AWS Braket and the Ocelot cat-qubit chip

What it is

The AWS Ocelot chip is Amazon’s entry into the race to build quantum hardware that can correct its own errors, and it approaches the problem from an unusual angle. Ocelot is built around cat qubits, a design that suppresses one whole class of error in the hardware itself, before any software correction is applied. The name is a nod to Schrödinger’s cat, the thought experiment in which a cat is held in a superposition of alive and dead; a cat qubit encodes information in a similar superposition of two stable states, arranged so that the most common kind of error becomes very unlikely.

How cat qubits help

To see why that helps, separate the two ways a qubit can go wrong. A bit-flip turns a 0 into a 1 or the reverse; a phase-flip corrupts the more delicate quantum relationship between states. Ordinary error correction has to guard against both, which is a large part of why fault-tolerant machines are expected to need so many physical qubits for each reliable logical one. Cat qubits are engineered so that bit-flips are strongly suppressed at the hardware level, leaving mainly phase errors to deal with. That is the pitch for the AWS Ocelot chip: cut the correction overhead needed for fault tolerance by handling one error type in the physics rather than the software.

Ocelot and Braket

Amazon runs its own silicon programme, of which Ocelot is the visible product, and this hardware effort sits alongside Braket, its multi-vendor cloud quantum platform. The two serve different purposes. The AWS Ocelot chip is a long-term research bet on a particular route to fault tolerance, while Braket is a working service today, letting customers run programmes on quantum machines from several suppliers through a single cloud interface.

The access layer

Braket is arguably the more immediately important of the two. It is how most enterprises will first touch quantum hardware, in much the same way most organisations first touched modern artificial intelligence through a cloud API rather than by buying their own machines. That framing captures Amazon’s instinct: make the technology available as a service, meet customers where they already are, and let them experiment without capital outlay. If quantum computing follows the path that cloud AI took, the companies that own the access layer stand to benefit regardless of which hardware modality eventually wins. Our AI hub traces how that cloud-first pattern played out in the previous wave of the technology.

What to watch

The combination is telling. With Braket, Amazon can sell quantum access now and stay neutral about which qubit technology prevails. With Ocelot and its cat qubits, it keeps a stake in the hardware itself, positioned around a specific and credible idea about how to reach fault tolerance more cheaply. Suppressing bit-flip errors in hardware does not solve everything, because phase errors still need correcting, yet it could meaningfully reduce the number of physical qubits a useful machine requires, and qubit overhead is one of the central obstacles the whole field faces. The fewer physical qubits each logical one demands, the sooner and the more cheaply a practical machine arrives.

Where the AWS Ocelot chip sits in the wider picture is as one of several competing answers to the same question: how to make quantum computers reliable enough to be useful without needing impossibly large numbers of physical qubits. Cat qubits are one route; topological qubits and photonic fusion are others. What to watch is whether Amazon’s approach carries from a promising chip into a system that demonstrably lowers the correction burden at scale, and whether Braket remains the default front door through which businesses meet quantum computing. For a broader view of the modalities in contention, our quantum explainer hub sets them side by side.