AWS's Ocelot chip bakes error resistance into qubits, cutting correction costs up to 90%

In February 2025 Amazon became the third tech giant in as many months to announce a quantum chip, unveiling Ocelot — a prototype built on so-called cat qubits, named for Schrödinger's cat. The target is one of quantum computing's biggest stumbling blocks: errors caused by heat, vibration, electromagnetic interference from phones and wifi, even cosmic rays.

Cat qubits, a technology pioneered by the French company Alice & Bob, work by engineering error resistance into the design of the qubits themselves rather than paying for correction around them. Ocelot carries five cat qubits among 14 key components, and Amazon says the chip could cut error-correction costs by up to 90% versus current approaches. The findings are published in Nature; Inria's Mazyar Mirrahimi called Ocelot an important step toward hardware-efficient fault-tolerant quantum computation.

The announcement lands in the middle of an industry pivot. IDC research manager Heather West, briefed before publication, called the result an advancement rather than a breakthrough — and notes the field is shifting from counting qubits to making systems usable at scale, which means solving error correction inside the machine. Michael Cuthbert, director of the UK's National Quantum Computing Centre, welcomed mechanisms that enable error correction without huge overheads in chip size, energy consumption and systems complexity.

AWS's endgame is offering quantum computing as a cloud service — and pointing it at Amazon's own logistics. Oskar Painter of the AWS center notes that a 1% improvement in global logistics optimization is worth large dollars, and that quantum could deliver it more effectively and in real time. On timing, he now calls an 'aggressive date' of about a decade to useful machines increasingly realistic; five years ago he would have said 20 or 30 years.

Attack the bottleneck in hardware: cat qubits engineered to resist errors natively, so fault tolerance stops costing an armada of extra qubits.

Error correction is the chokepoint of quantum scaling, so attacking it inside the qubit attacks the industry's real bottleneck.

Engineering error resistance into hardware trades correction overhead for design — up to 90% cheaper, by Amazon's own estimate.

AWS has a built-in first customer: its own global logistics network, where even a 1% optimization is worth large dollars.

When a defect is ruinous to patch around, redesign the component so the defect barely happens — error resistance beats error correction.

The work is published in Nature, and experts back the direction while cautioning that scaling remains hard — Painter agreed '100%' that scaling today's experimental systems will not be easy. The same quarter brought experimental chip announcements from Microsoft and Google, with the industry pivoting from qubit counts toward error correction as the measure that matters.