QuEra Announces Libra: 256 Logical Qubits on Amazon Braket by 2028

June 15, 2026 — QuEra Computing announced Libra, its first fault-tolerant quantum computer, targeting availability on Amazon Braket in 2028 through an expanded multi-year strategic collaboration with Amazon Web Services.

Libra is designed as a megaquop-class system: over 256 error-corrected logical qubits, a logical error rate of 10⁻⁶ (one in a million), and approximately one million reliable logical quantum operations. The system builds on QuEra’s neutral-atom architecture and the work of its scientific founders at Harvard and MIT.

The announcement was accompanied by a substantive AWS blog post from Eric Kessler, General Manager of Amazon Braket, who wrote that AWS sees “a clear technical pathway to large-scale fault-tolerant quantum processors” and envisions quantum processors eventually sitting alongside CPUs, GPUs, and AI accelerators in the AWS compute portfolio.

QuEra claims eight peer-reviewed papers in Nature and Physical Review Letters validate every building block of the Libra architecture, including logical qubits, below-threshold error correction, transversal gates, magic state distillation, real-time decoding, and continuous operation of thousands of qubits.

The collaboration deepens a relationship that began in 2022, when QuEra’s Aquila (a 256-physical-qubit analog quantum computer) became the first neutral-atom system on Amazon Braket. QuEra’s Gemini system, a digital gate-model machine with logical-qubit capabilities, is currently co-located with the ABCI-Q supercomputer in Japan.

AWS also emphasized its own complementary work: the Ocelot superconducting chip using cat-qubit architecture, which Kessler characterized as “deeply complementary” to neutral atoms rather than competitive.

My Analysis

In April, I updated the CRQC Scorecard, where I assessed the gap between each quantum computing modality and a cryptographically relevant quantum computer (CRQC). In that analysis, I wrote that neutral-atom qubits had the smallest combined gap to CRQC requirements across my three executive metrics (Logical Qubit Capacity (LQC), Logical Operations Budget (LOB), and Quantum Operations Throughput (QOT)) and the most complete set of demonstrated capabilities. I stated publicly that I believed neutral atoms would be the first modality to reach a CRQC.

QuEra’s Libra announcement reinforces this with a date, specs, and AWS co-signing the engineering roadmap.

But, just to be clear for my cyber readers: Libra is not a CRQC. Not even close. It is a megaquop machine, designed for one million reliable operations. Breaking RSA-2048 with Shor’s algorithm requires roughly 6.5 billion Toffoli gates. Libra is the first credible fault-tolerant quantum computer targeting cloud availability on a specific date with specific performance parameters, backed by peer-reviewed demonstrations of every required capability. The CRQC question and the Libra question are different questions, and conflating them will lead people to the wrong conclusions in both directions.

The Evidence Base: Eight Papers, Six Nature Publications

The strength of QuEra’s position is not the announcement itself (companies announce ambitious roadmaps constantly) but the peer-reviewed research behind it. The AWS blog post is more useful than QuEra’s press release here because it cites specific papers. Between the two sources, the evidence base includes:

The Architecture Paper. Dolev Bluvstein and collaborators at Harvard, MIT, and QuEra used reconfigurable arrays of up to 448 neutral atoms to implement all key elements of a universal, fault-tolerant quantum processing architecture in a single experiment (Nature, June 2025). I covered this paper in depth when it was published; it is, in my assessment, the most complete fault-tolerant demonstration on any quantum platform to date. The paper demonstrated 2.14× below-threshold error suppression in repeated surface code QEC, transversal gates and lattice surgery for logical entanglement, universal logic through transversal teleportation with 3D [[15,1,3]] codes, and mid-circuit qubit re-use that increased cycle rates by two orders of magnitude. The team ran deep-circuit protocols with dozens of logical qubits and hundreds of logical teleportations using [[7,1,3]] and high-rate [[16,6,4]] codes while maintaining constant internal entropy. The 96-logical-qubit result that made headlines in late 2025 comes from this paper.

In my CRQC Quantum Capability Framework, this single paper touches B.1 (QEC), B.2 (Syndrome Extraction), B.3 (Below-Threshold Operation), C.1 (Clifford Gates), C.2 (Magic State Production), and D.1 (Algorithm Integration). No other single experiment on any platform has covered that many capability dimensions simultaneously.

The Original Logical Processor. The December 2023 paper that started the neutral-atom fault-tolerance conversation: a programmable quantum processor with up to 48 logical qubits operating on 280 physical qubits (Nature, 2024). Surface code improvement from d=3 to d=7, color codes at break-even fidelity, fault-tolerant logical GHZ states, and computationally complex sampling circuits using 3D [[8,3,2]] codes.

Continuous Operation. The capability that eliminates the biggest engineering vulnerability of neutral-atom computing: atom loss. A team led by Neng-Chun Chiu demonstrated continuous operation of a coherent 3,000-qubit system for over two hours (Nature, June 2025), using optical lattice conveyor belts that reload atoms at 300,000 per second without disrupting the quantum coherence of stored qubits. I covered this result in September 2025. In the CRQC Framework, it directly addresses D.3 (Continuous Operation), which remains an open question for every other modality at comparable scale.

Magic State Distillation. Pedro Sales Rodriguez and collaborators demonstrated logical magic state distillation on the Gemini-class system using d=3 and d=5 color codes, observing improved logical fidelity of output states versus inputs (Nature, 2025). Without magic state distillation, you cannot do universal fault-tolerant computation; you are limited to Clifford gates, which are classically simulable. This addresses C.2 (Magic State Production) in the CRQC Framework.

Low-Overhead Transversal Fault Tolerance. Hengyun Zhou, Madelyn Cain, and collaborators introduced Algorithmic Fault Tolerance (AFT), combining transversal operations with correlated decoding to cut runtime overhead by a factor of d (the code distance, typically around 30) (Nature, September 2025). For large-scale logical algorithms, this translates to 10–100× reductions in execution time. I covered this in detail on PostQuantum.com when it was published. In the CRQC Framework, AFT addresses both D.2 (Decoder Performance) and the time component of LOB/QOT.

Constant-Overhead Fault Tolerance with qLDPC Codes. Qian Xu and collaborators proposed a hardware-efficient scheme for qLDPC codes on reconfigurable atom arrays that starts outperforming surface codes with as few as several hundred physical qubits (Nature Physics, July 2024). Their simulations show quantum algorithms involving thousands of logical qubits running on fewer than 10⁵ physical qubits. This is the theoretical paper that makes the physical-to-logical ratio argument for neutral atoms: where surface codes need roughly 1,000 physical qubits per logical qubit, qLDPC codes on atom arrays get the ratio into the low tens.

Ultra-High-Rate Quantum Error Correction. Chen Zhao and collaborators demonstrated qLDPC codes with encoding rates exceeding 1/2, co-designed for neutral-atom hardware (arXiv, April 2026). I analyzed this paper when it appeared. Under circuit-level noise simulation at 0.1% physical error rate, they achieved per-logical-per-round error rates of approximately 1.3×10⁻¹³ with a [[2304,1156,≤14]] code, approaching the “teraquop” regime. This is a simulation result, not hardware, and it covers quantum memory rather than full computation. But the code family works under realistic noise, and neutral-atom hardware can support it. [EDITOR: Verify — this paper is a pre-print as of June 2026; confirm peer-review status]

The STAR Architecture. Refaat Ismail and collaborators at QuEra and Los Alamos National Laboratory published the transversal STAR architecture for megaquop-scale quantum simulation (PRX Quantum, May 2026). The architecture delivers up to 250× faster execution and roughly 2× fewer physical qubits than conventional fault-tolerant approaches for structured simulation problems. This is the paper that makes the “megaquop on 10,000 physical qubits” argument, and potentially on as few as 1,000–3,000 qubits with high-rate codes.

What Libra Is — and What It Is Not

Libra’s specs (256 logical qubits, 10⁻⁶ error rate, one million operations) place it squarely in the megaquop regime. In my Quantum Utility Map analysis, this is the threshold where quantum computing transitions from physics demonstrations to early scientific applications. Kessler’s AWS blog post specifically names quantum chemistry, high-energy physics, and materials simulation as the initial target domains.

That aligns with what I found in The Narrow Advantage: the five industries where fault-tolerant quantum computing will deliver genuine advantages are chemistry/pharma, materials science, certain optimization problems, high-energy physics simulations, and cryptanalysis. Libra’s specs put it within range of the first three on that list, at least for structured simulation problems where the STAR architecture’s 250× speedup applies. It falls well short of the fourth (which requires gigaquop-scale machines) and is irrelevant to the fifth; cryptanalysis requires billions of non-Clifford gates, orders of magnitude beyond Libra’s design.

So no, Libra is not a CRQC. The LOB gap I identified in the CRQC Scorecard — roughly 650,000× between the best demonstrated logical circuit depth and the roughly 6.5 billion Toffoli gates required to factor RSA-2048 — remains the binding constraint. Libra’s million-operation budget closes perhaps one order of magnitude of that six-order-of-magnitude gap. The remaining five orders of magnitude still separate the quantum computing industry from a machine that threatens current cryptography.

But framing Libra purely through the CRQC lens misses the point. The megaquop regime matters because of what it enables, not what it threatens. If a 256-logical-qubit machine can simulate molecular systems that classical computers cannot, even for small molecules, even slowly, that is the proof point that the entire field has been chasing. AWS is investing in a multi-year partnership because quantum chemistry and materials simulation at the megaquop scale could generate real commercial value for their customers, and they want that workload running on their cloud.

How It Compares to the Field

Every major quantum hardware company is now targeting fault tolerance in the 2028–2029 window. The convergence is striking.

IBM’s Starling targets roughly 200 logical qubits running 100 million gates by 2029, using multi-module superconducting architecture with qLDPC codes. The logical qubit count is comparable to Libra’s, and the gate count is 100× higher, but the timeline is one year later and IBM has not yet demonstrated logical qubits on production hardware. The multi-module linking required (Flamingo, Kookaburra, Cockatoo processors connected via quantum interconnects) adds integration risk that QuEra’s single-module architecture avoids.

Google’s Willow chip delivered the first convincing below-threshold result on superconducting hardware in December 2024 and remains the only superconducting platform to demonstrate below-threshold error suppression at distance 7. Google targets a useful, error-corrected quantum computer by 2029 but has been vague about specs. Quantinuum’s Helios has the best physical qubit quality of any platform (99.921% two-qubit gate fidelity across all pairs) and the best demonstrated encoding ratio (48 error-corrected logical qubits from 98 physical qubits, a 2:1 ratio), but has not published a comparable “this system, this date, these specs” roadmap for a fault-tolerant machine.

The U.S. Department of Energy’s Grand Challenge, announced in April 2026 by Under Secretary Darío Gil, set a national target for the first fault-tolerant quantum computer by 2028. The timeline alignment with QuEra’s announcement may not be coincidental.

Why the Confidence Is (Partially) Justified

Three factors make QuEra’s confidence more grounded than the typical quantum roadmap announcement.

First, every building block has been demonstrated in peer-reviewed research. Most fault-tolerant roadmaps identify four to six required capabilities and have demonstrated one to three of them. The Bluvstein et al. architecture paper demonstrated all core elements in a single system. No other platform has done this.

Second, neutral atoms scale within a single module. Kessler’s AWS blog makes this point directly: Rydberg atoms can maintain 10,000 to 100,000+ qubits within a single module without the inter-module interconnects that superconducting architectures require. The 3,000-qubit continuous operation result, combined with the 6,100-qubit pulsed array at Caltech, suggests that the physical qubit count is no longer the bottleneck. Gate fidelity, decoder speed, and control system complexity are where the remaining engineering difficulty concentrates.

Third, AWS is co-signing with specifics, not platitudes. Kessler writes about quantum processors becoming “a natural part of the AWS compute portfolio” and names specific application domains. When a hyperscaler’s general manager writes with that level of technical specificity, it carries different weight than a startup’s press release.

Where the Risks Are

The gaps are real, and I want to name them clearly.

Gate fidelity. The ultra-high-rate code simulations assume 0.1% physical error rates (99.9% gate fidelity). Current best demonstrated two-qubit Rydberg gate fidelities on neutral atoms are in the 99.5–99.7% range. That 0.2–0.4% gap may sound small, but error correction performance is exponentially sensitive to physical error rates near the threshold. Closing this gap is an engineering problem (better laser control, improved atom positioning, enhanced readout), but it is the single highest-risk item on the Libra path.

Decoder latency at scale. Real-time decoding for qLDPC codes at the 256-logical-qubit scale has not been demonstrated on any platform. The AFT framework helps by reducing the number of syndrome rounds, but the decoder still needs to process complex syndrome patterns faster than errors accumulate. This maps to D.2 (Decoder Performance) in the CRQC Framework and is an area where IBM’s investment in the Relay-BP decoder architecture gives it a potential advantage.

Integration risk. Demonstrating building blocks individually, even all of them, is different from assembling them into a production system that runs reliably for customers on a cloud platform. The distance between “we showed this works in a lab” and “this runs as a managed service on AWS with SLAs” is measured in years of engineering, not months.

Physical qubit count ambiguity. The Libra announcement specifies 256 logical qubits but does not state the physical qubit count. At the [[16,6,4]] code ratio demonstrated in the architecture paper (roughly 2.7:1), you would need approximately 700 physical qubits. For surface codes at comparable distance, vastly more. The actual architecture will likely use a mix of code types, and the physical qubit overhead depends heavily on which codes are used for which operations and what error rates the system achieves.

The Bottom Line

Every company in quantum computing announces ambitious roadmaps. What separates QuEra’s is the evidence trail. Eight peer-reviewed papers in Nature and Physical Review Letters, covering every capability dimension from error correction to continuous operation. An AWS general manager writing that there is “a clear technical pathway” and committing Braket infrastructure to deliver it. A hardware architecture that scales within a single module, avoiding the multi-chip interconnect risk that IBM and Google face.

The risks are real: gate fidelity needs to improve by 2–4×, real-time decoding at scale is undemonstrated, and integrating lab results into a cloud-accessible production system is a different discipline from publishing papers. Libra is an announcement, and announcements are free.

But this is not a CRQC announcement, and evaluating it as one misreads the situation. A megaquop machine with 256 logical qubits is aimed at quantum chemistry and materials simulation, the domains I identified in The Narrow Advantage as the genuine near-term opportunities for fault-tolerant quantum computing. The 650,000× LOB gap to breaking RSA-2048 remains. Libra closes one order of magnitude of it. The other five orders of magnitude are a problem for a different generation of machines.

Kessler put it well in the AWS blog: “The Megaquop-scale is not the finish line. It is the starting line.” QuEra’s June 24 webinar is expected to reveal what comes after Libra. I will cover that when the roadmap is public.