Data centers, where large language models are trained and requests to AI services are processed, are consuming increasingly more energy. And it is not just about the computations themselves – a significant part of the power is lost even before reaching the servers. Ordinary copper cables heat up, waste energy as heat, require cooling, and take up space. Microsoft decided to check whether this problem could be solved not by optimizing algorithms, but by replacing the wires themselves.
The company published a note stating that it is testing high-temperature superconductors for powering server racks. Long story short: these are materials that, under specific cooling, transmit current almost without losses. In theory, this allows delivering more power within a smaller footprint, reducing heating, and simplifying the cooling system.
How High-Temperature Superconductors Work in Modern Computing
What Are Superconductors and Why Are They Back in the Spotlight?
Superconductivity is the property of certain materials to lose electrical resistance when cooled below a specific temperature. Regular wires always resist current, which causes some energy to turn into heat. Under the right conditions, superconductors work differently: current flows without losses, the wires do not heat up, and the energy reaches the consumer almost in its entirety.
The problem is that classical superconductors only work at temperatures close to absolute zero, which is expensive and difficult to maintain. The high-temperature superconductors Microsoft writes about transition into the necessary state under «warmer» conditions – for example, at the temperature of liquid nitrogen. This is significantly simpler and cheaper to maintain than using liquid helium.
The technology has been known for a long time, but it never reached industrial application in data centers. Now that the power density of server racks is growing along with the load from AI models, the question of energy efficiency is becoming less academic and quite practical.
Benefits of Using Superconductors for Data Center Energy Efficiency
Why Data Centers Need This
Modern servers, especially those involved in AI tasks, require increasingly more energy per rack. The copper busbars currently used to transmit electricity inside data centers have limitations: they are bulky, heavy, heat up, and require additional cooling. The higher the load, the more these drawbacks manifest themselves.
Superconducting cables, in theory, should solve several problems at once. Firstly, they transmit current with minimal losses, which means generating less heat and cutting cooling costs. Secondly, they can be more compact: with the same throughput, a superconductor takes up less space than a copper busbar. Thirdly, this reduces the overall load on the power supply system and simplifies scaling.
Microsoft does not disclose the technical details of the experiment but notes that interest in the technology is linked specifically to the growth in energy consumption of AI infrastructure. Simply put, if models continue to become more complex and servers continue to consume more energy, standard solutions may cease to cope.
Challenges and Barriers to Superconductor Implementation in Data Centers
What Is Hindering Adoption
Despite the potential, high-temperature superconductors have not yet become a mass-market solution. There are several reasons for this.
The first is cost. The materials themselves are more expensive than copper, and their production requires specialized processes. The second is the need to maintain low temperatures. Even «high-temperature» superconductors function at temperatures significantly below room temperature, which requires a cooling system. The third is reliability and longevity. It is still unclear how such cables will behave under conditions of round-the-clock operation over many years.
Furthermore, data center infrastructure is built for traditional solutions. Switching to superconductors would require not only replacing cables but also rethinking the entire power supply and cooling system. This is complex, expensive, and risky – especially while the technology has not yet proven its economic feasibility in real-world conditions.
Future Outlook for Superconducting Technology in AI Infrastructure
An Experiment or a Signal to the Industry?
Microsoft's publication is more a signal that the company is exploring alternatives than an announcement of a ready-made solution. For now, this is an experiment, and it is unknown whether it will lead to actual implementation. But the very fact that one of the largest companies in the field of cloud computing and AI has turned its attention to this technology speaks to the seriousness of the problem.
The growth in the power of AI models is not slowing down. Training large models, their inference, and working with multimodal data – all of this requires not only processor time but also stable, efficient power supply. If infrastructure does not keep pace with growing needs, it will become a bottleneck for the entire industry.
Superconductors are one possible path to a solution. Not the only one, and not necessarily the fastest, but interesting enough to test. Perhaps in a few years, we will see the first commercial implementations, or perhaps the technology will remain niche. For now, Microsoft is doing what companies at the cutting edge of progress do: testing hypotheses that might not pay off immediately but are capable of changing the rules of the game in the long term.