Within IBM Quantum, the debate has shifted from theory to proof. As hardware scale, error rates, and algorithms begin to converge, the challenge is no longer whether quantum computing works in principle but where it delivers credible separation and where it doesn’t.
In this interview, Oliver Dial, CTO of IBM Quantum, explains why chemistry and materials are emerging first, why hybrid models matter more than standalone machines, and why the next phase of progress depends as much on how quantum is used as on how it is built.
What is making quantum advantage around 2026 possible?
It’s really the convergence of three things. One is that the hardware is getting better. The main limitations to quantum hardware are the number of qubits and the error rate, because the error rates are quite high compared to classical computers. If the number of qubits is too small, it’s easy for a classical computer to simulate the hardware. If you can simulate my quantum algorithm on your phone, it’s probably not a quantum advantage.
Similarly, if the error rate is too high, there’s a different set of techniques you can use to model the hardware classically. Around 2023, the quantum hardware got both big enough and low enough in error rate to reach the point where you couldn’t simulate it directly using classical methods.
But you need to go a little beyond that, because the algorithms you want to use for advantage don’t map perfectly efficiently onto the hardware. There’s overhead. That leads to the second thing: algorithmic improvements. This includes finding algorithms where there’s the biggest possible difference between performance on quantum hardware and classical hardware. From that, you realise quantum advantage. It’s really a scientific question. You’re not looking at algorithms that are directly useful in a business sense; you’re looking for places where you can show separation.
There have also been advances in how we operate the hardware so we can get more accurate answers out of the same machines. When you put all this together, there are two general things that need to happen to get to quantum advantage. One is continuing to improve the infrastructure. The other is having clients and partners who are experts in applications and algorithms to use the hardware to demonstrate advantage.
There’s a third element as well. Advantage is defined as outperforming classical computing, so you also need an expert in classical simulation techniques. You need a head-to-head comparison between the best quantum techniques and the best classical techniques on the best possible infrastructure.
2025 was our rallying call: The hardware was good enough, and it was time to build these groups of algorithm experts and hardware teams and push towards an advantage. We’re already seeing academic publications edging up to that point. I’m more optimistic than ever; it may be early in 2026.
You studied physics at MIT and Caltech. Were you always working on quantum computing?
At MIT, I was doing condensed matter physics experiments. Specifically, I was studying what electrons do at very low temperatures and very high magnetic fields. It may seem like a big step to quantum computing, but there’s a type of qubit called a spin qubit, where you store information in single electron spins.
I transitioned from basic research into the behaviour of correlated electrons to working on spin qubits, and then came to IBM to work on superconducting qubits.
Different companies are pursuing different quantum technologies. How does IBM’s approach differ from others?
Google’s approach is fairly similar to ours. We both use what are called superconducting qubits, and specifically transmons. These are made from metals cooled to very near absolute zero, where they become superconducting.
We then use microwave photons as the qubits themselves.
How do you assemble teams that span physics, materials, and computer science?
This is exactly why the IBM Quantum Network is so important. It’s extremely difficult for any single institution to assemble all the expertise that’s needed.
Our focus is on making the infrastructure available — the hardware, software, and platform — and then partnering with others who bring domain expertise in areas such as chemistry, material science, and algorithms.
We participate in working groups across areas such as healthcare, life sciences, optimisation, and material science. We also have Qiskit add-ons, which let people contribute implementations of their expertise so we can assemble complete workflows.
Trying to bring all of that expertise together under one roof would be like trying to put all the scientists in the world into a single lecture hall. It just doesn’t fit. It has to be a community effort.
When it comes to practical applications, which areas do you see being impacted first?
My personal view is that chemistry and material science will see the first impact. These problems map very efficiently onto quantum computers, with relatively little overhead. We’re already close to parity between quantum and classical simulations for simple molecules.
If you think about human history, we name ages after materials. If we can advance materials discovery, the impact is enormous.
Other areas like cryptography, optimisation, and fluid dynamics require much larger, fault-tolerant quantum computers than we have today. That includes Shor’s algorithm. We expect the first commercial fault-tolerant machines in about four years, but even those won’t be large enough for some of these applications.
How large do quantum systems need to be for cryptography and other advanced applications?
The Starling system we’re targeting for 2029 will have about 200 qubits and around 100 million operations. For Shor’s algorithm, people often talk about around 3,000 qubits and roughly 10¹² operations. That’s far beyond what’s on our roadmap today.
What role do optimisation and machine learning play in quantum’s future?
Those are promising but controversial. Many classical optimisation algorithms are heuristic and work extremely well. It’s hard to prove why. We’re just entering an era where we can explore quantum heuristic algorithms, because now we finally have hardware that’s good enough to run them properly.
Chemistry and materials are the sure bet. Optimisation and machine learning are hopeful, but it’s hard to put a precise timeline on their impact.
AI and HPC dominate the enterprise today. How does quantum fit into that landscape?
Quantum computers are good for specific classes of problems. If a problem isn’t hard for classical computers, running it on a quantum computer would be a terrible choice.
The real opportunity is segmenting problems so that classical computing, AI, and quantum each do what they’re best at. That’s what we call a quantum-centric supercomputer. It’s about combining bits, neurons, and qubits in the most effective way.
We recently announced a partnership with AMD to explore this using their HPC hardware and our quantum hardware. In chemistry, for example, you might simulate most of a molecule classically and use a quantum computer only for the parts where entanglement matters. You get the best of both worlds.
So hybrid models make the most sense.
Yes. We’re exploring this concretely. We work with the University of Tokyo and RIKEN, where one of our quantum computers is paired with Fugaku. We also have a system at RPI working with their supercomputer.
Looking at long-term unsolved problems, where do you think quantum will have the greatest impact?
I still think material science, but I’m a condensed matter physicist. From a theory perspective, quantum computing introduces new complexity classes that don’t exist in classical computing. Demonstrating quantum advantage proves that these classes are real and useful.
Until that point, a skeptic could argue that everything is just classical simulation. Once advantage is demonstrated, that argument disappears.
Finally, where do you see IBM Quantum and the broader field heading over the next few years?
First comes quantum advantage. Then devices will continue to scale in qubits, quality, and speed. In parallel, we’re working on fault-tolerant machines in the lab. Those won’t be computationally useful for a few years, but 2029 is the next major milestone.
We’ll continue improving Qiskit and pushing paradigms like quantum-centric supercomputing. For the field as a whole, a major shift happened when quantum devices became accessible via the cloud. That separated building the hardware from using it.
Another important change is how we think about imperfect hardware. Even with noisy systems, it’s possible to statistically correct results to get accurate answers, particularly in chemistry.
Value comes in different forms. Early systems delivered scientific value before computational value. Today, we’re increasingly focused on computational value. Early systems delivered scientific value before computational value. Today, we’re increasingly focused on computational value.
