VEIR raised $110M to replace copper inside AI data centres — and bought a cryogenics problem

On 8 October, VEIR disclosed a $110 million oversubscribed Series C, co-led by Matter Venture Partners and Tyche Partners. The Woburn, Massachusetts company builds superconducting power delivery systems — cables that carry current with almost no electrical resistance once cooled, bundled with the cryogenics, controls and monitoring required to keep them running inside a data centre.

The physics works. The business question is harder, and it is not about whether superconductors conduct.

Dealroom places the round around the 91st percentile of 446 comparable US energy Series C rounds. Total capital raised since 2019 now stands at $225 million, following a $75 million Series B in January 2025. Valuation was not disclosed.

Copper stopped being the default

Superconductors are not new. The discovery is seventy years old. What changed is the operating environment on the demand side, where the constraint migrated from silicon to electrons.

AI server racks pull power densities that copper busway struggles to serve without getting physically larger. Every extra megawatt of density forces more copper, more space, more termination hardware and more installation labour inside a building whose floor area is already spoken for. The metal does not get better under load. It gets bigger.

AI is critically important and requires more efficient power delivery systems to continue to scale, while the physics of copper has become a fundamental limitation.— Weijie Yun, Managing Partner, Tyche Partners

Investors backing VEIR are making the same argument in different registers. Wen Hsieh, who led this round, frames it as an infrastructure ceiling rather than a materials problem.

The next generation of AI data center will require innovations beyond compute.— Wen Hsieh, Founding Managing Partner, Matter Venture Partners

Axios framed the same round more plainly, noting that cost and engineering hurdles have limited high-temperature superconductor deployment so far. That is the problem VEIR exists to solve, and the round is the market's answer to whether it has been solved enough to fund at scale.

What the 3 MW demonstration proves, and what it does not

VEIR says it has demonstrated a 3-megawatt superconducting power delivery system. Read precisely, that is a milestone in packaging rather than in physics.

The hard part of a superconducting installation is not the cable. It is the six other things bolted around it: cryogenic cooling equipment, electrical controls, termination units where the conductor leaves its cold state, and monitoring for a system whose failure modes are unfamiliar to facilities engineers. VEIR's system integrates all of them, and integrating them at three megawatts is genuinely hard.

The gap is upstream. High-temperature superconductor tape is manufactured in a handful of facilities, by a supplier set that is narrow and capacity-constrained. VEIR does not disclose who supplies it, or whether it holds an allocation. A company that cannot secure tape cannot ship systems, and the round funds manufacturing capacity for everything except the one input it does not control.

Three megawatts is roughly the scale of a single data hall's internal distribution requirement, not a campus. It is a credible pilot number. It is not a fleet number.

Who is actually underwriting this

The syndicate is more informative than the headline. Matter Venture Partners and Tyche Partners co-led; LG Technology Ventures, Gates Frontier, Sabanci Climate Ventures and Hui Capital came in new; Engine Ventures, Galvanize Climate Solutions, Piva Capital, Congruent Ventures and VXI Capital followed their money.

Two features stand out. First, LG Technology Ventures is a corporate venture arm buying exposure to a grid problem it faces as an equipment maker. Strategic capital arrives slower than venture capital and leaves later. Second, the cap table carries dedicated climate funds alongside deep-tech specialists, which is the combination a company needs when its economics depend on both a commodity input and an engineering roadmap.

Funding at that percentile is not the constraint. Sequencing is. The company chose to raise a large Series C before signing a named customer, which preserves optionality on valuation and concedes that the schedule now carries the risk.

As we wrote in September, off-grid power for AI data centres became its own asset class when TAR closed a $120 million Series A at a $1 billion valuation. That was generation-side capital chasing the same demand shock. Delivery is the next layer to be repriced.

The two dependencies nobody disclosed

This is where the round deserves a harder look, and where the disclosure thins out.

The first is thermal. A superconducting conductor resists nothing below its critical temperature, and the cooling plant that holds it there consumes continuous power — commonly framed as a fraction of throughput, but never quantified for this system by any source we could find. Over a ten-year asset life, parasitic load is not a rounding error. It compounds into the operating cost that decides whether the installation beats a copper busway it displaces.

The second is materials. Tape supply is the chokepoint that every superconducting power programme faces, and it is concentrated among producers whose capacity decisions are driven by demand from fusion and grid programmes competing for the same lines. A delivery specialist entering late inherits a queue it did not build.

Framing by Nexithon from disclosed round facts; no vendor cost or efficiency figures were published.

Neither dependency is disqualifying. Both are unquantified in public disclosure, which means any valuation built on the 2027 timeline is doing arithmetic with a missing term.

Behind-the-meter is a commercial choice

VEIR's first target is behind-the-meter power, meaning distribution inside the operator's own facility rather than on the utility side of the meter.

That is a deliberate commercial choice with two consequences. It sidesteps the interconnection queue, which as we reported in August is where large AI load projects stall for years. It also concentrates the customer base in operators rich enough, and desperate enough, to pay a premium for density.

The trade is straightforward and it cuts both ways. Fewer customers qualify, each deployment is larger, and the sales cycle runs through a buyer whose capital expenditure committee is currently allocating the largest infrastructure budgets in its history.

First commercial deployments are expected in 2027, a date that rests on a demonstration rather than a signed order. No customer has been named, though the company says it is working with customers and industry partners on AI data centre reference designs. Reference designs are how the industry has historically qualified components that then failed to reach volume orders.

What would make this a wrong bet

Watch the tape. If VEIR announces a supply agreement or a manufacturing partnership upstream, the chokepoint is loosening and the 2027 date becomes credible. If 2027 arrives with a demonstration extended rather than a customer site energised, the round bought a well-engineered system without a market.

And watch the thermal line. A published parasitic-load figure would let any analyst settle the copper comparison directly. Its absence after a $110 million round is the most telling omission in the disclosure.

Will superconducting power delivery reach commercial scale inside AI data centres by the end of 2028?

Development scenarios