BRENNEN

Industries / Technology / Quantum computing

Cloud access

Cloud access is how most people use a quantum computer: they write a program on their own computer and send it over the internet to a quantum processor in a data centre, paying for time or using a free allowance.

4

Milestones achieved, from 2016 to 2026.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2016

    A quantum computer online

    On 4 May 2016 IBM put a five-qubit superconducting processor at its T. J. Watson Research Center on the internet as the IBM Quantum Experience; IBM's release calls it the first time quantum computing was made available to members of the public. Thousands of people used it in its first months and years. In 2026 IBM's free plan gives 10 minutes of machine time a month, and paid use starts at USD 96 a minute, billed by the second.

    uk.newsroom.ibm.comibm.comibm.com

    Credit

    IBM Research

  2. 2020

    Several makers through Amazon

    On 13 August 2020 Amazon Web Services opened Amazon Braket to all customers, offering simulators and quantum machines from D-Wave, IonQ and Rigetti through one account. Users could test programs on simulated machines before paying for real ones.

    press.aboutamazon.com

    Credit

    Amazon Web Services

  3. 2021

    Trapped ions through Azure

    On 1 February 2021 Microsoft opened Azure Quantum to the public as a preview, with trapped-ion machines from Honeywell and IonQ, the IonQ one with up to 11 fully connected qubits, and optimisation software from Microsoft and 1QBit.

    azure.microsoft.com

    Credit

    Microsoft; Honeywell Quantum Solutions; IonQ

  4. 2026

    Over 100 qubits on Braket

    On 7 April 2026 Amazon added Rigetti's Cepheus-1-108Q, which AWS calls the first superconducting processor with more than 100 qubits on Braket. It is built from twelve 9-qubit chiplets, small chips joined into one processor, and offers pulse-level control for researchers.

    aws.amazon.com

    Credit

    Rigetti Computing; Amazon Web Services

Watch for

  1. Advantage from a rented machine

    Tigran Sedrakyan, Hrant Gharibyan and colleagues, whose work BlueQubit presents on its blog, ran random-circuit sampling (generating outputs from random circuits, a standard hard test) on 61 qubits of IBM's Nighthawk r2 through the ordinary cloud service with no special calibration. Two separate ways of estimating fidelity agreed. They estimate that one million samples, which took 19 seconds, would take more than a century on the Frontier supercomputer, and call it the first such advantage on a machine most users can easily rent.

    Achieved whenan outside group reruns the circuits through the cloud and gets similar fidelity, and classical-simulation researchers accept the cost estimate.

    arxiv.orgbluequbit.io

    Claimed by

    Sedrakyan and colleagues, September 2026 (preprint)

  2. China's Tianyan cloud

    The Tianyan Quantum Group says its public cloud offers Tianyan-287, a 105-qubit superconducting processor with two-qubit gate fidelity of 99.56%, and that it produced one million samples of a 74-qubit random circuit in 18.4 minutes, a task it estimates would take a supercomputer about 16,000 years. Users program it with Cqlib, an open-source toolkit.

    Achieved whenoutside users publish their own fidelity measurements on Tianyan-287, or a classical team checks the 16,000-year estimate.

    arxiv.org

    Claimed by

    Tianyan Quantum Group, December 2025

Open

  1. 01

    Cloud papers, rerun

    Take 25 published experiments that ran on public cloud machines between 2017 and 2024 and shared their code. Get each running again, rerun it on today's machines, and report which conclusions still hold.

    The brief
    Credit · unclaimed
  2. 02

    One job on every cloud

    Run one fixed set of small circuits every week for a year on every quantum machine anyone can rent, and publish for each run the queue wait, price paid, run time and result quality, next to the provider's posted calibration figures at that moment.

    The brief
    Credit · unclaimed

Open 01 · the brief

Cloud papers, rerun

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

Take 25 published experiments that ran on public cloud machines between 2017 and 2024 and shared their code. Get each running again, rerun it on today's machines, and report which conclusions still hold.

Why it is still open

In July 2026 Wolfgang Mauerer's group checked 127 quantum computing papers by hand: 24.4% shared code, and 64.5% of those failed to run in a clean setup. An automated screen of nearly 5,000 papers found code for 26.8%. The study checked whether code was shared and would run, not whether results hold on current hardware.

arxiv.org

Built with what exists

Current Qiskit and the Braket SDK; containers such as Docker to fix each setup; a coding agent such as Claude Code or GitHub Copilot to port old code (the Note "Effort, not model" covers GitHub Copilot Fusion); IBM's free 10 minutes a month and research credit programmes.

  1. First step

    Pick five papers from the study's sample that shared code for IBM hardware and get the first one running in a container.

  2. Done looks like

    25 papers, each with a public package that reruns from one command, a hardware rerun, and a verdict of holds, differs or cannot be rerun, checked by a second person; authors told before publication.

Open 02 · the brief

One job on every cloud

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

Run one fixed set of small circuits every week for a year on every quantum machine anyone can rent, and publish for each run the queue wait, price paid, run time and result quality, next to the provider's posted calibration figures at that moment.

Why it is still open

A January 2026 study by researchers at ARLIS recorded connection, run-time and cost data across cloud services, but for three months. Metriq publishes benchmark results across vendors, and IBM lets users download its current calibration figures. None keeps a continuing public record of wait, price and quality together.

arxiv.orgarxiv.orgquantum.cloud.ibm.com

Built with what exists

metriq-gym, the Unitary Foundation's open-source runner, which includes GHZ-state, mirror-circuit and layer-fidelity tests; the IBM, Braket and Azure software kits; GitHub Actions to run on a schedule; a small budget or research credits.

unitaryfoundation.github.io

  1. First step

    Run metriq-gym's GHZ test once on IBM's free plan and once on a Braket machine, and log the wait and the cost of each.

  2. Done looks like

    52 weekly runs on at least 6 machines from at least 3 clouds, with raw data public and every gap explained.