> ## Content Index
> Fetch the complete content index at: https://www.insidedeeptech.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The 12 Most Important Quantum Computing Companies in 2026
- URL: https://www.insidedeeptech.com/the-12-most-important-quantum-computing-companies-in-2026-ranked/
- Published: 2026-06-26T11:34:11.000Z
- Updated: 2026-08-27T00:35:55.000Z
- Description: IBM says it has more quantum computers deployed than the rest of the industry combined. Most of them sit in its own data centres. Seven of these twelve companies have machines standing on floors they do not own. Five have none.
- Author: Irina Peskisheva
- Tags: Quantum

IBM said in June that it has deployed more than 90 quantum systems worldwide, more than the rest of the industry combined. The claim is accurate and it hides something. Most of those machines sit in IBM's own data centres and are reached through the cloud.

Count only the machines standing on somebody else's floor, and the leader is a Finnish company founded in 2018 that appears on almost no list of this kind. IQM had delivered 17 systems to customers as of 30 June 2026, a figure from the company's own management report for the half year. IQM has been listed on Nasdaq since July.

That division runs through this page. Seven of the twelve companies below have machines standing on floors they do not own. Five have none, and two of those five hold the results the rest of the field measures itself against.

## How these twelve were chosen

Every company here designs and builds its own quantum processors. That leaves out firms whose business is reselling access to other people's machines, which is why Amazon appears nowhere below despite running one of the largest quantum cloud platforms.

Every company here also has a public record that can be checked against a document: a filing with a securities regulator, a peer-reviewed paper, or an announcement made by the customer as well as by the vendor. A press release on its own was not enough.

No order is imposed on the twelve, because the measures that would impose one do not survive contact with the field.

Qubit count fails first. The 156 superconducting qubits on an IBM Heron, the 98 trapped ions in a Quantinuum Helios and the more than 5 000 annealing qubits in a D-Wave Advantage are three different physical quantities, and putting them on one axis produces a ranking of vocabulary.

Two-qubit gate fidelity is comparable in principle and unusable in practice. Quantinuum publishes 99,921% measured across all qubit pairs. Most vendors publish their best pair. Annealing has no gate fidelity at all.

Logical qubits are the metric the field itself moved to, and they split in half. Quantinuum's Helios reports [94 logical qubits with error detection and 48 with error correction](https://www.insidedeeptech.com/the-state-of-quantum-computing-2026-a-field-report/), which are two numbers describing one machine. Order the field by that number and Google lands near the bottom while holding the strongest error-correction result anyone has published.

## The line that divides them

What does survive is a plain question about where the machines are.

A system counts as installed when it stands somewhere the manufacturer does not own. A machine in the vendor's own data centre, sold as cloud access, does not count, however large the customer base reached through it. That rule costs several companies here, and it is the only way the count stays comparable across superconducting circuits, trapped ions, neutral atoms and annealing.

Seven companies clear it. Five do not.

One warning about what the count does not say. A machine with 54 physical qubits and a machine with 156 physical qubits each count as one system. Delivery is a commercial fact, and it is not a claim about capability. Each entry below carries the size and the technology alongside the count for that reason.

The counts are also not all measured the same way, and the difference matters. IQM's figure is cumulative deliveries since the company was founded. IBM's is the number of third-party sites it currently names. Rigetti's covers 2026 alone, because the company has published no lifetime total, and D-Wave's is a floor for the same reason. Each entry says which it is.

The seven appear in order of those counts. That is an ordering of deliveries, and it is not a statement about which machine is better. The second group on this page exists because those two things come apart.

## Seven companies with machines on somebody else's floor

### 1\. IQM

Superconducting. 17 systems delivered to customers as of 30 June 2026, with 26 sold since the company was founded and eight more in production. Revenue for the first half of 2026 reached 8,9 million euros against 6,0 million a year earlier, and the order book stood at 67,3 million euros entering the year.

The Leibniz Supercomputing Centre near Munich runs four IQM machines, including Euro-Q-Exa, a 54-qubit Radiance system that went into service in February 2026\. Deliveries in the past year went to CINECA in Italy, to universities in Finland, Germany and Poland, and to Oak Ridge National Laboratory in Tennessee.

IQM sells machines outright to institutions that want to own and run them. That business model is why it heads the count and why it is almost absent from comparisons built on qubit counts, where 54 qubits looks unremarkable next to numbers three times larger.

### 2\. IBM

Superconducting. Seven third-party sites named by the company: Cleveland Clinic, Rensselaer Polytechnic Institute, PINQ² in Quebec, the University of Tokyo, RIKEN, Yonsei University in South Korea, and BasQ in Spain. Systems in Chicago and at Amaravati Quantum Valley in India were announced as coming.

The 156-qubit Heron processor is the workhorse of that fleet. In May 2026 two Heron machines, one at Cleveland Clinic and one at RIKEN, ran the quantum part of a simulation of a 12 635-atom protein alongside the Fugaku and Miyabi-G supercomputers, using up to 94 qubits and close to 6 000 quantum operations.

IBM committed more than 10 billion dollars to quantum in June 2026 and has published a roadmap that runs to Starling in 2029, a machine it describes as the first large-scale fault-tolerant quantum computer. Starling does not exist. The seven installed sites do.

### 3\. Pasqal

Neutral atoms. Three systems under the EuroHPC Joint Undertaking, out of eight quantum computers deployed under that programme in total: at CEA and GENCI in France, at the Jülich Supercomputing Centre in Germany, and at CINECA in Bologna, where a machine of 140 physical qubits arrived in February 2026 for integration with the Leonardo supercomputer.

Two of the three, Jade and Ruby, are analogue simulators. They run a narrower class of problem than a gate-based machine, and they are physically installed and in use, which is what this page counts.

### 4\. D-Wave

Annealing, with gate-based work underway. D-Wave sold its first Advantage system outright in the fourth quarter of 2024, an event it reported to the Securities and Exchange Commission in January 2025 as a change to its revenue model. The Jülich centre operates one. The company has not published a total for customer-owned machines, so the count here is a floor and not a figure.

Advantage carries more than 5 000 annealing qubits with 15-way connectivity. Those qubits solve optimisation problems by settling into low-energy states and they do not run gate-based circuits, so the number belongs in a different column from every other on this page.

The Leap cloud service reaches 42 countries. D-Wave also acquired Quantum Circuits, whose dual-rail superconducting qubits carry error detection in the hardware, which is the company's route into the gate-based market.

### 5\. Rigetti

Superconducting. Two nine-qubit systems delivered in 2026, both to commercial buyers, which management described as a change from a market that had been almost entirely government and academic. In August 2026 Rigetti created a separate Systems Delivery organisation for manufacturing and installing machines at customer sites.

The nine-qubit size is worth reading carefully. These are research instruments bought for experimentation, and Rigetti says as much. The company's larger Cepheus-1-108Q carries a target of 99,5% two-qubit fidelity, which is a target and not a result.

### 6\. IonQ

Trapped ions. One confirmed customer-owned installation, at QuantumBasel in Switzerland, which took delivery of a Forte Enterprise system in August 2024 and secured outright ownership of it in December 2025 under an agreement worth more than 60 million dollars. A 100-qubit Tempo system for the Korea Institute of Science and Technology Information was agreed in December 2025 for integration with the KISTI-6 cluster.

IonQ reports a two-qubit gate fidelity of 99,99% and has set a target of two million qubits by 2030\. It completed a 1,8 billion dollar acquisition of the semiconductor foundry SkyWater Technology in July 2026 and guides to revenue of 280 to 290 million dollars for the year, the largest of any company on this page.

Revenue and installed systems point in opposite directions here, because most IonQ customers reach the hardware through Amazon Braket, Microsoft Azure and Google Cloud.

### 7\. Quantinuum

Trapped ions. One installation outside its own facilities: a Helios system going into an Oracle Cloud Infrastructure data centre under an agreement announced in August 2026.

Helios is the most accurate machine in commercial service by the measure its maker publishes. It carries 98 physical qubits with all-to-all connectivity, a two-qubit gate fidelity of 99,921% measured across every pair, and single-qubit fidelity of 99,9975%. It reports 48 logical qubits with error correction and 94 with error detection.

Quantinuum listed on Nasdaq on 4 June 2026, selling 28 million shares at 60 dollars and raising 1,68 billion, the first traditional flotation by a full-stack quantum company. The prospectus that made that possible also disclosed revenue of 5,24 million dollars for the first quarter of 2026, down from 19,1 million a year earlier. A 73% fall in the quarter before a debut that closed with a market value above 15 billion dollars is the sharpest illustration available of how far ahead of revenue this market prices.

## Active, with no system at a customer site

Five companies on this page have installed nothing outside their own walls. That is a statement about business model and stage, and for two of them it says nothing at all about importance.

### Google

Google sells no quantum hardware and holds the result the rest of the field measures itself against. Its Willow processor ran a surface code below threshold, with logical error suppressed by a factor of 2,14 ± 0,02 for each increase of two in code distance, and an error per correction cycle of 0,143% ± 0,003% for a distance-7 code on 101 qubits. The work appeared in Nature in December 2024 and it is the reason error correction stopped being a projection.

### PsiQuantum

Photonic, and built on the argument that fault tolerance has to be designed in from the start and manufactured in an existing semiconductor fab. PsiQuantum has no machine at a customer site and publishes no benchmark result that can be set against the systems above. The bet is that the first useful machine will arrive whole.

### Microsoft

Microsoft runs Azure Quantum as an access layer for other companies' hardware and pursues topological qubits of its own. Its nearest deliverable is Magne, built with Atom Computing, described as around 50 logical qubits from roughly 1 200 physical atoms and scheduled for early 2027 with delivery to EIFO and the Novo Nordisk Foundation. That schedule has not been met yet, because it has not arrived yet.

### QuEra

Neutral atoms. QuEra and collaborators at Harvard reported 96 logical qubits at code distance 4 on 448 physical atoms in Nature, the largest confirmed logical qubit count published by anyone. The company's commercial machines are reached through cloud services.

### Intel

Silicon spin qubits. Intel distributes its 12-qubit Tunnel Falls research chip to laboratories, among them the Laboratory for Physical Sciences, Sandia National Laboratories, the University of Rochester and the University of Wisconsin-Madison, and it sells no systems. In July 2026 Hitachi, Intel's Japanese subsidiary and AIST began a NEDO-funded programme to build spin qubit processors on Intel's 18A process, targeting a 100-qubit prototype in the 2028 financial year and a 1 000-qubit platform in 2030.

Twelve qubits distributed to universities is a research programme. Intel is on this page because a foundry deciding that quantum belongs on its production line matters more over ten years than most of the announcements above.

## What changed in 2026

For most of this industry's life, comparing these companies meant reading press releases. That ended this year.

Quantinuum listed on Nasdaq in June. IQM listed on Nasdaq and in Helsinki in July, the first European quantum company to trade publicly. IonQ, D-Wave and Rigetti were already listed. Five of the twelve companies here now file with a securities regulator, which means the numbers in their filings carry a legal consequence that a press release does not.

The filings are already more informative than the marketing. IQM's delivery count comes from a management report. Quantinuum's revenue decline comes from its own prospectus. Neither number would have appeared in an announcement. The same thing happened outside quantum this year. [Unitree's filing for its Shanghai listing](https://www.insidedeeptech.com/unitree-humanoid-robot-filing/) did it for humanoid robots.

## Questions readers ask

### Which company has the most quantum computers?

IBM, by its own count of more than 90 systems deployed worldwide as of June 2026\. Most of those are reached through IBM's cloud. Counting only machines installed at a customer's own site, IQM has the most, with 17 delivered as of 30 June 2026.

### Which companies sell quantum computers outright?

IQM, D-Wave, IonQ, Rigetti and Pasqal have all sold machines outright to institutions that host and operate them. Most buyers so far are supercomputing centres, national laboratories and universities. Rigetti reported its first two commercial buyers in 2026, both purchasing nine-qubit systems for research.

### Which quantum computer is the most accurate?

Among machines in commercial service, Quantinuum's Helios, at 99,921% two-qubit gate fidelity measured across all pairs. IonQ reports 99,99% for its own systems. The two figures are measured differently and should not be read as a like-for-like ordering.

### How many logical qubits exist?

The largest confirmed count is 96, at code distance 4 on 448 physical atoms, reported by QuEra with collaborators at Harvard in Nature. Quantinuum reports 48 with error correction on Helios. Every larger number in circulation is a roadmap.

### Is any of this making money?

Very little. Quantinuum reported revenue of 30,9 million dollars for 2025 and a net loss of 192,6 million. IQM reported 8,9 million euros for the first half of 2026\. IonQ guides to 280 to 290 million dollars for 2026, the largest figure in the field and one that includes businesses beyond quantum computers.

### Why is Google in the second group?

The first group counts installed machines and Google sells none. Its below-threshold error correction result on Willow is the most consequential piece of work any company on this page has published. One list cannot carry both facts, which is why this page is divided in two.

### What happened to Intel and Microsoft?

Both are still in the field and neither sells a system. Intel ships research chips to laboratories and has moved its manufacturing argument onto the 18A process with Hitachi. Microsoft runs an access platform for other people's hardware and has a machine of its own scheduled for 2027.

## What to watch

Watch whether IQM's delivery count keeps climbing at the rate its order book implies. Eight systems were in production at the end of June, and a company that converts those cleanly will have roughly doubled its installed base inside two years.

Watch Quantinuum's next two quarters. The first quarter of 2026 brought in 5,24 million dollars. A company valued above 15 billion has to show that number moving, and the prospectus is now a public benchmark against which it will be measured.

Watch what happens when the first customer-owned machines come up for replacement. Nobody in this industry has yet been through a refresh cycle with a customer who owns the hardware. That will reveal more about the economics than any qubit count.

## Sources

Figures for IQM come from the company's management report for the half year ended 30 June 2026 and its own announcements. Quantinuum figures come from its Form S-1 registration statement and the pricing of its Nasdaq offering on 4 June 2026\. D-Wave's first outright system sale comes from a Form 8-K filed with the Securities and Exchange Commission in January 2025\. IBM's fleet size, site list and investment commitment come from its announcement of 2 June 2026, and the Cleveland Clinic and RIKEN simulation from the joint announcement of 5 May 2026\. Google's error correction result is published in Nature, volume 638, pages 920 to 926\. QuEra's logical qubit count is published in Nature under doi 10.1038/s41586-025-09848-5\. Remaining figures come from each company's own announcements and are attributed in the text.

Where a number is a company statement and not an independently verified result, the text says so. Where a company has not published a total, the count is described as a floor. No figure on this page comes from an aggregator.