What’s a quantum computer good for, anyway?

The 21st-century fervor about building the first industrial-scale quantum computer, pioneering theoretical physicist Peter Zoller says, is akin to the 20th-century obsession with becoming the first to conquer Mount Everest. “When you’re climbing, you look around worrying, ‘Who is number one?’” he says. “When you reach the top, that’s when you ask yourself, ‘Why the hell did we actually do this?’”

In 1995 Zoller and Ignacio Cirac, then a postdoctoral researcher in Zoller’s group at the University of Colorado Boulder, proposed the first realistic blueprints for a quantum computer. Their idea was to use trapped ions as “qubits”—the quantum equivalent of digital bits, able to exist in a superposition that simultaneously represents 0, 1 and all positions in between. More than a decade earlier physicists Paul Benioff and Richard Feynman had independently suggested that machines harnessing the quantum realm’s weirdness could, in theory, outperform classical computers at some tasks.

Today teams around the world are developing ever bigger quantum processors using qubits made from ions, neutral atoms, superconducting loops, and more. IBM and Berkeley, Calif.–based company Atom Computing currently lead the charge with quantum computers hosting more than 1,000 qubits, and last year a research group at the California Institute of Technology reported that it had built a record-breaking array of more than 6,000 qubits.

On supporting science journalism

If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.

“It’s an exciting time because people are fielding quantum computers with hundreds and thousands of qubits,” says Nobel Prize–winning quantum physicist John Martinis, a professor emeritus at the University of California, Santa Barbara, and co-founder of quantum hardware company Qolab.

In 2019 Google researchers, led by Martinis, reported that their 53-qubit processor, Sycamore, had become the first to achieve “quantum advantage,” performing a calculation in 200 seconds that they estimated would take the best classical supercomputers around 10,000 years to solve. That supercomputer number was later disputed—IBM argued that its best classical computer could actually perform the task in just two and half days—but even if it held up, the calculation was of only academic interest as a proof of principle. “Google’s 2019 demonstration of quantum [advantage] was an important milestone, but many people would say that it did not yet constitute a breakthrough on a problem of broad practical significance,” notes quantum physicist Kihwan Kim, now at the Institute for Basic Science in South Korea.

“We can now simulate things like superconductivity, artificial photosynthesis and small drug designs.” —Michelle Simmons, Silicon Quantum Computing

Experts agree that to tackle useful problems that lie beyond the reach of even the best possible classical supercomputers, we need qubit numbers to jump significantly, potentially to a million or more. In addition, quantum physicists will need to engineer robust qubits that maintain their quantum properties for a longer time, and they’ll have to find ways to fix errors introduced during calculations. In March a multi-institutional group of researchers reported that IBM’s superconducting Heron processors could accurately predict the results of neutron-scattering experiments that measured the structure of a specific antiferromagnetic crystal to an unprecedented scale, using 50 qubits or fewer; the physicists noted, however, that classical computers could perform the same feat faster and more accurately.

So what will quantum computers really be good for—and when? Experts say we are still years away from quantum computers able to handle practical applications that classical computers cannot, which might include breaking common data-encryption schemes, simulating quantum processes for fundamental physics, and designing better drugs and materials. That said, Martinis—an expert on scaling up quantum hardware—notes there are no guarantees that million-qubit computers will ever be created. “The proof,” he says, “will be building them and seeing that they work.”

Cryptography

The most notorious promise of quantum computers has been that they will one day break RSA encryption, a long-standing protocol used worldwide to secure bank transfers, cryptocurrencies and digital communication. That day may come surprisingly soon. It was long thought that cracking encryption would require a processor with at least a million qubits. But in February a team from Iceberg Quantum in Sydney, Australia, dramatically reduced that estimate, calculating that with careful optimization and error correction, hackers might need fewer than 100,000 qubits for the feat. In March, Google announced a new commitment to migrating its systems by 2029 to protect them from quantum hacking.

Although the Iceberg claims have yet to be peer-reviewed, they are credible and have caused a stir, says Artur Ekert, a cryptography expert at the University of Oxford. “Many think that the threat to encryption from quantum computers is just mumbo jumbo—and I have also been skeptical—but it could just take a few more papers like this one to make the notion of breaking RSA relevant,” he says. Martinis’s opinion has also shifted in recent years. “If you are worried about RSA encryption—as you should be—I would say it might be broken in five to 10 years,” he says.

RSA encryption leverages the fact that it’s easy to create a secret key by multiplying two large prime numbers together but effectively impossible for any classical computer to determine the key by efficiently factoring it back into those constituent primes. A classical computer could test successive numbers sequentially, remembering each value and looking for a pattern, but this approach is infeasible with large numbers, Ekert explains. Modern classical algorithms use other methods but remain inefficient because the execution time increases exponentially with the size of the number to be factored.

The quantum world, however, is not so constrained. Qubits can take on multiple values simultaneously and become entangled with one another, amplifying their power. “Essentially it can cover all possible computational paths at the same time,” Ekert says. In 1994 theoretical computer scientist Peter Shor, now at the Massachusetts Institute of Technology, proposed that a hypothetical quantum computer could use this property to crack RSA encryption. If and when one does, it will probably use the algorithm Shor developed.

There are proposals for quantum-resistant cryptographic algorithms; the U.S. National Institute of Standards and Technology (NIST) published three such schemes in 2024. Zoller thinks such work largely defuses the threat because it suggests the world can move away from RSA encryption before quantum hackers arrive. “Shor’s algorithm may ultimately be remembered as a landmark scientific achievement of great historical importance for inspiring the development of quantum computers, as much as for its implications for breaking encryption,” he says.

Ekert feels less reassured. Last year, he notes, a computer scientist in China proposed—it turns out incorrectly—a quantum algorithm capable of breaking NIST’s top candidate, which is called lattice-based encryption. “It took the brainpower of the whole quantum cryptography community [more than a week] to find a mistake, showing you how close those things are,” Ekert says. “Maybe next time it will be correct.”

Fundamental Physics

One domain where quantum processors are already seeing success is in modeling particle interactions to solve mysteries at the heart of fundamental physics. “It goes back to Feynman, who articulated that you can’t really understand how nature works unless you build it at the same length scale,” says quantum physicist and material scientist Michelle Simmons, founder and CEO of Silicon Quantum Computing (SQC) in Sydney.

Simulating interactions of multiple particles rapidly becomes impossible with a classical computer, explains Daniel González-Cuadra, a quantum physicist at the Institute for Theoretical Physics in Austria. “The information that it takes to describe the state of these systems grows exponentially with the size of the system, and at some point you just don’t have enough memory,” he says. Capturing all the complexities requires an equally complex quantum machine.

Groups around the world are making progress on this front. One big area of focus is increasing the “coherence” of qubits so they stay in superposition long enough to carry out their calculations. In 2021 Kim’s China-based team demonstrated that trapped-ion qubits can maintain coherence for more than an hour, which he says was “a very important benchmark for scaling up meaningful quantum simulations.”

Last year two teams independently published real-time quantum simulations of the creation of matter and antimatter during a process called string breaking. According to the Standard Model of particle physics, pairs of strongly interacting subatomic particles, such as quarks, behave as though they are joined by an elastic string, “like a violin string that vibrates,” explains quantum physicist Pedram Roushan of Google Quantum AI in Santa Barbara. Roushan’s team ran its simulation on Google’s Sycamore chip, which uses superconducting loops as qubits. The simulation showed how pulling two particles apart increases the string’s tension until it finally snaps, releasing the stored energy by generating a new pair of matter and antimatter particles. “These theoretical concepts were known since the 1970s, but we were able to visualize them and take a picture of the strings and their breaking,” Roushan says.

The Sycamore experiment was an example of a digital simulation, meaning it was performed on a multipurpose chip with circuits of qubits designed to do many different tasks. In contrast, González-Cuadra, Zoller and their colleagues worked with a team at QuEra Computing in Boston to create an analog simulator—a lattice of neutral-atom qubits specially built to simulate string breaking. The two string-breaking simulations are among the first to model particle interactions in two spatial dimensions, González-Cuadra says. “The physics is richer, so we could see how these strings fluctuate,” he explains.

“We’re optimistic that we’re going to see the first practical applications in five years.” —Sergio Boixo, Google Quantum AI

These kinds of simulations won’t replace particle physics experiments. Instead they will help physicists hone their theories and make testable predictions that can be checked at particle accelerators. And so far they have simulated only simple models that can also be checked with classical computers. But González-Cuadra believes quantum simulators will start to surpass their classical counterparts in a couple of years, marking an era of true quantum advantage. This possibility raises the question of how physicists can be sure their quantum simulations are spitting out reliable results. To answer it, last year Zoller and his colleagues posted a preprint paper on arXiv.org describing a strategy for developing an analog quantum machine that not only makes predictions but also quantifies the uncertainty in those predictions. “If you ask me what the big challenge for quantum simulation is, it is the frontier of verification,” Zoller says.

Materials Design

The dream, Zoller says, is for quantum simulators to shift from a passive “discovery mode”—in which they are used to model nature—to an “active design mode” in which quantum computers would spit out recipes for synthesizing new molecular structures with specific desirable properties. The quantum engineering of new molecules could lead to better drugs and to batteries that don’t use costly, environmentally damaging commodities such as rare earth elements. “These things take billions of dollars, so if you just make something even a few percent cheaper or a few percent better, then that’s really worth it,” Martinis says. “It could be a huge thing not just monetarily but for changing how things are built to make them more ecological.”

Room-temperature superconductivity is one high-priority target. Superconductivity, the free flow of electricity without resistance, typically requires a material to be cooled to extremely low temperatures, which makes it impractical for many applications. But certain materials exhibit the phenomenon at higher temperatures, and some researchers hope quantum engineering can help them find new superconducting materials that don’t need any cooling at all. “These systems consist of 10²³ particles, whereas classically we can model only about 100 particles,” says Henrik Dreyer, a quantum physicist at Quantinuum in Munich. Physicists would need to reduce the error rates in quantum processors to just one in a million to make it possible; at the moment the best chips are down to slightly below one in 1,000, Dreyer explains.

Dreyer and his colleagues have been performing digital simulations of cuprate superconductors using Quantinuum’s Helios chip, which employs 98 trapped-ion qubits. In carefully controlled laboratory conditions, shooting these materials with a laser can very briefly—and surprisingly—create a superconducting state at a relatively high temperature. “The first question is: Why?” Dreyer says. Last year Quantinuum posted a preprint on arXiv saying its two-dimensional simulation modeling the material shows that under laser fire, its electrons pair up—a condition necessary for flow without resistance. “The ultimate question is,” as posed by Dreyer, “Can we engineer it to do this at room temperature for a minute, an hour, 10 days, or more?”

Meanwhile Simmons and her SQC colleagues in Australia have developed a simulation system called Quantum Twins—a 2D array of 15,000 clusters of phosphorus atoms embedded in silicon—to create analogs of various materials. In February the team reported that the platform can simulate the transition between insulating behavior and metallic conduction. “We can now start to simulate things like superconductivity, different battery materials, artificial photosynthesis and small drug designs,” Simmons says.

Google Quantum AI’s Sergio Boixo notes that the company has collaborated with BASF on battery design, Sandia National Laboratories in Albuquerque on fusion energy, and German chemical company Covestro on pharmaceutical development. Last year it implemented an algorithm for modeling molecular structure on Willow, Google’s 105-qubit superconducting processor, that can be used in combination with nuclear magnetic spectroscopy. The technique, which works by bouncing signals onto qubits and effectively listening for their echoes, runs 13,000 times faster on Willow than an equivalent algorithm would on the best classical supercomputer. One important aspect of the algorithm’s design is that it allows results to be corroborated by another quantum machine. “Quantum Echoes is the world’s first quantum-verifiable algorithm with quantum advantage,” Boixo says. “We’re optimistic that we’re going to see the first practical applications in five years.”

Quantum AI

If you really want to generate hype, combine the word “quantum” with “AI,” jokes Jacob Biamonte, an expert on quantum machine learning at ÉTS Montreal. Indeed, as quantum processors get bigger, some physicists are focusing on using them to boost the performance and energy efficiency of classical artificial intelligence.

Last year SQC launched Watermelon, a quantum-enhanced AI processor, to help speed up machine learning. Classical AI systems are already adept at finding patterns in vast datasets, which makes them particularly useful for optimizing communications and energy networks, for instance. SQC’s quantum technique builds on classical reservoir computing, a method for taking input data points and transforming them onto a higher-dimensional neural network, making it easier to find patterns. In 2017 scientists in Japan predicted that the classical nodes of the neural network could be replaced by a smaller number of qubits subject to quantum interference. “The advantage of having a quantum reservoir is that you get an exponential increase in dimensionality,” Simmons says, enabling a quantum reservoir to achieve the same training results as a classical reservoir but potentially faster and using fewer resources.

Watermelon’s first commercial trial—in collaboration with Australian telecommunications company Telstra—has shown promising results. Telstra already uses AI to monitor latency and bandwidth patterns on its networks. It takes about three weeks to train the company’s models using standard classical methods. With Watermelon’s help, Telstra achieved the same training results in just two days. “In the grand scheme of the world, that is quite significant because at the moment, data centers are very power hungry,” Simmons says, noting that similar optimizations could be rapidly rolled out to other energy-intensive tasks, such as training AIs for image recognition, fraud detection and market prediction. “I feel like I’m in this freight train that’s going at superhigh speeds,” she says.

Ekert, however, remains cautious about the longer-term benefits of using quantum AI processors to analyze classical datasets. “Turning classical data into a quantum form is terribly inefficient,” he says. Where quantum computing and machine learning are already being combined most helpfully, Ekert argues, is in physicists’ use of classical AI to design quantum error-correcting codes and better quantum hardware. Last year, for example, Finnish company QMill launched a classical AI service for compressing quantum circuits, reducing the number of gates needed for operation by 20 to 50 percent. Biamonte also thinks the current vision is too small. “If the goal is to use quantum computers to do machine learning for classical data, it doesn’t even make sense, because classical machine learning is already so good,” he says.

If quantum processors could one day be used to analyze quantum data directly, however, that would be a game changer. “There should be these wonderful patterns that classical computers cannot detect because there are just too many data for their memory,” Biamonte says. A quantum AI could riff on the molecular structure of an existing patented drug, for instance, to generate multiple different configurations with the same benefits. It could then assess those molecules to see whether they could be synthesized and patented before a company committed funds to trying to make them. “That’s the exciting future that doesn’t exist yet,” Biamonte says.