IBM's Copper Replacement Research Could Reshape AI Server Farms

The short answer

IBM is not watching the copper-replacement materials race from the sidelines. IBM researchers have demonstrated cobalt monosilicide nanowires on 200 mm wafers, developed ruthenium interconnect structures for advanced chip nodes, coauthored the Cornell-led niobium arsenide nanowire study, and built a co-packaged optics prototype aimed at moving AI data with far less electrical power.

IBM and semiconductor researchers often call this post-Cu or beyond-copper work. Copper replacement is the plain-English description, although most designs would replace copper only at specific interconnect levels rather than everywhere.

None of those projects means copper disappears from IBM servers next year. They solve different problems, sit at different readiness levels, and still face difficult manufacturing tests. The important part is that IBM is already treating copper as a limit to work around rather than a permanent answer.

Copper is becoming three different bottlenecks

Copper remains an excellent bulk conductor. The problem changes when wiring dimensions shrink. Below roughly 20 to 30 nanometers, electron scattering at surfaces and grain boundaries drives resistance up. A material that looks worse than copper in a large wire can become more useful when both are reduced to chip-interconnect dimensions.

Copper also carries heat away from processors, and it carries electrical signals between chips, boards, and servers. AI infrastructure is pressing all three limits at once. More accelerators create more microscopic wiring, more data movement, and more concentrated heat. That is why the apparent collection of unrelated materials stories is actually one infrastructure story.

IBM's topological conductors get better as they get smaller

The Cornell-led niobium arsenide study is the most counterintuitive result. The team produced single-crystal NbAs nanowires down to 40 nanometers and measured lower resistivity as wire diameter decreased. At 40 nanometers, room-temperature resistivity was 10.5 plus or minus 1.9 micro-ohm-centimeters, about 70 percent below bulk NbAs. IBM Research in Albany and the IBM Thomas J. Watson Research Center appear in the author affiliations.

IBM's own cobalt monosilicide work goes smaller and moves closer to manufacturing reality. Researchers fabricated CoSi nanowires on 200 mm wafers with CMOS-compatible processes, reaching cross-sectional areas of 35 square nanometers, or about 6 nanometers in diameter. The near-epitaxial wires showed roughly an 80 percent resistivity reduction below 100 square nanometers compared with much larger CoSi wires.

That does not mean either material is ready to replace copper in a product. NbAs contains arsenic and the reported wires were formed through a specialized nanomolding process. CoSi still needs complete integration, yield, reliability, and cost evidence. What the work proves is more fundamental: surface-dominant conduction can reverse the scaling penalty that hurts copper.

Ruthenium may be the nearer post-copper step

IBM has also been developing ruthenium interconnects, a less exotic option with a more mature integration story. In 2025, IBM-affiliated researchers reported 16 nanometer pitch subtractive ruthenium lines with electrical resistivity below 20 micro-ohm-centimeters. IBM's earlier work with Samsung also demonstrated 18 nanometer pitch ruthenium top-via structures with embedded air gaps and reported reliability advantages over comparable copper structures.

The reason ruthenium remains interesting even though its bulk resistivity is higher than copper is architectural. Copper needs barrier and liner material around it. Those layers consume an increasing share of an extremely narrow interconnect. Ruthenium can be patterned differently and may deliver lower total line resistance and better reliability after the entire structure is counted.

The next replacement may be light rather than another metal

At longer distances, IBM is not only looking for a better electrical conductor. It is moving the signal into light. IBM's co-packaged optics prototype brings optical links much closer to the chip package, where it can complement short electrical connections and extend high-bandwidth links from meters to hundreds of meters.

IBM estimated more than a fivefold power reduction compared with mid-range electrical interconnects and modeled large AI-training gains from keeping accelerators fed with data. Those are prototype projections, not shipping-system specifications. The direction is still important. When processors sit idle waiting for data, adding more compute does not solve the real constraint.

Heat is a separate materials race

Theta-phase tantalum nitride belongs in the same discussion for a different reason. A UCLA-led team measured room-temperature thermal conductivity near 1,100 watts per meter-kelvin in single crystals, nearly three times copper's roughly 400 watts per meter-kelvin. That makes the material interesting for heat spreaders, packaging, power electronics, and local hot spots around AI accelerators. IBM was not listed as a participant in this work.

This result should not be translated into commercially available server cooling. The phase is difficult to stabilize, and the reported crystals required several gigapascals of pressure and temperatures of several thousand kelvin. Defects, grain boundaries, mixed phases, repeatable dimensions, attachment methods, and cost all stand between a record measurement and a data-center component.

The battery story is related, but it is not the same technology

Graphene Manufacturing Group is working on a graphene aluminium-ion battery with the University of Queensland, UniQuest, Rio Tinto, and the Battery Innovation Center. The company reports six-minute charging in laboratory testing, no lithium or rare-earth materials, and a 49 watt-hour-per-kilogram pouch-cell result in April 2026. It also identifies the program as Battery Technology Readiness Level 4, with pouch-cell optimization still underway.

That is graphene, not graphite, and it is not an IBM program. It matters to server farms because batteries sit behind uninterruptible power supplies, peak-shaving systems, and grid storage. Faster charging, long cycle life, and alternative chemistry could eventually change how facilities buffer power. At its current readiness and energy density, it is a development program rather than a replacement for production lithium-ion data-center systems.

What this could change inside server farms

The first impact will not be wholesale replacement of copper bus bars and building wire. It will happen where physical limits are most expensive: inside chips, at chip-package boundaries, between accelerators, and around concentrated hot spots.

Better nanoscale conductors could reduce resistance and delay in the dense wiring above transistors. Ruthenium could extend electrical interconnect scaling when copper structures become mostly liner. Optical links could move more data across boards, racks, and clusters for less power. Better thermal materials could remove heat at the package before the facility cooling system has to deal with it.

Taken together, those changes affect how many accelerators can be packed into a rack, how much power is lost moving data, how often processors wait for remote memory or peers, and how much cooling capacity is required for useful compute. The server-farm effect is therefore larger than the value of the conductor itself. It changes the system around the conductor.

This is also a national-security materials strategy

The commercial case is performance and energy. The strategic case is optionality. The United States has formally treated copper import dependence as a national-security and industrial-resilience issue, while the Department of Energy has funded conductor-manufacturing work through its CABLE competition. A July 2026 Government Accountability Office assessment also described batteries and semiconductors as strategically important industries exposed to vulnerable mineral supply chains.

Substitution does not automatically create independence. Niobium, arsenic, cobalt, ruthenium, and tantalum each bring their own sourcing and processing questions. GAO's conclusion is appropriately cautious: semiconductor substitutes are generally years from maturity, and domestic production facilities are expensive and specialized.

My read is that the strategic asset is not one miracle material. It is the ability to discover candidates, prove them on relevant wafers, integrate them into advanced packaging, and manufacture them through more than one supply chain. IBM's research role matters because that integration knowledge is much harder to replace than raw material alone.

A realistic readiness map

These projects should be compared by the job they perform and the evidence already demonstrated, not by treating every laboratory result as a copper wire replacement.

What this does and does not say about Power 12

It says nothing definitive about a Power 12 bill of materials. IBM has not published a Power 12 processor process, package design, interconnect metal, optical interface, thermal system, model number, or launch date. Any claim that Power 12 will use CoSi, NbAs, ruthenium, or theta-phase TaN would be fabrication.

It does say something useful about IBM's engineering priorities beyond the current generation. Power systems are becoming more dependent on AI acceleration, dense packaging, chip-to-chip communication, memory bandwidth, and energy efficiency. IBM's work on topological conductors, ruthenium, optics, and advanced packaging belongs beside its confirmed chiplet direction when evaluating the constraints a future Power design must solve. The materials are not confirmed. The problems are.

The evidence that would move this from research to deployment

The next meaningful milestones are not another dramatic conductivity number. They are repeatable thin-film deposition, compatibility with 300 mm manufacturing, stable interfaces with surrounding dielectrics, electromigration lifetime, thermal cycling, contamination control, high yield, and a cost that a foundry can defend.

For IBM Power, the decisive evidence would be an IBM announcement letter, Redbook, technical manual, foundry disclosure, or product-level packaging statement. Until one of those appears, this is a strategic research direction with real industry consequences, not a product specification.

Frequently Asked Questions

Is IBM replacing copper in its current Power servers?

No public IBM product documentation says current Power servers use these post-copper materials. IBM's published evidence describes research prototypes and advanced-node development.

Why can a conductor improve as it gets smaller?

Topological semimetals can carry a larger share of current through protected surface states. As the wire shrinks, those surface channels become more important, while conventional copper suffers more scattering at surfaces and grain boundaries.

Will Power 12 use a copper replacement?

IBM has not published a Power 12 interconnect material or package specification. The research identifies relevant engineering directions, but it does not confirm a Power 12 implementation.

Is theta-phase tantalum nitride ready for server cooling?

No. The reported thermal result came from carefully synthesized single crystals under extreme pressure and temperature. Commercial use would require scalable manufacturing, repeatable material quality, component integration, and long-term reliability evidence.

Why is post-copper research a national-security issue?

Batteries, semiconductors, power systems, and defense equipment depend on vulnerable mineral supply chains. Alternative materials and domestic manufacturing knowledge create more sourcing options, although every substitute introduces its own material and processing dependencies.

Related Resources

• IBM's Copper Replacement Research Is a Power 12 Roadmap Signal

Sources

• IBM Research: Resistivity scaling of topological CoSi wires
• PubMed: Cornell and IBM NbAs nanowire study in Science
• IBM Research: 16 nm pitch subtractive ruthenium interconnects
• IBM Research: Copper evolution and post-copper interconnects
• IBM Newsroom: Co-packaged optics for AI data centers
• Science via PubMed Central: Theta-phase tantalum nitride thermal conductivity
• Graphene Manufacturing Group: Graphene aluminium-ion battery program
• U.S. GAO: Critical minerals substitution and recycling technologies
• U.S. GAO: Copper imports and national security review
• U.S. Department of Energy: CABLE conductor manufacturing prize