Haiqu Built an Agentic OS for Quantum. Most of It You Cannot Verify.

Haiqu launched an agentic operating system for quantum R&D on May 6, 2026 — and is asking customers to trust it with real money based on claims nobody can verify independently.

The company says its middleware can cut the cost of quantum computing workflows by 100x to 1000x. Those claims are Haiqu's. The software generating them is proprietary, and Haiqu has not published the code or submitted it for independent audit. The only component of Haiqu's stack that outside researchers can actually examine is Rivet, an open-source transpiler for quantum circuits on the company's public GitHub account. Rivet handles circuit translation — converting quantum programs written in one framework into a form that works with another. It does not handle circuit scheduling, job orchestration, error mitigation decisions, or runtime management. Those are the layers where the cost claims would have to originate, and they are not on GitHub.

The gap matters because quantum middleware sits between researchers and the quantum hardware they rent by the minute. If the middleware is doing something genuinely novel — better circuit scheduling, smarter error mitigation, more efficient compilation — that could be worth paying for. If it's a well-tuned wrapper around existing tools, the economics look very different. Without an auditable baseline, there is no way to know.

The teams Haiqu is targeting are the ones currently running quantum experiments on raw Qiskit or similar frameworks. That workflow means booking time on remote quantum processors, manually optimizing circuits, managing error mitigation, and rewriting portions of the stack every time a new hardware update lands. A working abstraction layer that handles circuit optimization and job orchestration automatically could eliminate a real bottleneck — which is why the claimed cost reductions are worth scrutinizing carefully rather than taking at face value.

If such a layer works at the claimed price point, the competitive bottleneck shifts: advantage goes not to whoever has the most quantum hardware access, but to whoever can best build on top of good middleware. That is the same dynamic that played out in cloud computing, where value migrated from owning servers to knowing how to deploy software efficiently. The specific skill that matters now is circuit scheduling optimization — knowing how to structure a quantum program so it finishes within hardware coherence times and minimizes wasted shots. Teams that can do that well will get more useful results from the same machine. Haiqu is betting that its proprietary layers automate that skill.

Two scientific demonstrations accompanied the launch. The first ran the Single-Impurity Anderson Model on 40-qubit hardware using sample-based Krylov quantum diagonalization to identify the ground state of a 20-site SIAM system Haiqu blog. The second recovered the magnetic excitation spectrum of CuDCl, a quasi-one-dimensional spin-1/2 Heisenberg antiferromagnet, on an IBM quantum processor using 160 circuits and thousands of two-qubit gates in under 10 minutes of QPU time — the processor time quantum computers charge researchers to rent The Qubit Report. A molecular dynamics simulation Haiqu estimates previously cost roughly $30,000 and required more than nine hours of quantum processor time; Haiqu says the same calculation now runs in about 30 seconds for approximately $25 The Qubit Report.

Haiqu is not without credentials. The company published a peer-reviewed paper with HSBC in Physical Review Research in April 2026 demonstrating efficient encoding of real-world probability distributions into quantum circuits HPCwire. Founded in 2022 by Stanford-trained engineer Richard Givhan and former Max Planck/Weizmann researcher Mykola Maksymenko, Haiqu raised an $11 million seed round in January 2026 The Quantum Insider. Early access has gone to Capgemini and Deloitte The Quantum Insider.

The enterprise-access claims and scientific demos are real. The magnitude of the cost reduction is a company estimate attached to closed-source software. Rivet — the circuit transpiler on GitHub — is the only part of the stack independent researchers have been able to verify for themselves.