Quantum Chip Race

Three quantum approaches, different evidence

Google's Willow, Microsoft's Majorana program and Amazon's Ocelot explore different routes toward reliable quantum computing. Comparing their physical qubit counts as if they were ordinary processors misses the central challenge: useful calculations must survive errors long enough to produce a trustworthy result.

The evidence has advanced since the original chip announcements. Google has reported a verifiable quantum-advantage experiment, Microsoft has published a new device-lifetime preprint, and AWS has demonstrated a particular approach to error correction. Each result needs to be judged by what the experiment actually measured.

Physical qubits and logical qubits

A physical qubit is a noisy component. A logical qubit encodes information using an error-correction scheme so that the system can detect and manage errors. Adding hardware helps only when the underlying quality and correction process are good enough. Below a suitable error threshold, larger codes can suppress logical errors; above it, extra components can create more problems than they solve. Google's error-correction paper

A reliable memory is also different from a complete computer. Keeping encoded information alive, executing logical gates, connecting many logical qubits and completing an application are separate milestones. Hardware that demonstrates one stage has not automatically demonstrated all the others.

Willow: error correction and Quantum Echoes

Google's Willow research demonstrated below-threshold surface-code error correction. The result supports the idea that increasing code size can improve the reliability of a quantum memory. It does not mean errors have disappeared or that every application can run economically on the chip. Published results

In October 2025, Google reported Quantum Echoes, an algorithm using out-of-time-order correlators and a reversible evolution experiment. The team described verifiable quantum advantage for a specific set of circuits on Willow. This is more informative than a generic claim that a quantum machine is faster, because it identifies the task and comparison. It still does not establish a speedup for ordinary cloud workloads or every chemistry problem. Google research explanation

For investors, the right progression is from a credible experiment to repeatable useful workloads, then to customers willing to pay enough to cover the system's costs. A benchmark can advance the first step without settling the last.

Microsoft's Majorana approach

Microsoft's program pursues semiconductor-superconductor devices and a proposed topological architecture. Its June 2026 preprint reports a roughly 20-second parity lifetime in an indium-arsenide/lead tetron device. That is a measurement of a device property under the reported conditions. Tetron preprint

The result should not be expanded into proof of a complete topological quantum computer. Parity lifetime, the identification of the underlying states, control operations, logical error rates and fault-tolerant computation answer different questions. The preprint's status and experimental scope belong beside the result, especially when it is used to support claims about Microsoft's broader roadmap.

The business opportunity remains conditional on those engineering steps. A promising architecture can justify research investment while its eventual production scale, cost and commercial timing remain uncertain.

Ocelot: reducing the correction burden

AWS's Ocelot work uses cat qubits, which encode information in oscillator states, together with an outer repetition code. The published experiment describes a logical memory and the operations used to detect errors. The approach seeks to exploit a strong bias between different error types to improve hardware efficiency. Ocelot research

That bias is useful precisely because error correction has a cost. If the device suppresses one error type effectively, the surrounding code can focus resources on the remaining errors. Claims about greatly reduced future overhead are architectural projections; they are not a measured discount on operating a complete commercial quantum computer.

What the next comparisons should measure

An informative comparison specifies the workload, logical error rate, number and quality of operations, runtime, classical baseline and total resources. It should also state whether the result has been independently reproduced and whether improvements in classical methods change the claimed advantage.

The commercial comparison adds integration, availability and price. Scientific progress does not transfer directly into near-term cloud revenue. These companies also have large existing businesses, so a quantum milestone should not be substituted for an assessment of their current earnings.

Northwise's Alphabet, Microsoft and Amazon pages connect this technology discussion with company research. Dated valuation models remain separate from this evolving scientific overview.