Quantum Computing Progress in 2026: IBM, Google, and the Race to Fault Tolerance

Two hardware announcements from late 2024 and late 2025 generated more boardroom quantum anxiety than anything since NIST published its post-quantum standards in August 2024. Google's Willow chip demonstrated below-threshold error correction for the first time on real hardware. IBM announced Nighthawk, a 120-qubit processor targeting quantum advantage in practical computation by end-2026. Both results are genuine scientific progress. Neither changes the 2033–2035 Q-Day central estimate that security planners use for migration scheduling, and understanding why requires a short tour of what these announcements actually showed.

Where Quantum Hardware Actually Stands in 2026

Quantum computing hardware in 2025–2026 operates in what researchers call the NISQ era: Noisy Intermediate-Scale Quantum. NISQ devices have between 50 and a few hundred physical qubits with error rates that make sustained deep circuit execution unreliable without error correction. The word "noisy" here is technical: individual gate operations on physical qubits have a non-trivial probability of introducing errors that compound through long computations. A circuit that requires thousands of sequential gate operations, as Shor's algorithm does, will produce garbage output before it reaches the result.

The distinction that matters for security planning sits between physical qubits and logical qubits. A logical qubit is an error-corrected unit built from many physical qubits working collectively to detect and correct faults as they occur. Running Shor's algorithm at RSA-2048 scale would require millions of logical qubits with very low error rates. Gidney and Ekerå's 2021 analysis, the authoritative published estimate, puts the physical qubit requirement at approximately 20 million under current error rate assumptions. IBM's best deployed hardware is at 120–156 physical qubits. The gap is several orders of magnitude, and closing it requires fault-tolerant quantum error correction, not just more qubits.

That is the context in which both the Willow and Nighthawk announcements should be read.

Google Willow: What the Error Correction Result Means (and Does Not Mean)

Google Quantum AI announced Willow in December 2024. The chip has 105 physical qubits and produced two results that drew significant attention.

The first was a random circuit sampling (RCS) benchmark: Willow completed a computation that Google's team estimates would take the Frontier supercomputer at Oak Ridge National Laboratory approximately 1025 years. That number is so large it is essentially meaningless as a practical comparison, and RCS benchmarks are designed to be hard for classical computers while being tractable for quantum hardware. They are not practically useful computational tasks. Google's own researchers noted in the same paper that the outstanding challenge remains demonstrating beyond-classical performance on "an application with real-world impact." The RCS result does not mean Willow can do useful things faster than classical computers.

The second result is the one that matters scientifically. As the team scaled up the surface code error correction from distance-3 to distance-5 to distance-7 (adding more physical qubits to the error correction scheme), the logical error rate decreased. Previous hardware had always shown the opposite behaviour: adding qubits added more errors than the code could correct, because the qubits themselves were too noisy. Willow is the first superconducting device to show that scaling the code produces longer-lived logical qubits rather than shorter-lived ones. This is the "below-threshold" result. It demonstrates that the physics of quantum error correction works as theory predicts on real hardware, not just in simulations.

The significance is real and the nuance is important. Willow demonstrated that scalable error correction is physically achievable on superconducting qubits, which removes a legitimate scientific objection to fault-tolerant quantum computing timelines. It does not demonstrate fault tolerance at useful scale. A distance-7 surface code uses 49 physical qubits to produce one logical qubit with reduced error rate. Shor's algorithm on RSA-2048 requires millions of them. The pathway is now credible. The pathway is still long.

IBM Nighthawk, Loon, and the 2029 Fault-Tolerance Target

IBM's November 2025 announcements were different in character from Google's: less about a single milestone and more about a detailed engineering roadmap with named deliverables by year.

Nighthawk is a 120-qubit processor using 218 next-generation tunable couplers in a square lattice. Its design allows circuits with approximately 30 per cent more complexity than the Heron processor family that preceded it, and IBM's target for Nighthawk is quantum advantage in a commercially useful computation by end-2026. Alongside it, IBM announced Loon, an experimental processor designed to validate the hardware components required for fault-tolerant quantum computing assembled on a single chip. Loon is a proof-of-concept for architecture, not a production system.

The IBM roadmap beyond these two processors is unusually specific. Kookaburra (2026) is the first module designed to store information in quantum low-density parity-check (qLDPC) memory and process it with an attached logical processing unit. Cockatoo (2027) targets entanglement between modules. The full fault-tolerant system target remains 2029. IBM publishes gate count targets alongside processor names: Nighthawk supports up to 5,000 two-qubit gates at launch, scaling to 7,500 gates by end-2026 and 10,000 gates in 2027.

Those gate count targets are meaningful engineering milestones. Shor's algorithm on RSA-2048 requires gates numbered in the billions, not thousands. The targets show IBM is on an upward trajectory; they also quantify how far the trajectory has yet to travel. IBM's roadmap is the most detailed public hardware development schedule from any major quantum developer, which makes it the most verifiable over time.

It is worth noting that Heron came in two versions: Heron r1 at 133 qubits and Heron r2/r3 at 156 qubits. Nighthawk at 120 qubits is not larger than Heron in qubit count. The advance is in circuit complexity and coupler design, not raw qubit count, which reflects a shift in what IBM considers the primary progress metric.

"Quantum Advantage by 2026": The Right Interpretation

IBM's target of quantum advantage by end-2026 is genuine and worth understanding precisely. "Quantum advantage" in IBM's roadmap language means a calculation with practical commercial utility that runs faster on a quantum processor than on the best available classical hardware. This is a specific and achievable goal for particular problem classes.

The problem classes where near-term quantum advantage is most plausible include specific chemistry simulations, structured optimisation problems, and machine learning tasks with suitable quantum feature maps. These are useful applications. They are not cryptanalytic applications. Breaking RSA or ECC requires a fundamentally different scale of computation, an entirely different class of algorithm, and error rates that near-term quantum advantage demonstrations do not target or require.

The press cycle around quantum advantage announcements tends to compress "faster than classical for chemistry" and "could break encryption" into a single undifferentiated category of "quantum threat." They are not the same thing. A drug discovery company has a reason to care that Nighthawk hits its 2026 advantage target. A security planner's Q-Day assessment does not change because of it.

What This Means for Security Planning

The correct interpretation of 2024–2026 hardware progress for security teams is not that Q-Day has accelerated. The NCSC, NSA, and NIST maintain migration timelines built on the 2033–2035 range and have not revised those timelines upward in response to Willow or Nighthawk. No major standards body has changed its deprecation schedule or migration guidance as a direct consequence of either announcement.

What has changed is the credibility of the timeline, not its position. Willow's below-threshold result removes the objection that scalable error correction might prove physically impossible on superconducting hardware. IBM's roadmap removes the objection that fault-tolerant timelines lack engineering milestones. Both developments strengthen the case that the 2033–2035 central estimate is realistic rather than speculative. For the organisation that was treating Q-Day as a distant theoretical concern rather than a planning input, these announcements close that rhetorical escape route.

For harvest-now-decrypt-later (HNDL) risk, the hardware announcements are directly relevant in a different way. HNDL does not require Q-Day to have arrived. Adversaries collecting encrypted traffic today are acting on the probability that Q-Day eventually occurs. Progress reports that increase the engineering credibility of that probability should accelerate hybrid ML-KEM deployment in the organisations that have been waiting to see how the hardware trajectory develops. The case for deploying X25519+ML-KEM hybrid key exchange now, rather than at some future point in the migration programme, is stronger because of Willow and Nighthawk, even though neither machine can run Shor's algorithm.

For the fuller picture of what Shor's algorithm actually requires and which encryption systems it threatens, see the companion article on why RSA and ECC will not survive a quantum computer. For the migration timeline and how the Q-Day range maps to organisational planning phases, see the CRQC timeline analysis.

The hardware race is real. The timelines are not in dispute. What remains consistently misread is which type of quantum progress matters for which planning decisions. Willow matters for the physics community and for any security planner who was waiting for proof that fault-tolerant quantum computing is physically achievable. Nighthawk's advantage target matters for the organisations whose commercial workflows could benefit from near-term quantum computation. Neither announcement makes a 2028 deadline appropriate for something that does not need to be done until 2030. The 2033–2035 window, and the NCSC's Phase 1 target of having discovery and planning complete by 2028, remains the right frame.

Quantum technologies are evolving quickly and new developments emerge regularly. This page was last updated on 18/05/2026. For the most current information, we recommend contacting us directly.