Real machines exist. Real limits too. Here's what quantum computing can and can't do right now.
Short answer: yes, they exist. Longer answer: they exist, they run, and they still don't do most of what popular science headlines promise. Both things are true at once, and conflating them is where most of the confusion around quantum computing comes from.
Estimates put the number of operational quantum systems worldwide somewhere between 100 and 200, ranging from experimental prototypes locked away in national labs to commercial machines you can rent time on through a cloud dashboard. Some estimates go higher — up to roughly a thousand — once you count smaller research and educational units, NMR-based desktop machines included. The wide range itself tells you something: nobody has a clean, agreed-upon definition of what counts.
What "existing" actually means here
A classical computer processes information as bits — a 0 or a 1, nothing in between. A qubit can be a 0, a 1, or a superposition of both at once, and qubits can become entangled across distance in ways that have no classical equivalent. That's the theoretical engine behind the whole field. It's also, and I want to be direct about this, not magic — it's a different computational model that happens to be extraordinarily good at a narrow set of problems and mostly useless for the rest.
Hardware built on this principle is real. Google's Willow chip runs 105 qubits and demonstrated below-threshold error correction in late 2024 — a meaningful step, not a finished product. IBM's Condor processor passed 1,121 qubits. D-Wave's Advantage system, a quantum annealer rather than a general-purpose machine, ships with over 5,000 qubits and is already sold commercially, first as the Orion system back in 2007, then as a fully commercial product from 2022 onward. IonQ has offered commercial trapped-ion systems since 2019. Microsoft took a different bet entirely, unveiling its Majorana 1 chip in early 2024 — the first processor built on topological qubits, a design meant to be inherently more stable rather than requiring layers of error correction bolted on afterward.
The catch: most of this is still NISQ
NISQ. Noisy Intermediate-Scale Quantum. That acronym is doing a lot of quiet work in every honest article about the field. These machines are extremely sensitive — a stray vibration, a slight temperature drift, ambient electromagnetic noise, and the calculation degrades. Physicists call this decoherence: qubits losing their quantum state through interaction with the environment, and it's the single biggest obstacle standing between where we are and where the marketing usually implies we already are.
Which is why most quantum hardware needs cooling to near absolute zero. Not "very cold." Near absolute zero — colder than deep space, maintained by dilution refrigerators the size of a room, for a chip that might fit in your palm. The infrastructure cost dwarfs the chip cost, and that ratio isn't changing anytime soon.
The theorized endpoint — a universal quantum computer capable of general-purpose applications — is estimated to need well over 100,000 physical qubits with robust, scaled error correction layered on top. Nobody has built one. Not close, honestly. Current machines top out in the low thousands of physical qubits, and physical qubits aren't the same as logical, error-corrected qubits — you typically need dozens or hundreds of noisy physical qubits to synthesize one reliable logical one. Do the math on what 100,000 logical qubits implies in physical hardware, and the gap gets uncomfortable fast.
So what are they actually good for, today, not eventually
Optimization problems — routing, scheduling, logistics — where quantum annealers like D-Wave's already see commercial use. Chemistry and drug discovery, where simulating molecular interactions is a problem classical computers handle poorly by nature, since molecules are themselves quantum systems. Cryptography research, both building next-generation security and, less comfortably, probing how current encryption might eventually be broken. You can access several of these machines yourself right now through AWS Braket, Azure Quantum, or IBM's cloud platform, run a real experiment on real quantum hardware, and get a real (probably noisy) result back.
What they are not good for: replacing your laptop. Running your CRM. Doing anything a classical computer already does efficiently. This isn't a temporary marketing gap — it's structural. Quantum computers only outperform classical ones on a specific class of problems where quantum parallelism gives a genuine mathematical advantage. Outside that class, a qubit is just an expensive, fragile bit.
Here's the part worth sitting with for a second: Google's 2019 quantum supremacy demonstration — beating a classical supercomputer at a deliberately constructed random-circuit sampling task — proved the principle. It did not solve a problem anyone actually needed solved. That distinction between "we can prove this is faster" and "this is useful" gets collapsed constantly in casual reporting, and it shouldn't be.
Where this is headed — and where it honestly might not
McKinsey projects roughly 5,000 operational quantum computers by 2030, with hardware and software mature enough for the hardest problems not expected before 2035 at the earliest. I'd treat that second date the way I treat every long-horizon tech forecast: directionally useful, not something to plan a business around.
Whether decoherence gets solved through better qubit materials, through Microsoft's topological bet, or through some combination nobody's tried yet — that's genuinely an open question, and anyone who tells you they know the answer with confidence is selling something.
