Proceedings · Session S-222 · filed November 17, 2026

AI & Emerging Tech in R&DSession paper

QuEra CCO: AI-Designed Circuits Cut Quantum Costs, But 2028 Stands

QuEra reports Claude-designed circuits cut gate counts up to 9.9x on benchmark building blocks, yet its Libra fault-tolerant machine stays on the 2028 roadmap, says CCO Yuval Boger.

By Amara Osei4 min read748 words

Summary

  • An October Harvard–Anthropic–QuEra preprint reports AI-designed circuits beat hand-built designs on seven of eight tasks, improving a Coulomb-potential routine 9.9x (6,355 to 1,140 Toffoli gates; 252 to 142 qubits).
  • A Claude-built laser controller restored frequency locks in 695 of 700 trials; it now runs as a standalone production script without AI in the loop.
  • QuEra's Libra system, 256 logical qubits and about a million reliable operations, remains planned for 2028 on Amazon Braket and at HPC centers.
  • CCO Yuval Boger expects strongly correlated materials, not drug discovery, to yield quantum computing's first scientific wins.
  • A 2025 Merck, Amgen, Deloitte and QuEra study ran quantum reservoir computing on more than 100 qubits to predict molecular properties.
QuEra exec on AI-designed quantum circuits, and why he doesn’t expect drug discovery to be quantum’s first big win
FigureQuEra exec on AI-designed quantum circuits, and why he doesn’t expect drug discovery to be quantum’s first big win — Brecht Corbeel / unsplash

QuEra Computing says AI-designed quantum circuits beat hand-built designs on seven of eight core tasks, cutting a Coulomb-potential routine's gate count from 6,355 Toffoli gates to 1,140 and its qubit requirement from 252 to 142. The October preprint, authored with Harvard and Anthropic researchers, reports a 9.9x improvement on that building block and 4.1x on an Fe2S2 iron-sulfur cluster benchmark. Chief Commercial Officer Yuval Boger is careful about what these numbers mean: they are verified circuit designs and resource estimates, not hardware runs and not full applications.

The work sits inside a broader pattern of AI at the Boston-based neutral-atom quantum developer. Claude models have touched three areas: algorithm design, error correction and system operation. In August, QuEra reported a Claude-built controller restored a laser's frequency lock in 695 of 700 testbed trials; the company now runs that controller as a standalone production script, with no AI in the loop at execution time.

What has been measured, and what has not?

The laser-lock figures come from 100 timed trials for each of seven fault types. The five misses all traced to one test-rig problem, not the control logic, according to Boger. During testing, the laser hopped modes spontaneously 43 times, and the controller caught all 43.

The safety envelope is tight. A laser running logical-qubit operations must hold frequency within about ±1 kHz. Boger said Claude never had free rein: a Model Hardware Standard enforces limits, interlocks and emergency stops at the hardware interface, and the model cannot override them. Engineers set the scope, reviewed every step and decided what counted as proof.

Boger draws a concrete historical comparison for the algorithm work. Researchers cut FeMoco's estimated cost roughly 150,000-fold — from about 50 trillion Toffoli gates to about 340 million — over eight years. He asks whether AI can compress that timeline to months for some building blocks; the preprint, in his framing, is an early sign that it can.

Does AI pull the roadmap forward?

No, Boger says plainly. He will not claim AI has pulled the roadmap in by some number of years. Libra, QuEra's fault-tolerant system designed for 256 logical qubits and about a million reliable operations, remains planned for 2028, with availability on Amazon Braket and at HPC centers. Until then, the new circuits can be tested only in emulation.

He adds two cautions. AI is improving classical methods at the same pace, so the bar quantum must clear keeps moving. And a full application only gains as much as the AI-optimized building blocks contribute to its total cost, so each application must be checked individually. On error correction, he credits the October cascade-decoder result entirely to Harvard collaborators, not QuEra.

Where does the first scientific win land?

"People often expect drug discovery to be the first big win. I don't," Boger said. He expects strongly correlated materials to pay off first — problems where different classical methods give different answers — citing Fermi-Hubbard lattice models up to 10×10 as fitting Libra's 256-logical-qubit, 10⁻⁶ logical error rate, roughly million-operation budget.

Molecular chemistry comes later: ground-state energies of strongly correlated molecules with small-to-medium active spaces, such as metalloprotein active sites and transition-metal chemistry. That work belongs to a next-generation system targeting 1,000+ logical qubits at 10⁻⁹ error — about a billion reliable operations — in 2028/2029. Boger flags this himself as a roadmap target, not a delivery date.

Is pharma already involved?

Yes, but not with quantum-generated training data. A 2025 study by Merck, Amgen, Deloitte and QuEra used quantum reservoir computing to predict molecular properties, running on more than 100 qubits, with a classical model learning from quantum-produced features. Two of six finalists in Wellcome Leap's Quantum for Bio Phase 3 run on QuEra hardware: one on covalent inhibitors for myotonic dystrophy with Nottingham and Phasecraft, the other on binding affinity and NMR estimation with Harvard and MIT. Chemistry specialists in the QuEra Quantum Alliance, notably QC Ware and Phasecraft, are mapping which molecular problems break classical methods before the hardware arrives.

The training-data vision, Boger concedes, remains undemonstrated: a small set of accurate energies with error bars for molecules where classical methods fail, used to correct cheap classical predictions. Producing data classical methods cannot, he notes, requires fault-tolerant hardware — which means waiting for Libra.

via anthropic.com (Original)

Filed under

  • ai-in-quantum-computing
  • quantum-circuit-optimization
  • neutral-atom-quantum-computing
  • fault-tolerant-quantum-computing
  • quantum-error-correction
Share this article:

More from Amara Osei

Amara Osei

Show full bio

News editor covering money, funding and market shifts around research and development at Hypothesis Wire.

253 articles

References

  1. Sandia and Quantinuum Validate 98-Qubit Helios Trapped-Ion System
  2. Sandia Verifies Quantinuum's 98-Qubit Helios at 99.921% Two-Qubit Fidelity
  3. IBM Commits Over $10 Billion to Fund Fault-Tolerant Quantum Roadmap
  4. Edinburgh's Quantum Software Lab secures £20m to build UK quantum advantage
  5. Q/C Technologies taps Sandia's CINT to verify optical AI hardware roadmap

« Previous articleNext article »