NVIDIA launches Ising — the first open AI models for quantum computing — with day-one adoption by IonQ, IQM, Atom Computing, Infleqtion, and Q-CTRL
On World Quantum Day 2026-04-14, NVIDIA announced Ising, billed as the world's first open AI-model family for quantum computing. The initial two open models (Ising Calibration 1 and Ising Decoder SurfaceCode 1) target quantum error correction decoding and system calibration respectively, with reported 2.5× decoding speed-up and 3× accuracy improvement, plus automated calibration reducing setup from days to hours. Models are distributed via Hugging Face and NVIDIA NIM, with training-data generation on cuQuantum and cuStabilizer. Adopter roster at launch: IonQ, IQM Quantum Computers, Atom Computing, Infleqtion, and Q-CTRL.
Material platform release: open-source AI-for-quantum tooling from the dominant AI-infrastructure vendor, with day-one adoption by five tracked OEMs. Scored 7 on the platform-expansion tier, adjacent to IBM Qiskit-Runtime launches and AWS Braket expansions — a concrete developer-ecosystem advance, not a fault-tolerance threshold crossing. The 2.5× decoder speedup and 3× accuracy gain are incremental improvements over prior neural-decoder work (e.g. the AlphaQubit class of results), not a real-time-at-scale decoding capability that would unblock a class of FTQC experiments previously infeasible; the calibration automation is a workflow quality-of-life win rather than a landscape move. Below the score-8 FIPS-finalization anchor because Ising has no regulatory forcing function, and below the score-8 H2-class-hardware anchor because it is tooling rather than named-processor architecture.
Compresses the QEC-decoder software race (previously an in-house moat for companies like Riverlane, Qedma, and IBM) into an open baseline, shifting differentiation to hardware and custom topology specializations. Likely catalyzed the April-14-through-mid-April rally across the quantum-sector stock complex.