Latest quantum computing breakthroughs

The latest quantum computing breakthroughs point to one shift that matters more than any qubit count: for the first time, adding more qubits made a quantum processor more accurate, not less. Google demonstrated this with its Willow chip in December 2024, and the result reshapes how the field talks about progress. For years the story was a race to pile up qubits. The real contest now is error correction, and 2024 through 2026 delivered the first hard evidence that it can work at scale.

This page separates what labs have actually shown from what press releases imply. Both halves are worth reading. The demonstrated results are real and hard-won. The marketing around them runs well ahead of what these machines can do today.

What changed recently

The standout event is Google’s Willow processor, a 105-qubit superconducting chip built in Santa Barbara and announced on December 9, 2024. Willow crossed a line the field had chased for nearly three decades, called the error-correction threshold.

Here is the idea. Physical qubits are fragile. They lose their quantum state to heat, stray fields, and manufacturing flaws. Error correction spreads one unit of reliable information, a logical qubit, across many physical qubits so the group can detect and fix faults. The catch is that every extra qubit you add also brings its own errors. Below a certain physical error rate, correction wins and the logical error keeps dropping as you scale up. Above it, you just add noise. Getting below that threshold is the difference between a machine that improves as it grows and one that drowns in its own mistakes.

Willow got below the threshold. Google’s team tested grids of qubits at increasing size, from a 3x3 array to 5x5 to 7x7, and each time they made the grid larger the logical error rate fell by roughly half. That halving with each step is the signature engineers wanted to see. The full result was published in Nature under the title “Quantum error correction below the surface code threshold.” It is a measured, peer-reviewed claim, not a demo reel.

Google paired this with a second announcement that drew more headlines and deserves more skepticism. Willow ran a random circuit sampling task in under five minutes that, by Google’s estimate, would take one of today’s fastest supercomputers 10 septillion years, written as 10 to the 25th power. The number is staggering and mostly beside the point. Google states plainly that random circuit sampling has not shown a practical commercial use. It is a physics benchmark chosen because it is hard for classical machines, not a job anyone needs done. Google also assumed generous conditions for the comparison and expects classical methods to keep improving. Treat the five-minute figure as a stress test of the hardware, not a measure of usefulness.

IBM spent 2025 laying out a different kind of marker: a dated engineering plan. In June 2025 the company published a plan toward a large-scale fault-tolerant machine called Starling, targeted for 2029 and built at a new data center in Poughkeepsie, New York. Starling is meant to hold 200 logical qubits and run 100 million quantum operations, which IBM frames as 20,000 times more operations than current quantum computers manage. Behind Starling sits a chain of stepping-stone processors: Loon in 2025 to test the wiring, Kookaburra in 2026 to combine quantum memory with logic, and Cockatoo in 2027 to link two modules together. A larger follow-on, Blue Jay, is sketched for 2,000 logical qubits and 1 billion operations.

IBM’s plan rests on a change in error-correcting code. The company is moving to quantum low-density parity-check codes, or qLDPC, which it says cut the number of physical qubits needed per logical qubit by about 90 percent compared with the surface codes used by most rivals. If that holds up in hardware, it lowers the biggest cost in building a useful machine.

Why these breakthroughs matter

The reason Willow matters is narrow and real. Error correction was the open question that no amount of extra qubits could answer on its own. A 1,000-qubit chip with a bad error rate is a physics curiosity. A smaller chip that gets quieter as it grows is a path to something useful. Willow is the first strong experimental sign that the second story is possible with real hardware. IBM’s 1,121-qubit Condor processor, built in 2023, makes the contrast plain. Condor pushed the qubit count higher than anything else at the time, yet IBM never opened it to public use. Sheer count without fidelity does not buy much.

That is why the field’s language shifted from physical qubits to logical qubits. A logical qubit is the unit that actually does protected computation, and today building even one good logical qubit takes dozens to hundreds of physical qubits. Willow showed the error rate moving the right way as the logical qubit grew. IBM’s plan is a bet on shrinking the physical-qubit tax so that 200 logical qubits become buildable. Both are steps toward the same destination: machines where you can run a long calculation without errors piling up faster than you can fix them.

The stakes reach past physics labs. A fault-tolerant machine at scale would threaten the encryption that protects internet traffic and financial systems, which is why standards bodies are already moving to post-quantum encryption. The same hardware could model chemistry and materials that classical computers struggle to simulate, from battery electrolytes to catalysts. None of that is available now. The point of 2024 through 2026 is that the engineering path to it stopped being purely theoretical.

Where quantum still falls short

Set the plans aside and the honest picture is sobering. Today’s quantum computers are noisy, small in usable terms, and unable to beat classical machines on any problem that industry actually cares about. No commercial task runs faster or cheaper on a quantum computer today than on a conventional one. That sentence is worth rereading, because most coverage implies the opposite.

The gap between physical and logical qubits is the reason. A machine with 156 physical qubits, like IBM’s Heron, does not give you 156 reliable qubits. Error correction consumes most of them. Estimates for breaking widely used encryption call for thousands of stable logical qubits, which translates to millions of physical qubits at current overheads. We are orders of magnitude away.

Benchmarks also invite confusion. Random circuit sampling, the task Willow aced, was designed to be hard for classical computers rather than useful to anyone. Claims of a supercomputer needing longer than the age of the universe describe that one artificial task, not a spreadsheet or a drug-discovery run. When you see an astronomical speed figure, check whether the problem being solved is one that people outside a quantum lab would ever want solved.

Dates deserve the same caution. IBM’s 2029 target for Starling is a corporate goal, not a delivered product, and quantum timelines have slipped before. Willow’s result, though genuine, was demonstrated on small grids. Scaling error correction from a 7x7 patch to a full machine with hundreds of logical qubits is a large, unproven jump in engineering. The physics is no longer the blocker. The manufacturing, wiring, and control systems are.

What to watch next

A few concrete signals will tell you whether the field is delivering on 2024’s promise. Watch for IBM’s Loon and Kookaburra chips to appear on schedule in 2025 and 2026 and to show that qLDPC codes work in hardware, not just in simulation. That is the load-bearing claim in IBM’s plan.

Watch for the first demonstration of a logical qubit with a lifetime far longer than its physical parts, sustained through many rounds of correction rather than a brief window. Watch, too, for any team running a real algorithm end to end on error-corrected qubits, even a small one, since that would move the story from components to computation.

Be wary of three things: qubit counts quoted without error rates, speed claims tied to random circuit sampling, and any suggestion that today’s machines threaten encryption now. Progress in quantum computing is real and, for once, measurable against a clear physics target. It is also slower and narrower than the headlines suggest. Both facts are true at the same time.

For more coverage of the technologies shaping orbit and beyond, visit our technology hub. For a related look at how another fast-scaling field is reshaping the sky above us, read our explainer on satellite megaconstellations.