How the Debate Began
When D-Wave introduced its first commercial quantum annealing systems in the late 2000s and early 2010s, the company made an audacious claim. Rather than building a universal gate-based quantum computer, it would pursue a specialized machine designed to solve optimization problems through quantum annealing.
The approach was based on a simple idea. Many difficult optimization problems can be represented as an energy landscape containing countless peaks and valleys. The goal is to find the lowest valley, known as the ground state, which corresponds to the optimal solution. Classical optimization methods often become trapped in local minima. Quantum annealing attempts to exploit quantum tunneling, allowing the system to pass through energy barriers rather than climb over them. In principle, this could allow the machine to find better solutions more quickly than classical methods.
The promise was compelling. The challenge was proving that it worked.
One of the earliest and most cited comparisons came from Catherine McGeoch, a computer scientist at Amherst College, whose 2013 study found D-Wave outperforming certain classical solvers by large margins on specific problem instances.¹ The study attracted significant attention but was subsequently criticized for using a poorly tuned classical baseline that did not represent the best available classical optimization methods. The episode established a template for the benchmark disputes that followed: a favorable result for D-Wave, methodological criticism, and subsequent studies with better classical baselines showing smaller or no advantage. Understanding that template is essential for evaluating the claims made by both sides throughout the debate.
The 2014 Paradox
The most important moment in the early quantum annealing debate came in 2014, when two significant studies appeared within months of each other and reached apparently contradictory conclusions.
Sergio Boixo and collaborators at Google published evidence that D-Wave's hardware exhibited genuine quantum mechanical behavior, specifically quantum annealing signatures consistent with quantum mechanical predictions rather than classical simulation.² The qubits were entangled. Quantum tunneling was occurring. The machine was doing something that could not be explained classically.
Almost simultaneously, Troyer and collaborators published a study titled "Defining and Detecting Quantum Speedup" in Science, comparing D-Wave's hardware against carefully tuned classical optimization methods.³ Their central finding was that simulated annealing with parallel tempering, a classical algorithm, matched D-Wave's performance on the benchmark problems used. No scaling advantage was found.
Both papers were methodologically sound. Both findings were accurate. Together they established the critical distinction at the heart of the quantum annealing debate: quantum behavior and computational advantage are separable questions. A machine can exhibit genuine quantum phenomena and still fail to outperform the best classical algorithms on the problems it is designed to solve. That distinction has structured the debate ever since.
The Skeptics
Among the most prominent skeptics has been Scott Aaronson. Aaronson has never argued that D-Wave's machines are fraudulent or non-quantum. Instead, his criticism has focused on the evidence required to demonstrate quantum speedup. From his perspective, proving quantum advantage requires more than outperforming a weak classical baseline. A quantum system must be compared against the best available classical algorithms running on the best available classical hardware.⁴
This standard is demanding because classical optimization algorithms have become extraordinarily sophisticated. Simulated annealing, parallel tempering, tensor-network methods, and numerous problem-specific heuristics have benefited from decades of development and optimization. The existence of quantum effects in a device does not automatically imply computational advantage. Many researchers across the field share this view.
The broader context in which this skepticism operates includes a phenomenon researchers call dequantization. In 2018, Ewin Tang demonstrated that a classical algorithm could match the performance of a quantum algorithm for recommendation systems that had previously been cited as a candidate for quantum speedup.⁵ While Tang's result was not about annealing specifically, it illustrated a recurring pattern across quantum computing: algorithms that appear to provide quantum advantage sometimes inspire classical algorithms that close or eliminate the gap. Quantum annealing skeptics point to this pattern when evaluating claims of advantage from D-Wave's systems.
The Preskill Perspective
John Preskill has generally taken a more measured position on quantum annealing. He has emphasized that demonstrating quantum advantage is inherently difficult because classical algorithms continue to improve. Every time researchers identify a benchmark where a quantum device appears superior, classical researchers often discover new techniques that narrow or eliminate the gap. This phenomenon has played out repeatedly throughout the history of quantum annealing.
Preskill has argued that the important scientific question is not whether quantum effects are present, but whether those effects produce scaling advantages as problem sizes grow.⁶ A machine that is slightly faster on small problems is interesting. A machine that becomes increasingly faster as problems become larger is revolutionary. The latter remains the standard by which quantum annealers are ultimately judged.
Andrew Childs, the Adiabatic Theorem, and Complexity Theory
Andrew Childs is best known for his work on quantum algorithms and computational complexity. From the perspective of theoretical computer science, his work and that of his collaborators has focused on a precise and difficult question: what classes of problems can quantum annealing solve more efficiently than classical computation, and can those advantages be proven?
Quantum annealing is theoretically grounded in the adiabatic theorem, which states that a quantum system will remain in its lowest energy state, the ground state, if it evolves slowly enough from an initial configuration to a final one encoding the solution to the target problem. In principle, this provides a path to solving hard optimization problems by finding the ground state of a carefully designed Hamiltonian.
The difficulty is that the adiabatic theorem provides no efficient guarantee for hard problems. The time required to maintain adiabatic evolution, meaning slow enough evolution to stay in the ground state, can grow exponentially with problem size for the hardest optimization instances. This means that for the most difficult problems, quantum annealing may face the same exponential wall that classical optimization faces, just approached from a different direction.⁷
For gate-based quantum computers, the theoretical advantages of algorithms such as Shor's and Grover's are mathematically proven within complexity theory. Quantum annealing lacks comparable theoretical guarantees for broad classes of practically important problems. This absence of strong theoretical foundations has contributed significantly to ongoing skepticism among theoretical computer scientists.
The D-Wave Position
Researchers at D-Wave have consistently argued that critics often ask the wrong question. From D-Wave's perspective, the goal is not necessarily to prove a universal speedup across all optimization problems. The goal is to provide practical computational advantages on commercially relevant workloads.
D-Wave researchers point to evidence of quantum tunneling, entanglement, quantum coherence, and improved performance on selected optimization problems, as well as commercial deployments in logistics, scheduling, manufacturing, and finance. The company argues that these results demonstrate meaningful value regardless of whether a broad theoretical quantum advantage has been formally established.
This position reflects a practical engineering perspective rather than a purely theoretical one. A business cares whether a problem can be solved faster, cheaper, or better. It may care less about whether the underlying speedup satisfies the strictest standards of computational complexity theory.
D-Wave's more recent strategy has shifted significantly toward hybrid classical-quantum algorithms, where the quantum annealer handles part of a problem and classical computation handles the rest.⁸ This shift changes the central question from whether a quantum annealer alone can outperform classical computation to whether a hybrid system outperforms purely classical approaches. Several of D-Wave's commercial deployments now use hybrid workflows, and the company has invested heavily in hybrid solver infrastructure. Whether this hybrid approach ultimately demonstrates quantum advantage, or merely provides a practical computational tool that combines quantum and classical resources effectively, remains an open question.
The Recent Arc: From Coherence to Supremacy to Challenge
The debate entered a new phase between 2022 and 2026 as D-Wave's hardware improved substantially and the scientific claims it supported became more ambitious, before meeting significant classical counter-results.
In 2022, Andrew King and collaborators at D-Wave published evidence of coherent quantum annealing in a programmable 2,000-qubit Ising chain in Nature Physics, demonstrating quantum coherence at a scale beyond any previous programmable quantum system. ⁸ The following year, a companion paper in Nature extended those results to a 5,000-qubit programmable spin glass, providing evidence of quantum critical dynamics that the authors argued went beyond what classical simulation could efficiently reproduce. ⁹
These results built toward what D-Wave described as its most significant scientific claim to date. In March 2025, D-Wave published "Beyond-Classical Computation in Quantum Simulation" in Science, reporting that its Advantage2 prototype performed simulations of quantum dynamics in programmable spin glasses in minutes that would have taken nearly one million years on a classical supercomputer built with GPU clusters. ¹⁰ D-Wave described this as the world's first demonstration of quantum computational supremacy on a useful, real-world problem.
The claims sparked immediate debate. Researchers at École Polytechnique Fédérale in Lausanne argued that the problems D-Wave solved can be tackled without any need for quantum entanglement, and that their classical approach did not require even simulating the effects of quantum entanglement. ¹¹ D-Wave disputed this characterization, arguing that the counter-results did not challenge the core supremacy claim.
The most pointed rebuttal came in May 2026, when researchers at the Flatiron Institute's Center for Computational Quantum Physics demonstrated that a 3D tensor-network classical algorithm, combined with a technique called belief propagation, could reproduce D-Wave's Advantage2 results on standard workstations and even ordinary laptops. ¹² The Flatiron study, published in Science, argued that D-Wave's benchmarking methodology had failed to adequately explore the classical algorithm solution space.
The pattern was recognizable to anyone who had followed the debate since 2014. D-Wave demonstrates results that appear to exceed classical capability. Classical algorithm researchers develop new techniques that match those results. The boundary of what constitutes quantum advantage shifts. Neither side concedes. The science advances.
What Everyone Agrees On
Despite years of debate, there is considerably more agreement among researchers than outsiders often realize.
Most researchers agree that D-Wave's systems are genuine quantum devices in which quantum tunneling plays a real role. Entanglement has been observed within the hardware. Quantum annealing is fundamentally different from classical simulated annealing, even if the two share a conceptual resemblance. Optimization remains one of the most important practical computational challenges, and any approach that addresses it more effectively than existing methods has genuine value.
The disagreement concerns magnitude, not existence. The question is not whether quantum effects occur. The question is whether those effects provide large, scalable, and economically meaningful advantages over the best classical alternatives. That question has not been definitively answered in either direction, and the most recent exchange in the pages of Science suggests it will not be resolved soon.
What may ultimately be the most important lesson of the quantum annealing story is that scientific progress is rarely a straight line. New technologies often spend years inhabiting an uncomfortable middle ground between promise and proof. Quantum annealing has occupied that space longer than most, and the exchanges of 2025 and 2026 suggest it will continue to do so. Whether future generations of annealers eventually demonstrate undeniable quantum advantage, or whether the field's center of gravity shifts entirely toward gate-based systems and fault-tolerant computation, remains one of the most fascinating open questions in modern computing.
References
¹ McGeoch, C. C., and Wang, C. (2013). Experimental evaluation of an adiabatic quantum system for combinatorial optimization. Proceedings of the ACM International Conference on Computing Frontiers. ACM, New York.
² Boixo, S., Rønnow, T. F., Isakov, S. V., Wang, Z., Wecker, D., Lidar, D. A., Martinis, J. M., and Troyer, M. (2014). Evidence for quantum annealing with more than one hundred qubits. Nature Physics, 10, 218–224.
³ Rønnow, T. F., Wang, Z., Job, J., Boixo, S., Ghosh, S., Isakov, S. V., Wecker, D., Martinis, J. M., Lidar, D. A., and Troyer, M. (2014). Defining and detecting quantum speedup. Science, 345(6195), 420–424.
⁴ Aaronson, S. Shtetl-Optimized (blog). Various posts on D-Wave and quantum speedup: Feb 2012, May 2013, Feb 2014, Aug 2015, Dec 2015 and Mar 2017. https://scottaaronson.blog reviewed online on 7/12/2026.
⁵ Tang, E. (2019). A quantum-inspired classical algorithm for recommendation systems. Proceedings of the 51st Annual ACM Symposium on Theory of Computing (STOC). ACM, New York. (Based on 2018 preprint.)
⁶ Preskill, J. (2012). Quantum computing and the entanglement frontier. Rapporteur talk at the 25th Solvay Conference on Physics. arXiv:1203.5813.
⁷ Farhi, E., Goldstone, J., Gutmann, S., Lapan, J., Lundgren, A., and Preda, D. (2001). A quantum adiabatic evolution algorithm applied to random instances of an NP-complete problem. Science, 292(5516), 472–475. See also Aharonov, D., van Dam, W., Kempe, J., Landau, Z., Lloyd, S., and Regev, O. (2007). Adiabatic quantum computation is equivalent to standard quantum computation. SIAM Journal on Computing, 37(1), 166–194.
⁸ King, A. D., et al. (2022). Coherent quantum annealing in a programmable 2,000 qubit Ising chain. Nature Physics, 18(11), 1324–1328.
⁹ King, A. D., et al. (2023). Quantum critical dynamics in a 5,000-qubit programmable spin glass. Nature, 617, 61–66.
¹⁰ King, A. D., et al. (2025). Beyond-classical computation in quantum simulation. Science, 388(6743), 199–204.
¹¹ Mauron, L., and Carleo, G. (2025). Classical simulation of D-Wave's spin glass dynamics. École Polytechnique Fédérale de Lausanne. Reported in New Scientist, March 2025.
¹² Tindall, J., Mello, A., Fishman, M., Stoudenmire, E. M., and Sels, D. (2026). Dynamics of disordered quantum systems with two- and three-dimensional tensor networks. Science. Flatiron Institute, Center for Computational Quantum Physics and Boston University.