- A 10 MW class data center with ultra-dense high-efficiency energy distribution: Design and economic evaluation of superconducting DC busbar networksEnergy, 2022
- Construction and the Circulation Test of the 500-m and 1000-m DC Superconducting Power Cables in IshikariIEEE Transactions on Applied Superconductivity, 2016
- Research on configuration scheme and principle of HTS cable protectionIEEE, 2021
- IEC 63075:2019, Superconducting AC power cables and their accessories for rated voltages from 6 kV to 500 kVInternational Electrotechnical Commission, 2019
- 500mの直流超電導送電に成功！Ishikari Superconductivity and DC Transmission System Technology Research Association, 2015
- データセンターへの超電導送電を開始Sakura Internet and Ishikari project consortium, 2015
SuperMatics: AI built for superconductors.