Unlock quantum with AI.

The AI-native infrastructure layer for scalable quantum computing. We turn quantum data into the AI advantage that scales quantum computing.

Errors are the wall between today’s devices and useful quantum computing.

Reaching roughly 10,000 physical qubits unlocks the first practically useful applications. Every hardware modality — superconducting, trapped ion, neutral atom — faces the same wall: quantum error correction.

0.37 µsFPGA atom-rearrangement latency — about 160× faster than prior published work
~8:1Physical-to-logical qubit overhead with qLDPC codes, against roughly 100:1 for surface codes
<10KParameters in the decoder that runs in real time — small enough for FPGA deployment

We don’t hand-engineer decoding rules. Our AI learns them.

Trained on real quantum data, our decoder reaches a lower logical error rate, faster, than rule-based algorithms — while staying small enough to run inside the control loop.

01

More logical qubits from fewer physical qubits

We build on qLDPC codes: one code block can carry hundreds of logical qubits, where traditional QEC yields a few or even one.

02

The yardstick is the logical error rate

Useful computation needs it below 10-11. That threshold is where rule-based decoders run out of road — and where learning takes over.

03

Learning beats rules

Published results already show AI-native decoding ahead by up to two orders of magnitude on key control tasks.

04

Small enough to be real-time

Research models use millions to billions of parameters — too large for real-time hardware. We compress to below 10K parameters and run on FPGA.

Every experiment compounds.

Quantum begins with a data-generation problem. Each experiment produces control and error data; aggregated across modalities, it trains better models, which in turn improve the systems that produce more data.

STEP 1

Generate

Every experiment creates new control and error data.

STEP 2

Aggregate

Knowledge combines across modalities — what no single hardware player can see alone.

STEP 3

Train

More diverse data produces better control, calibration and correction.

STEP 4

Compound

Better performance drives more deployments — and more data.

Selected, partnered, underway.

U.S. Department of Energy selectionSelected from more than 5,000 proposals — the top 6%.
NVIDIA — Ising ecosystem decoding partnerWorking within the accelerated-computing ecosystem for quantum decoding.

Additional partnerships are confidential and are not listed here.

Recent

Jul 2026

DOE Genesis Mission award

With UC San Diego and Berkeley Lab: a $740K+ project to advance AI-accelerated quantum computing — selected from more than 5,000 applications, 278 funded (<6%).

DOE announcement →
May 2026

74-qubit IBM hardware benchmark

Error detection benchmarked on IBM superconducting hardware at up to 74 physical qubits, mapping the practical pseudothreshold for near-term devices.

arXiv:2605.02861 →
Apr 2026

Graph Transformer Decoder released

An attention-based decoder approaching the theoretically optimal error threshold on surface codes — and the first neural scaling laws in quantum error correction. Open source.

GitHub →

The direction

AI will guide the quantum breakthrough. We are building the intelligence and control layer that every useful quantum computer will need — whichever hardware path wins first.

Building quantum infrastructure?

We work with hardware teams, research groups and infrastructure partners.