How this actually runs
We do not build quantum hardware and we are not going to. The group works one layer up: how a classical GPU fleet schedules around, verifies and absorbs work from partner QPUs. Four researchers, part-funded by rental margin, work alongside the GPU fleet. Results fold into the production scheduler when they hold up under load. Nothing here is a standalone SKU.
Hybrid scheduling treats a quantum call as a long-latency operation so the GPU queue does not sit idle waiting for the round-trip. Error-mitigated sampling is classical post-processing that makes noisy intermediate-scale output usable for small optimisation problems, run on the same fleet. Simulator capacity is already in service for circuits that are not worth a QPU slot.
Access is not an open queue. Partners get time because the problem is real and the classical fleet can absorb it. You do not buy quantum hours. You may get a research collaboration plus ordinary GPU rent; partner QPU time is through those providers’ own contracts. If the honest answer is ‘use the simulator’, that is the answer.
Notes are published when there is something to say - methods, failure modes, what we will not claim. There is no weekly blog. Citing a note does not imply we endorse a commercial quantum product. Scientific computing customers who just need CUDA on H100 / A100 use Rentals like everyone else; this programme is optional and small.
What a note is
Methods, failure modes, what we will not claim. No announcement prose.
There are not weekly posts. Silence means the work is still in the lab.
Notes are public. Access to run on the hybrid path is not automatic - that is the access model.