A billion dollars in quantum revenue sounds like a company that has cracked the market. Then you read the fine print. Pierre Jaeger, IBM Quantum’s CTO, confirmed the figure, and it is a real headline: nobody else in the industry has publicly claimed anything near that commercial threshold. But it is cumulative, built up over multiple years, and at least one informed industry observer has publicly noted that it folds in unconditional grants counted as revenue. IBM has not disclosed the annual breakdown or when it even started tracking quantum revenue.
Set it against the rest of the ledger and the scale gets brutal. IBM reported $62.8 billion in full-year 2024 revenue, which makes quantum a rounding error in the P&L. The generative AI book of business hit $5 billion cumulative in that same reporting period and was growing roughly $2 billion per quarter. Quantum took years to reach one-fifth of that.
That does not make the number empty. It reframes it. This is a signal about ecosystem maturity and enterprise commitment, not a statement about how big quantum computing is as a business today. To understand what IBM has actually built, you have to go back a lot further than the 2025 earnings call.
IBM’s connection to quantum computing predates the field itself. Richard Feynman’s 1981 proposal that quantum mechanical systems could simulate physics problems intractable for classical machines came out of a conference co-organized with MIT, but Feynman had deep ties to IBM through the Physics of Computation series that seeded that work. The more concrete claim to lineage came in 1993, when IBM researchers Charles Bennett and colleagues published the first experimental demonstration of quantum teleportation, showing that quantum information could move between particles without the particles physically moving. That was foundational theory rather than engineering, and it put IBM researchers at the center of the field’s intellectual development.
The engineering pivot came in 2016, and in hindsight it defined the entire commercial strategy. IBM put a 5-qubit quantum processor on the cloud and let anyone use it for free. IBM Quantum Experience, now IBM Quantum Platform, was the first publicly accessible quantum computer in the world, and it worked as a research tool, a talent pipeline, and a market-creation play running years ahead of any actual market. By making quantum accessible before it was useful, IBM seeded a whole generation of researchers, students, and enterprise tinkerers with hands-on time on IBM hardware running IBM’s software. The Qiskit open-source SDK launched alongside it and became the most widely used quantum programming framework in the world.
The commercial layer got formalized in 2017 with the IBM Q Network, a membership program that gave enterprises, national labs, and universities dedicated cloud access plus Qiskit support. It now runs to over 100 members across Fortune 500 companies, startups, and government research institutions. Years of contracts, partnerships, and access fees like these are the structural foundation of that billion-dollar figure. It was never a sudden breakthrough.
The hardware story ran in parallel, and IBM’s processor roadmap has been unusually public and unusually consistent, which in an industry drowning in vaporware is itself a differentiator. Canary opened the line at 5 qubits in 2016. Then came Falcon, Hummingbird, and Eagle, the 127-qubit chip that broke the 100-qubit barrier in 2021. Osprey followed at 433 qubits in 2022, Condor at 1,121 qubits in 2023. Few organizations have matched that kind of sustained execution. The IBM Q System One, unveiled at CES 2019, was the first commercially integrated quantum system, a cylindrical cryogenically isolated unit built to leave the research lab.
Inside one of those machines sits a specific bet. IBM uses superconducting transmon qubits, a circuit design that encodes quantum information in the energy states of a Josephson junction. That differs from the trapped-ion systems at IonQ and Quantinuum, from photonic systems, and from the topological qubits Microsoft is chasing, and each of those approaches has its own tradeoffs.
Transmon qubits run at roughly 15 millikelvin, achieved with dilution refrigerators, colder than the cosmic microwave background and colder than interstellar space. That extreme cooling is not optional: thermal noise at room temperature would flatten the quantum states instantly. The refrigerators are large, expensive, and mechanically fussy, which is exactly why a quantum computer looks like an elaborate chandelier of gold-plated copper tubing. Q System One folds the refrigerator, control electronics, and shielding into one unit, but the cryogenics are non-negotiable.
Connectivity follows a heavy-hex lattice, where each qubit touches at most three neighbors in a hexagonal grid stitched together with extra “bridge” qubits. Fewer connections means correlated errors are less likely to propagate. A fully connected topology would let you build more complex circuits, and it would also let errors spread freely, so the heavy-hex layout trades circuit flexibility for lower error rates. On noisy hardware, that trade is worth a great deal.
Error is the whole problem. Quantum states are fragile, and any interaction with the environment causes decoherence, collapsing superpositions into plain classical states before the computation finishes. IBM’s current answer is error mitigation: run the circuit many times, then use statistical methods to infer what the ideal noiseless result would have been. That is not error correction. Full fault-tolerant error correction encodes one logical qubit across many physical ones, using redundancy to catch and fix errors in real time. Which is why Condor’s 1,121 physical qubits do not add up to a 1,121-logical-qubit computer. It is 1,121 physical qubits with error rates that still rule out most useful computation.
That gap drove one of the more revealing moves in IBM’s recent strategy. Condor was the headline, the first chip past 1,000 qubits, and then IBM’s own engineers pivoted the emphasis to Heron r1, a 133-qubit processor with substantially lower error rates. Heron is modular, built with fixed-frequency qubits and tunable couplers that cut the crosstalk plaguing larger chips. Vanguard’s collaboration with IBM, exploring bond-portfolio construction under real constraints like liquidity requirements and regulatory limits, ran on Heron r1. HSBC’s quantum work on bond-trading efficiency, which reportedly showed a 34% improvement in trading efficiency metrics, leaned on current-generation hardware too. Neither is quantum advantage over a classical machine. They are proofs of concept that quantum approaches can slot into real financial workflows.
To score its own systems IBM built two metrics. Quantum Volume rolls qubit count, connectivity, gate fidelity, and measurement error into a single figure: a 2^n circuit the system can run with better than 2/3 probability of success defines a Quantum Volume of 2^n. CLOPS, Circuit Layer Operations Per Second, is a throughput metric added later because Quantum Volume tells you nothing about how fast a system chews through repeated circuits. Both exist to drag the industry conversation off raw qubit counts, which are trivially easy to inflate and easier still to misread.
The roadmap to fault tolerance centers on Starling, targeted for 2029 and described as IBM’s first large-scale fault-tolerant quantum computer. The claim attached to it is a 20,000x performance improvement over today’s systems, a figure that captures the distance between error-mitigated NISQ computation and genuine logical-qubit computation with error correction.
That target is four years out, and the whole industry is sprinting at the same line. Google’s Willow chip, announced in late 2024, demonstrated exponential error suppression as qubit count scales: adding more physical qubits to the error-correcting code actually lowers the logical error rate, which is precisely the theoretical requirement for fault tolerance to function at all. That was a serious experimental result. Google’s separate “quantum supremacy” and “beyond classical” claims have long been contested, since the benchmark problems were picked specifically to be hard for classical machines and easy for quantum ones.
Microsoft is playing a different game entirely with topological qubits built on Majorana zero modes, which should be more inherently stable than transmons because their quantum information is encoded non-locally. The catch is enormous: Majorana-based qubits have not been demonstrated at scale, and Azure Quantum currently runs on third-party hardware from IonQ, Quantinuum, and Rigetti while the topological program matures. If topological qubits work as advertised, they could need far fewer physical qubits per logical qubit than any superconducting approach. That “if” is doing a lot of load-bearing.
IonQ and Quantinuum run trapped-ion qubits, which hit higher gate fidelity than superconducting qubits at today’s scales but operate more slowly and hit their own scaling walls. Quantinuum’s H-series claims the highest two-qubit gate fidelity in the industry. IonQ trades on the NYSE and ships through AWS Braket, Azure Quantum, and Google Cloud. D-Wave sits off in its own corner, using quantum annealing rather than gate-based computing, a paradigm aimed squarely at optimization problems and not really comparable to what IBM is doing.
For a straight spec reference, the field lines up like this:
| Company | Qubit Technology | Key System | Cloud Access | Commercial Status |
|---|---|---|---|---|
| IBM | Superconducting transmon | Heron r1 (133q), Condor (1,121q) | IBM Quantum Platform (since 2016) | $1B+ cumulative revenue |
| Superconducting | Willow (2024) | Google Cloud (limited) | Research-primary | |
| Microsoft | Topological (Majorana, pre-commercial) | Azure Quantum (third-party HW) | Azure Quantum | Pre-commercial on own HW |
| IonQ | Trapped ion | IonQ Forte | AWS, Azure, Google Cloud | Publicly traded (NYSE: IONQ) |
| Quantinuum | Trapped ion | H2 series | Quantinuum cloud | Honeywell spinout, private |
| Rigetti | Superconducting | Ankaa series | AWS Braket | Publicly traded, smaller scale |
| D-Wave | Quantum annealing | Advantage2 | D-Wave Leap | Publicly traded; optimization niche |
| Amazon Braket | Cloud aggregator | No own hardware | AWS | Hosts IonQ, Rigetti, QuEra |
IBM’s real edge in that field is not qubit quality, since Quantinuum’s trapped-ion systems beat it on fidelity right now, and it is not raw qubit count either. It is the pile-up of everything else: the largest deployed fleet at 75-plus systems worldwide as of Q4 2024, the longest continuous cloud-access history, the most mature software ecosystem in Qiskit, and the deepest enterprise network through the Q Network. Those are moats that take years to dig and cannot be copied on a quarter’s notice.
The quantum-adjacent win almost nobody covers may matter sooner than any of the hardware. In August 2024, NIST published its first finalized post-quantum cryptography standards, and two of the three primary standards, ML-KEM (based on CRYSTALS-Kyber) and ML-DSA (based on CRYSTALS-Dilithium), are algorithms developed by IBM researchers. These defend against a threat that needs no quantum computer to exist yet: the “harvest now, decrypt later” attack, where an adversary hoovers up encrypted data today and waits to crack it once a powerful enough quantum machine arrives. Financial records, health data, government communications, and long-lived infrastructure secrets are all sitting ducks. The NIST process ran eight years and chewed through 82 candidate algorithms. IBM taking two of the three primary slots is the payoff of sustained cryptographic research, and it lines up IBM’s security consulting and software businesses to bank near-term revenue as organizations migrate to quantum-resistant encryption. That revenue may well outpace what quantum hardware access earns for years.
The Q Network’s 100-plus members engage at wildly different depths, from active research programs running on IBM hardware to accounts kept open just to poke around. Vanguard and HSBC sit at the substantive end, with specific financial use cases and measurable, if not yet commercially decisive, results.
The most structurally important recent move is the National Quantum Algorithm Center, announced in 2025 with the State of Illinois and slated for the Illinois Quantum and Microelectronics Park in Chicago. Algorithm development is the bottleneck hardware progress cannot break on its own: even a finished fault-tolerant machine is dead weight without algorithms that exploit it on problems where quantum genuinely beats classical. This government-industry center reflects IBM’s bet that the software layer needs dedicated institutional infrastructure, not just more cloud endpoints. IBM Research Zurich anchors the European side, wired into academic institutions and national labs across the continent; at home, IBM works with Argonne, Oak Ridge, and other Department of Energy facilities under the U.S. National Quantum Initiative.
So what does the billion actually signal? IBM has built the most commercially mature quantum computing ecosystem on the planet, measured by deployed systems, software adoption, enterprise relationships, and booked revenue. That ecosystem makes real money, even if the accounting behind the money is fuzzy and the absolute number is small next to the rest of IBM.
The technology is still NISQ: noisy, error-prone, and short of demonstrating unambiguous quantum advantage on any problem someone actually needs solved. The road to fault tolerance runs to 2029 at the earliest, and every player faces the identical physics. IBM’s Starling roadmap is credible because IBM has hit its hardware targets for the better part of a decade, but credible and certain are not the same thing, and the people behind it know it. Jay Gambetta, IBM Fellow and VP of IBM Quantum, has architected the roadmap strategy. Dario Gil, SVP and Director of IBM Research, runs the pipeline that produced both the hardware line and the cryptographic standards. Arvind Krishna’s executive backing, unusual for a technology with a payoff horizon this long, has kept the money flowing through budget cycle after budget cycle.
The billion is a milestone in ecosystem building, and IBM has earned it. But the one thing worth doubting is the timing of the real prize. Quantum computing that provably beats a classical machine on a problem somebody needs solved is still out ahead of everyone, and I would bet IBM’s software and cryptography businesses pay real dividends long before a single Starling qubit does useful work.