The hardest idea in quantum mechanics to hold in a non-physicist’s head is that a thing can be in two states at once. Not flickering between them too quickly to see but both, genuinely, at the same time, until someone measures and it settles into one. That contradiction turns out to be the best way to describe the decades-old argument over the state of quantum computing itself.

Quantum has been a technology that the Nobel Committee, the National Security Agency and half of Silicon Valley agree will rewrite what a computer can do. It is also, simultaneously, a thing so far from finished that it’s been safe to file away along with cold fusion and, until recently, artificial intelligence as technological marvels that are always at least 30 years beyond the horizon.

Read More on Technology

An ordinary computer stores information in bits, which are switches with exactly two states: on or off, one or zero. A quantum computer uses quantum bits, or qubits, which can be both at once. String enough of them together and the number of states they can hold simultaneously grows so fast that a few hundred of them can represent more possibilities than there are atoms in the observable universe. What you could do with a computer that runs on qubits has been a promise that science has been chasing for a very long time.

It has moved from theory into practice. Real-world commercial applications are no longer the stuff of imagination—from high-speed magnetic-levitation trains to quantum-powered navigation systems without GPS, and from better batteries and climate-friendly fertilizer to personalized drugs and precise tornado predictions. The people on the front line of the technology say this is the year it is really starting to happen.

“Useful quantum computing is here right now,” Jay Gambetta, IBM’s director of research, told a packed conference center in Boston this spring.

The Unreachable World

In May 1981, at a conference on the physics of computation co-hosted by MIT and IBM, Richard Feynman stood up and told a room of computer scientists that they were going about it wrong. If you want to understand how the world actually works, he argued, ordinary digital machines will never be enough, because the world does not run on ones and zeros. “Nature isn’t classical, dammit,” he said, “and if you want to make a simulation of nature, you’d better make it quantum mechanical.”

Forty-five years after Feynman’s pronouncement, one of the leaders in the quantum computing field, IBM, and a constellation of partners it has spent a decade recruiting are trying to force the measurement. Nearly a decade of public predictions produced a name for what it would take for this technology to prove itself as something beyond an academic curiosity—quantum advantage—and a year marked on the calendar: 2026.

Put simply, quantum advantage means the point at which a quantum computer solves some real problem—not a rigged puzzle built to showcase the hardware, but something someone needs an answer to—faster, cheaper or more accurately than the best classical computer can. What this means and how to measure it is contested. What is not contested is that the claim is now being tested in public instead of argued over in a seminar room.

Quantum’s strangeness—arguably the whole reason the field exists—has been driving people to distraction since the idea of quantum first surfaced. Albert Einstein spent the last three decades of his life insisting that quantum mechanics could not be a complete explanation for the way the universe operates. The theory said two particles could be linked such that measuring one instantly determined the state of the other, however far apart they were, a result Einstein derided as “spooky action at a distance.” He was wrong, as it turned out, and the experiments confirming he was wrong won a Nobel Prize in 2022. But his objection was not stupidity. It was the reasonable reaction of a person confronting a description of reality that works perfectly and makes no sense.

That is the peculiar condition of quantum mechanics. It is the most successful physical theory ever written: It predicts experimental results to 11 decimal places, and it describes a layer of the world that human beings cannot see, touch or intuit. For most of the last century, we could write down the equations governing molecules, materials and chemical reactions, but not solve them. Quantum mechanics opens up the possibility of solving those equations. Paul Dirac, one of the theory’s architects, said as much in the 1920s: The equations describe essentially all of chemistry and most of biology, and they are too hard to calculate. Everything since has been approximation.

Not all of it stayed on paper. Some quantum effects have already been dragged into the ordinary world, and one of them carries passengers. Magnetic levitation trains—Japan’s test track hit 603 kilometers per hour (375 miles per hour) in 2015—float on magnetic fields generated by superconductors, materials that conduct electricity with zero resistance when cooled far enough. Superconductivity is a quantum phenomenon, one of the few that operates at a scale humans can stand next to. It is also, not coincidentally, what IBM’s quantum chips are made of, chilled in refrigerators to a hundredth of a degree above absolute zero for exactly the same reason.

Changing the Terms

That ambition to reach into what feels like an alien dimension is what pulled Gambetta into the field. His doctorate was in interpretations of quantum mechanics, the unresolved fight over what actually happens at the moment of measurement. “I realized we were arguing mathematics,” he told Newsweek at IBM’s New York headquarters in June. “It was the same mathematical equation, but we were just arguing how to interpret it.” So, he changed the terms. “Rather than arguing the interpretations of the math, let’s see if I can build a quantum computer, and that will at least allow me to answer which interpretation was correct.”

What such a machine could do, if it worked, follows directly from what makes it different. When a conventional computer models a molecule, it cannot represent the thing as it actually is; the mathematics is too large, so it approximates, discarding detail until the problem fits. Michael Biercuk, who founded the quantum software company Q-CTRL, compares the result with a photograph enlarged past its resolution. “When we try to put that very complicated quantum picture into a classical computer, it pixelates,” he said. “You lose information. You’re throwing stuff away to make it fit.”

His other analogy is audio compression: a classical simulation is an MP3 to quantum’s vinyl richness. For most purposes, an MP3 is fine, but for the hardest problems in chemistry and materials science, it is not. A quantum computer, built from the same physics as the thing it is modeling, does not have to discard anything.

The consequences, in theory, land in four broad families. Simulating physical systems—molecules, materials, chemical reactions—is the nearest and the most obvious, because that is the problem Feynman described. Optimization is the second: routing, scheduling, logistics, the class of problems that get impossibly hard as they grow. Machine learning is the third, still the most speculative. And the fourth is differential equations, the mathematics that governs how fluids and heat and air move. Gambetta’s own view is that differential equations are where an ordinary person will first notice quantum computing, because that road leads to fluid dynamics and, from there, to forecasting.

Made concrete, that list is a catalog of things the modern world cannot currently calculate with the desired precision: Better batteries, because you could model what happens inside one rather than inferring it. Fertilizer manufactured without the enormous energy cost of the century-old industrial process, because you could finally simulate how bacteria do it at room temperature. Drugs designed by computing how a molecule will behave in the body instead of testing thousands of candidates to find out. Fusion fuel. Aircraft surfaces. Weather, eventually.

None of which is what much of prior quantum investment has been for. The use case that has drawn the most attention from the government agencies backing quantum is a proof published in 1994 by a Bell Labs mathematician named Peter Shor. It demonstrated that a sufficiently large quantum computer could unravel the encryption protecting essentially all internet traffic, financial records and state secrets. The industry calls the moment that becomes possible Q-Day, and it is still far off.

But, Biercuk says, “It was the promise of that kind of machine that spurred the National Security Agency to start funding IBM, my time at Harvard, and others.” The mission has been to ensure the United States builds a code-breaking quantum computer before an adversary does. Commercial applications were never the objective. “We want to make sure China doesn’t do it first,” he said. “But what can we do in the interim?”

What Counts As Winning?

Quantum advantage sounds like it should mean one thing. In practice, it has led to a lot of academic arguments. The precedent everyone is still arguing over arrived in October 2019, when Google announced it had achieved “quantum supremacy.” Its 53-qubit Sycamore processor had performed a sampling calculation in about 200 seconds that Google estimated would take the world’s fastest supercomputer 10,000 years.

Within days, IBM published a rebuttal arguing the classical estimate was wrong by orders of magnitude, and that a better-designed conventional approach could do the same job in two-and-a-half days. The dispute mattered more than the result, because the problem Google chose had no use to anyone. It was constructed to be hard for classical computers and nothing else.

At the other extreme are the researchers who will accept no claim short of proof that no classical algorithm could ever do better—a standard that can only be failed, never met. Between the rigged benchmark and the impossible proof sits what this year’s claims are measured against: solve a problem someone already pays money to solve, beat the best method currently available and make the result reproducible by anyone renting time on a public cloud.

The first step toward quantum advantage came in 2023, in work with the University of California, Berkeley, with a quantum computer large enough that no conventional supercomputer could brute force a copy of the calculation. IBM called the milestone quantum utility, and the result made the cover of scientific journal Nature that June under the headline “Cutting Through the Noise.”

But utility was not advantage, and a quantum computer being unusually good at being a quantum computer is of interest mainly only to quantum physicists. So, in 2024, Gambetta made a sharper prediction: A quantum machine would beat a classical one at a problem people actually pay to solve by 2026—but only if two communities that had spent years talking past each other started working together. “We were only going to get quantum advantage by 2026,” he said, “if we bring the [high-performance computing] and the quantum communities together.”

What IBM has conspicuously not done is claim the prize itself, and Gambetta was blunt about why. “I think it’s hard for someone that does the hardware to be the one that claims it,” he said, “because then you make it feel like a flagpole moment, which in my opinion has undermined the scientific debate that has to happen.”

The Chorus

Those claims have arrived this year in a rush, from institutions with almost nothing in common except the quantum machines they were using.

Q-CTRL’s claim involved working on the physics of how electrons move through a material—the calculations underneath batteries, solar cells and power transmission, and a category that consumes something like a third of the world’s supercomputer time. The company pushed an IBM processor to 120 qubits and insisted on running everything through IBM’s public cloud so that anyone could repeat it. The best conventional software, on a rented cluster, took as long as 160 hours. The quantum machine took two minutes and 46 seconds.

Then, on a single day at the end of July, three more claims landed at once, and all three were arguing about something other than speed. A team at the University of Chicago, led by the computer scientist Bill Fefferman, used a novel error-correction method to run a problem beyond the practical reach of classical computers with 10 times fewer errors than prior methods—and did it in about 15 minutes.

Qedma, an Israeli company whose founders include Dorit Aharonov, one of the theorists who established decades ago that quantum error correction could work at all, benchmarked its results directly against Fugaku, a supercomputer in Kobe, Japan, and one of the fastest on earth, and found that several conventional methods disagreed with each other while the quantum results held steady.

Algorithmiq, based in Milan, posted its solution to a materials science problem using quantum computing to IBM’s advantage tracker—a scoreboard that sorts claims into those still standing, those a classical method later matched and those used for baselines—and after eight months, no classical computing method has been able to reproduce it.

What unites them is not that quantum computers went faster. It is that they started producing answers you could check. “Verification remains one of the biggest challenges in firmly establishing experimental quantum advantage,” Fefferman said, which is a polite description of what went wrong in 2019 when Google’s announcement of quantum supremacy was met with doubts about how much better its solution was than classical computing.

Not everyone with quantum breakthroughs is wading into the debate over what is or isn’t quantum advantage. Lara Jehi, the Cleveland Clinic’s chief research information officer, declined to describe her own institution’s work as quantum advantage at all. “The value of quantum is not by exceeding classical,” she told Newsweek. “It is not a race of one versus the other. It’s how do we get farther by riding these two horses together, like in a carriage.” Her preferred analogy is not a faster car but a different vehicle. No matter how much sleeker and quicker you make a car, she said, it will never become a plane. And getting anywhere that matters means using both—you drive to the airport, you fly and you drive again at the other end.

Using a hybrid quantum-classical supercomputing approach, Cleveland Clinic staff scientist Kenneth Merz spent 18 months running past a milestone the field expected to take five to seven years. In October 2024, the largest molecule his team could simulate on a quantum computer was 10 atoms. By last December it was a real protein at 300. By this May it was more than 12,600—a protein bound to a drug-like molecule and suspended in water, which is the condition proteins actually exist in, and the condition that makes drug discovery hard. Merz thinks he undersold it. “We thought we were really going big,” he said at a conference this spring in Boston, “but we quickly realized we could have been two or three times larger.”

Similarly, Cleveland Clinic worked with Oak Ridge National Laboratory, the largest open-science laboratory in the Department of Energy system, on the chemistry of fusion fuel—specifically, use of a molten salt blanket inside a reactor to breed tritium, of which only a few pounds exist naturally on Earth. Sarp Oral, who leads the lab’s advanced technology section, is less interested in the fusion result than in the apparatus that produced it. “This workflow is a template,” he said, “that will not only transfer to fusion energy, but also will let us apply the same template to different scientific problems, like drug discovery, like climate.” He compared it to building a telescope: one instrument, pointed at whatever question comes next.

Jerry Chow, IBM’s chief technology officer for quantum-centric supercomputing, had told Newsweek in June that more results were coming in mid-July, and that the shape of the thing mattered more than any single one. “It’s hard to pick any one of them and say, ‘That one,’” he said. “It’s the fact that we have the capability for this group to exist, to be pushing all at the same time. That is the story itself.”

What IBM Is Betting

IBM’s bet on quantum has been years in the making, but the stakes are now considerably larger than they were six months ago. On July 14, the storied technology company suffered its worst day on the market in its history when its stock fell 25 percent in a single session, after the company warned that quarterly results would miss badly because firms were redirecting spending from the mainframes and software that IBM sells to purchasing their own AI infrastructure. “Numerous large deals failed to close on the timelines we expected,” chief executive Arvind Krishna wrote to investors.

Two weeks later, on the same day three of his partners announced quantum advantage, Krishna went on CNBC and put a number on the other bet. Quantum would have “a measurable impact on our top line and bottom line” by 2028 or 2029, he said, and the market for it would be worth far more than that: “By the end of the 2030s, we are now pretty convinced this is a trillion dollars of value.” His case for why now was the same one his partners had spent the summer making. A quantum computer, he said, “can do things better, faster, cheaper, in a way that normal classical computers cannot do at this time.”

That has made the roadmap IBM has set out for itself even more crucial. There are two more big dates on it. In 2029, IBM intends to deliver a machine code-named Starling, built around error correction at scale. The distinction is not speed. Today’s quantum computers can sustain roughly a thousand operations before accumulated errors ruin the calculation; Starling is designed for a hundred million. That is not a faster version of the current machine, it is a different kind of machine, and everything IBM has said publicly about the 2030s—a data-center-scale system called Blue Jay—depends on getting there.

Gambetta is unembarrassed about what he wants IBM to become along the way. “Our desire is to build the infrastructure like Nvidia is for GPUs [graphics processing units],” he said. “So, I would like IBM to be the Nvidia for QPUs.”

The model is deliberate. Graphics processors were built to render video games but went on to become the engine of the artificial intelligence boom because a generation of researchers—pushed hardest by AI pioneer Geoffrey Hinton, who spent years urging the machine-learning field onto hardware designed for something else entirely—discovered they were extraordinarily good at the mathematics underneath neural networks.

Gambetta has his own version of that history. He was programming graphics cards to solve physics equations as a graduate student, before anyone outside a game studio cared. “I would say the history of Nvidia and the choices they made have inspired us,” he said, though he expects a different outcome—quantum, conventional and AI processors working alongside each other rather than one displacing the rest.

On May 21, the Commerce Department announced that the United States would take equity stakes in nine quantum computing companies as part of roughly $2 billion in CHIPS Act grants. IBM’s share, $1 billion, was the largest single allocation, matched by about $1 billion of IBM’s own money and directed at a facility called Anderon in Albany, New York—the first factory in the country purpose-built to manufacture quantum chips, running the same 300-millimeter wafer lines the semiconductor industry uses, adapted for superconducting qubits. Seventy-five miles south, IBM is expanding its Poughkeepsie campus with a plant to assemble the Starling systems themselves.

The Believers

Strip away the hardware and the money and what remains is a small number of people who staked careers on an argument most of their colleagues thought was premature.

Chow’s career began in a hallway. His father was a physics professor; as a Harvard sophomore he read a paper about quantum dots—the nanoscale crystals that glow under electricity and now sit in television screens—and walked to the author’s office and knocked. “I said, ‘Hi, I read a paper about some of the work you do. I’d love to learn more,’” he remembers. “He invited me in right at that moment.” Chow built hardware at Yale and then IBM before taking the job he holds now, which is less about physics than about persuading three incompatible kinds of processors to cooperate.

He passes along a line from Charles Bennett, an IBM fellow who shared the 2025 Turing Award (often described as computer science’s Nobel Prize) with Gilles Brassard for their contributions to quantum cryptography. Instead of thinking of quantum computers as exotic upgrades of classical computers, Bennett’s formulation is the reverse: “A classical computer is a quantum computer handicapped.” What that means, Chow said, is that “the most fundamental computer they can build is a quantum computer.”

Chow does not hedge. Asked if quantum is, finally, really coming, he said, “I mean, we are saying right now that it’s here. Quantum is now, quantum is real.”

Biercuk’s career is evidence that quantum is not one story. Part of Q-CTRL’s business is quantum computing. The other part is quantum-powered navigation systems now under contract with Airbus, Lockheed Martin and the Defense Department, that let aircraft and ships determine position without GPS. Satellite navigation is jammed or spoofed constantly in contested airspace—“a thousand times a day” over the Persian Gulf alone, Biercuk said—and every rerouted flight costs an airline thousands of dollars.

And then there is where the Cleveland Clinic put its quantum computer, which was Jehi’s decision. It sits in the cafeteria. The engineering allowed for it, but that was not the reason. “There was an inspiration reason,” she said, “which is it had to be there, in their faces, so they think about it.” The clinic signed its partnership with IBM in 2021, the year the institution turned 100, when nobody could say whether any of this would work. Five years on, with the protein results published, Jehi still describes the decision as “a calculated risk.”

There may never be a single moment when the world agrees quantum advantage has been reached. Instead, it will be the growing accumulation of results, and then one day the arguing will stop, and the date on the pile will be somewhere in the recent past. Which is to say, the measurement may not have been taken just yet. The thing remains unsettled, for now, in two states at once—a technology that has already changed what is computable, and a promise that still has to be believed. Gambetta thinks he knows what the consensus will be. “I think in two years we’ll look back and say 2026 is when it happened.”