By Sophie Lindqvist | Chip Dispatch

QuEra will place a 256-logical-qubit fault-tolerant computer called Libra on Amazon Web Services' Braket platform by 2028, the company said this week.

The neutral-atom machine will be QuEra's first fault-tolerant product. It targets the "megaquop-class" tier, defined as one million reliable logical quantum operations per run, according to QuEra.

How large is Libra?

The system pairs more than 256 logical qubits with 10,000 to 15,000 physical qubits. QuEra said the physical qubits consist of neutral Rubidium atoms held in optical tweezers. A logical error rate of 10⁻⁶ is the design target. That figure translates to one error per million operations.

Libra runs at room temperature. No cryogenic cooling is required. The hardware fits standard 19-inch server racks.

What does "megaquop-class" actually buy a user?

QuEra frames the metric in two parts. Logical qubit count sets application breadth. Logical error rate sets calculation depth. The deeper the circuit, the longer an algorithm can run before noise overwhelms the result, which is why the 10⁻⁶ figure matters more than a simple qubit headline.

The company quoted Chief Commercial Officer Yuval Boger on the practical difference. "The real metric is how many good qubits you have – error-corrected qubits – and how good they really are," Boger said. He drew the gap in numbers: today's leading physical-qubit systems produce one error every 1,000 operations. QuEra's target is one per million. That is the line between unusable output and long calculations.

Why does the physical-to-logical qubit ratio matter so much?

It dictates hardware scale and cost. Industry assumptions as recently as two years ago placed the ratio near 1,000-to-1, meaning roughly 1 million physical qubits would have backed 1,000 logical ones. A QuEra, Harvard, and MIT paper from April showed ratios as low as two physical qubits per logical qubit. That single result cuts quoted machine sizes by orders of magnitude.

A second compression compounds the first. Algorithm qubit counts have shrunk. Shor's original factoring estimate required roughly 1 million qubits. Current research, Boger said, points to 20,000 to 30,000. Convergence of three threads — better qubits, lighter error correction, and leaner algorithms — in Boger's view, now puts useful commercial applications within roughly two years.

Who competes with QuEra on Braket?

Libra follows Aquila, which QuEra launched on Braket in November 2022. Aquila exposes 256 physical qubits from Boston for 130 hours a week. QuEra now extends that AWS partnership into fault tolerance.

Alice & Bob disclosed its own step days earlier. The Paris-based company unveiled the Helium Quantum System, integrating 18 physical cat qubits into a complete rack with cabling, control electronics, and software, marking a transition from chip development to full system delivery. The published power draw is 40 kilowatts. Cat qubits encode information into quantum states of light rather than discrete particles. The company says fault tolerance is built into the hardware layer.

Google, working in superconducting, said in March it would expand into neutral atoms. The industry, Boger noted, has not converged on one modality. He expects trapped-ion, photonic, and superconducting platforms to remain in the running.

What is the catch?

Libra ships in 2028, a roadmap date rather than a delivered product. QuEra provided no power budget, no per-shot Braket pricing, and no published benchmark against classical supercomputers. "Megaquop-class" is a QuEra definition. Whether the two-qubit-per-logical figure holds at 15,000 physical qubits under realistic workloads remains an open question that no independent benchmark has yet answered in a published study.

What to watch next

The Libra milestone QuEra plans to hit first is intermediate-scale fault tolerance, ahead of the full 256-logical-qubit build-out. Braket pricing and a published logical-qubit benchmark against existing 100-qubit superconducting systems will mark whether the 2028 roadmap stays on schedule.