A quantum hardware vendor and an HPC analyst say the bottleneck is software. Read the small print on who is asking
There are now two documents in the Alice & Bob and Hyperion Research series on quantum and high-performance computing, and they are frequently conflated. Getting them apart matters, because the coverage of each has recycled figures from the other.
The first, "Seizing Quantum's Edge: Why and How HPC Should Prepare for eFTQC," was announced in a joint press release datelined St. Paul, Minnesota and Paris on Sept. 8, 2025. The second, "HPC-Quantum-AI: Shaping the Next Compute Era," was covered on Sept. 9, 2026 by The Quantum Insider and EE Times, and is the one that generated this week's headlines. The widely quoted claims that up to half of current HPC workloads at US government research labs could benefit from the speedups of early fault-tolerant quantum computing, that the resource requirements for algorithms like Shor's have been compressed by an estimated factor of 1,000, and that machines with 100 to 1,000 logical qubits at logical error rates between 10⁻⁶ and 10⁻¹⁰ arrive within five years — all of those belong to the 2025 document, not the 2026 one. They appear in the 2025 press release, and Data Center Frontier's write-up of that first report, published Sept. 12, 2025, carries them. Neither the Quantum Insider nor the EE Times coverage of the 2026 report repeats any of the three.
What the 2026 report adds is a different diagnosis. Where the 2025 edition argued that quantum accelerators were coming and centres should get ready, the 2026 edition argues that the hardware has in fact moved and the software has not.
The research base is interviews. According to The Quantum Insider's account, the 2026 report draws on 15 leaders at supercomputing institutions including Argonne National Laboratory, Lawrence Berkeley National Laboratory and NERSC, Oak Ridge National Laboratory, RIKEN in Japan, Munich Quantum Valley and the Pawsey Supercomputing Centre in Australia. EE Times, covering the same release, names Oak Ridge, Berkeley and RIKEN among the 15 and describes the sample as directors and engineers. Fifteen interviews is a defensible qualitative sample for this kind of exercise; it is not a survey, and no one has presented it as one.
Juliette Peyronnet, general manager for the US at Alice & Bob, put the thesis to EE Times as follows: the quantum industry and the hardware have been progressing quite significantly, but one gap identified in the work is in regard to the software stack. In the Quantum Insider account she framed the same point differently, saying that what determines whether a quantum processor delivers value is its ability to execute applications.
That second formulation is worth pausing on, because it is a competitive claim dressed as an observation. Alice & Bob builds cat-qubit superconducting hardware and has spent two years arguing that raw physical qubit counts are a poor proxy for capability. A report concluding that qubit counts matter less than the surrounding stack is a report whose conclusion aligns with its sponsor's positioning. That does not make it wrong. It does mean the finding should be read as advocacy supported by interviews rather than as an independent audit.
Hyperion Research, for its part, is an established HPC market analyst firm, and Bob Sorensen, its senior vice president of research and chief analyst for quantum computing, is quoted in both editions. Sorensen told EE Times that HPC end users everywhere are looking for guidance and insights for what it will take to integrate the computational power of quantum into the current and projected advanced computing ecosystem. In the 2025 release he called quantum technologies a pivotal opportunity for the HPC community.
The 2026 report's three recommendations, as reported by The Quantum Insider, are concrete enough to be testable. First, centres should navigate the quantum-HPC-AI design space by testing how tightly quantum has to couple to classical and AI compute. Second, the software stack should be unified through open standards covering device management and job submission. Third, workflows should be co-designed with HPC staff, AI specialists and quantum vendors working together, with the report citing a quantum chemistry workflow example targeting a 10,000-fold speed-up.
The second recommendation is the one with teeth, and also the one most likely to be resisted. Open standards on device management and job submission across vendors would let a supercomputing centre swap one QPU for another without rewriting its stack. That is unambiguously good for the centres. Whether it is good for any individual quantum vendor depends entirely on whether that vendor expects to win on hardware. A company confident in its physics has every reason to push for interchangeability at the software layer.
The coupling question connects to work Alice & Bob published separately in June. On June 26, 2026, Kevin D. Kissell, credited as a senior architect at the company, put out a document Quantum Computing Report describes as a computer architecture blueprint for decoupled AI topologies. As the write-up has it, the budget to decode error syndromes and execute feedback before decoherence is bound to a rigid one-microsecond threshold, met by processing raw syndrome measurements inside FPGA or ASIC blocks adjacent to the qubit readout hardware; a second, asynchronous loop is exempt from that budget and lets GPUs and heavier machine-learning classifiers work on longer data blocks, routed over NVIDIA's NVQLink and CUDA-Q using remote direct memory access.
The same write-up records that quantum low-density parity-check codes compress physical-to-logical qubit overhead from as much as 1,000 to 1 down to something nearer 100 to 1 — a figure it attributes to the wider field rather than to any measurement of the company's own. Readers should note the provenance either way: this is an architecture proposal authored by a vendor, and nothing in the write-up reports a benchmark comparison, a named party who ran one, or a measured performance result. It is a design argument, not a measurement.
Taken together, the vendor's June architecture paper and the September report tell a coherent story: hardware control loops are tight, classical decoding is the constraint, and the fix is co-design between quantum vendors and the people who run supercomputers. Alice & Bob has been building the commercial relationships that story implies. Its own newsroom lists a partnership with Bull to bring quantum computers into HPC on June 24, 2026, and the unveiling of its first quantum system, Helium, on June 18, 2026.
For anyone tracking this sector as an investment theme rather than a research one, the practical takeaway is narrow. There is no listed pure-play here — Alice & Bob is private, Hyperion is an analyst firm — and the report contains no revenue figures, no procurement commitments and no named contracts. Its value is as a read on what national laboratory buyers are actually worried about, and this year they are apparently worried about integration plumbing rather than qubit counts. That is a slower and less headline-friendly problem than a new processor announcement, and it is the sort of thing that shapes procurement two and three years out.
Markets gave the theme no obvious verdict this week. The last completed session was Tuesday, Sept. 8, when the Nasdaq Composite closed at 26,421, down 0.32%, and the Russell 2000 at 2,960, down 0.52%, per Investrade's review of the day. As of roughly 11:15 a.m. ET on Wednesday the US session remained open and had produced no closing level.
