Quantum computing’s promoters talk about qubits. The power company will want to talk about refrigerators. 

Exhaustive  pages have been written about what AI data centers will demand from the grid at scale, but little is out there on what quantum computers will demand at scale.

And while quantum has been five years away for thirty years (to quote the great Molly Wood), it may finally be getting closer. On Oct. 7 the Defense Advanced Research Projects Agency moved Atom Computing, Diraq, IBM (IBM: NYSE) and IonQ (IONQ: NYSE) into the final stage of its Quantum Benchmarking Initiative, where they join Microsoft (MSFT: NASDAQ) and PsiQuantum. DARPA made a big deal of it, but none of the six will meet the program’s test, “utility-scale operation – meaning its computational value exceeds its cost – by the year 2033,” without power infrastructure built for quantum.

The AI power debate is about size. Data centers used 1.5% of the world’s electricity in 2024, and the International Energy Agency expects that to more than double by 2030. The arguments are over megawatts, GPU racks and places in the queue for a grid connection. Quantum brings a different kind of load. It runs constantly. It costs a great deal of electricity for each unit of useful work. Some designs depend on helium and on refrigerators that take a year or more to arrive. Regulators drafting rules for AI data centers want those loads to cut back when the grid is short. A dilution refrigerator that warms up takes 24 to 48 hours to cool down again, so it cannot cut back.

The qubits themselves use almost no power. The energy goes to the equipment that keeps them cold and keep them quiet. A a great deal differs on the design of the quantum machine. 

That means refrigeration, lasers and vacuum pumps, control electronics and the conventional computers that correct the machine’s errors as it runs.

Cold is the most expensive concern. Thermodynamics sets a floor: removing one watt of heat at 4 kelvin from a room at 300 kelvin takes at least 74 watts of electricity, and real refrigerators do worse. At millikelvin temperatures, the range where most quantum chips run, the ratio climbs much higher. Every control wire that runs into the cold stage carries heat with it, so the cooling bill rises with the number of qubits. 

In a paper posted in May (not yet peer reviewed) 16 researchers, including Raja Yehia of the Institute of Photonic Sciences in Barcelona, found that most of the energy today’s machines use goes to “maintaining a stable qubit environment, through cooling or temperature control.” They warned that “should current architectures scale up without new fundamental discoveries and innovative engineering, the consumption related to qubit control electronics will rapidly outgrow this cost.”

The equipment is still bespoke. For example, the dilution refrigerators most designs use depend on helium-3, a rare isotope that comes largely from nuclear decay. So cryogenic refrigerators come with a 12- to 18-month lead time. 

Today’s machines are small loads – so far. Researchers at NASA and Google  (GOOG: NASDAQ) estimated in 2019 that a 53-qubit superconducting processor drew 15 kilowatts. QuEra has said its 256-qubit neutral-atom machine draws less than 7 kilowatts. 

Useful machines will be much larger, and nobody knows by how much. Ezratty said in a presentation last year that six designs under development, each scaled to 4,000 logical qubits, would draw from under 1 megawatt to over 100 megawatts. A logical qubit is an error-corrected unit built from many physical qubits. His estimate was not peer reviewed. A peer-reviewed study led by David McCollum of Oak Ridge National Laboratory, which modeled superconducting machines, projected that integrated quantum and classical data centers in the 2040s would draw power of an order of magnitude similar to today’s data centers. The authors wrote that water and helium-3, more than electricity, could limit how far quantum computing grows. For a sense of what industrial cold costs, CERN’s cryogenic system uses 40 megawatts of electricity to keep the Large Hadron Collider’s magnets at 1.9 kelvin.

AI campuses are now planned in gigawatts, so quantum will add a new load to the grid rather than compete with AI for power, unless machines become very large or very numerous. The open question is whether quantum replaces some GPU work or adds to it.

This nascent technology seems to be coalescing around seven types of design, each with different power characteristics 

Superconducting. IBM, Google, a unit of Alphabet (GOOGL: NASDAQ), Rigetti Computing (RGTI: NASDAQ) and D-Wave Quantum (QBTS: NASDAQ) build qubits from circuits that lose all electrical resistance near absolute zero. They are least noisy at 10 to 35 millikelvin, inside dilution refrigerators, and their wiring heat load grows as they add qubits.

Topological. Microsoft’s Majorana 1 chip sits in a dilution refrigerator that keeps its qubits “much colder than outer space,” the company said. Microsoft said the machine is designed to be deployed inside its Azure data centers.

Phosphorus atoms in silicon. Silicon Quantum Computing places single phosphorus atoms in silicon, one at a time. Its engineers test processors at millikelvin temperatures, and one said in a case study published by refrigerator maker Bluefors that SQC’s “commercial scale quantum computer fits into one dilution fridge.” Nisewaner said at the panel that SQC’s first product, Watermelon, runs at 3 kelvin. I could not confirm that figure.

Silicon chips. Diraq builds qubits in silicon using processes compatible with existing chip fabrication. It said in March 2024 that its processors had operated above 1 kelvin, 20 times warmer than previously demonstrated, which it said cuts the cost and complexity of refrigeration without eliminating it.

Photonic. PsiQuantum computes with particles of light, but its chips still need cold. The company says its systems run at temperatures more than 100 times higher than superconducting qubits require, cooled by liquid helium in cabinets that each supply 30 watts of cooling. Its first utility-scale site will hold one of the largest cryogenic plants built for quantum computing, ordered from Linde (LIN: NASDAQ) and due in the second half of 2027.

Trapped ion. IonQ and Quantinuum (QNT: NASDAQ) hold charged atoms in electric fields inside a vacuum, with no dilution refrigerator. IonQ said its newest system, Superion 256, occupies a standard server rack, draws less power than a rack of GPUs and has a cooling system that fits typical data centers. It has published no power figures for the system. Some trapped-ion designs add cryogenic cooling, according to Physics Today.

Neutral atom. Atom Computing, Infleqtion (INFQ: NYSE), QuEra and Pasqal (PSQL: NASDAQ) hold uncharged atoms in beams of laser light. Their power goes to high-powered lasers, vacuum systems and electronics, with no dilution refrigerator.

Four of DARPA’s six finalists use cryogenic refrigeration: IBM, Microsoft, Diraq and PsiQuantum. The cold-power problem becomes the grid’s problem only after one of them wins.

DARPA’s Quantum Benchmarking Initiative runs in three stages. In the first, companies had to show a plausible design for a machine whose computational value exceeds its cost by 2033. In the second, which lasts about a year, DARPA reviews each company’s detailed research plan, prototype roadmap and manufacturing plan. In the third, DARPA tests the company’s hardware independently, with awards that can reach $300 million. 

Australia’s Silicon Quantum Computing was one of 11 companies DARPA moved into the second stage in November. It was not among the four promoted to the third stage on Oct. 7, but DARPA is promoting companies in waves and has said more will follow. 

Silicon Quantum Computing was founded in Sydney in 2017, and few U.S. investors have looked at it. It is private, and its biggest backer is the Australian government: the National Reconstruction Fund Corp. has committed A$60 million, including A$40 million in June, when London’s Firgun Ventures also invested. 

Unlike most quantum chipmakers, Silicon Quantum Computing makes, assembles and tests its chips in-house rather than at an outside foundry. It places each phosphorus atom to within 0.13 nanometers and can design, build and test a new chip in under a week, the company says. Its chief executive, Michelle Simmons, is a Scientia Professor at the University of New South Wales, where she directs the Centre of Excellence for Quantum Computation and Communication Technology.

Its first commercial product, “Watermelon”,  is already on the market. In a trial with Schneider Electric (SU: EPA), hybrid models built with it improved next-day energy forecasting accuracy by an average of 20% over classical baselines, with peak gains of 41%, the two companies said. SQC targets a full-scale machine in 2032 or 2033, and its claim that the machine fits in one dilution refrigerator puts a ceiling on how much cooling each one needs.

Their advanced approach bowled over venture investor Bill Tai, whom I interviewed as part of a panel during San Francisco’s Tech Week last week “I’ve looked at a bunch of these things as a venture guy for thirty years, and I was just not interested,” Tai told me. “[Silicon Quantum Computing] seems like the most scalable, reliable, reproducible at scale, taking advantage of the trillions of dollars that have been invested in semiconductors for forty years.”

That’s led Tai to invest in the power to run quantum machines and the AI data centers beside them. He’s focused on Western Australia. 

The country is nearly as large as the contiguous United States and has 27.9 million people, against 341.8 million in the United States. Western Australia covers a third of its land and has 3. 1 million residents. That leaves lots of room for power generation.

Sovereign Green Compute, a Perth company Tai chairs, describes itself as an origination and governance platform for Australia’s AI infrastructure buildout, lining up power, land, connectivity and partnerships with Traditional Owners. The company has secured a 700-megawatt site that will draw on solar and wind with natural gas as backup.

His fund group, a nonprofit community of athletes, conservationists, technologists, artists and innovators spends some important time in Perth

From that vantage point, this look different. The Australian Energy Market Operator’s main grid runs separately from the national market. AEMO forecasts that minimum demand on that grid could fall below zero, to minus 93 megawatts, by 2035-36, as rooftop and other small-scale solar grows from 3,100 megawatts to 6,000 megawatts. That could well power a whole lot of quantum computer – a short fiber optic run to Singapore and then the rest of Asia and the world.

To be sure, if a trapped-ion or neutral-atom design wins, quantum’s power problem shrinks to lasers and electronics. Western Australia is far from most users, and much of the case for its power comes from people building businesses on it. SGC was formed last fall, Tai said.

DARPA’s test will settle part of the question. A machine whose “computational value exceeds its cost” by 2033 will have to count its electricity, and the six finalists draw it in very different ways.

I joined Nicole Petallides on Schwab Network to talk about where quantum computing stands headed into the last quarter of the year. With revenues of $80.1 million, IonQ (IONQ: NYSE) was more than the other four listed hardware companies combined, and that it came with a $337.2 million operating loss. We talked about IonQ’s claim that its Superion 256 fits a standard server rack, but I probably should’ve pointed out that the claim is unverified because the company has published little by way of power, cooling or fidelity figures.

WATCH THE SCHWAB NETWORK HIT HERE →

I joined Roger Peterson on Bloomberg Markets to talk about what we know and what we still don’t know about Anthropic after Reuters got the first look at its IPO prospectus. There are so many questions that could’ve been answered. Safe to say news of 2026 revenue at $4.6 billion yielding a net loss of about $42 billion and $518 billion of compute obligations wasn’t enough for me.

WATCH THE BLOOMBERG HIT HERE →

In San Francisco all week. I’m available for interviews, background or client calls — and preparing for IPO filing from Anthropic! 👆

Written reports are available to clients, our IPO reports (dubbed “The Quiet Period”) for $5,000. Video summaries on YouTube, and of course our popular summaries of the summaries on Instagram, TikTok and YouTube Shorts.

^{We certify that (1) the views expressed in this report accurately reflect our views about all of the subject companies and securities and (2) no part of my compensation was, is or will be directly related to the specific recommendations or views expressed in this report.}

^{Important disclosures}

^{Important disclosures are available by calling (347) 619-2489 or writing to Epistrophy Capital Research, One Ferry Building, Suite 201, San Francisco, CA 94111.}

^{Epistrophy Capital Research is an independent research provider and does not operate as a financial institution. Epistrophy explicitly does not provide investment advice, stock recommendations, or solicitations to buy or sell any securities.}

^{The research reports provided by Epistrophy Capital Research contain opinions derived from publicly available information, issuer communications, recognized statistical services, and other reputable sources considered reliable. However, Epistrophy does not independently verify the accuracy or completeness of such information and explicitly disclaims responsibility for any errors or omissions.}

^{Opinions and analysis contained within Epistrophy's research reports are current only at the time of publication and are subject to change without notice. Readers must independently verify facts and conduct their own due diligence before making investment decisions.}

^{Epistrophy Capital Research does not consider or evaluate individual investor circumstances, including investment objectives, financial situations, or risk tolerance. Investing in securities, particularly small-cap and micro-cap stocks, involves substantial risks, including significant volatility and potential loss of principal. Readers are strongly advised to consult their financial advisor or another qualified professional before acting on any information provided. Readers should assume Epistrophy Capital Research, its principals or its contributors may have positions, long or short, any of the companies discussed and the Epistrophy Capital Research principals or contributors may have had or currently have business interests in the companies discussed.}

^{Past performance referenced in Epistrophy reports is not indicative of future results. Security prices can fluctuate widely and investors should be aware that investments can result in significant financial losses. Unless explicitly stated otherwise, prices quoted in reports reflect market closing prices from the previous trading day.}

^{Epistrophy Capital Research publications are intended solely for direct recipients and should not be redistributed or shared with third parties without explicit permission from Epistrophy Capital Research LLC.}

^{Epistrophy Capital Research reports are provided strictly for informational purposes and do not constitute a comprehensive analysis of any company, security, or industry. No content within these reports should be considered accounting, tax, legal, or professional advice.}

^{Links or references to third-party websites or external resources are provided solely for informational convenience. Epistrophy Capital Research expressly disclaims endorsement of, and responsibility for, the content, accuracy, or reliability of such external information. Accessing third-party information is done entirely at the user's own risk.}

^{For additional details, clarification, or specific inquiries regarding Epistrophy Capital Research reports, please contact Epistrophy Capital Research LLC directly}