Diraq, a quantum computing company in Sydney founded by the physicist and engineer Andrew Dzurak, is building toward a single chip carrying millions of qubits — and betting that almost nobody who uses one will ever be in the same building as the hardware. Dzurak's model puts quantum processors in data centers beside large AI factories, reached remotely the way cloud AI services are reached today, which he argues lowers cost and energy use and takes the technology out of the exclusive hands of governments and well-funded organizations. He says meaningful commercial returns are still several years away. Asked in January 2025 when useful quantum computers would arrive, Nvidia chief executive Jensen Huang said 15 years would probably be too optimistic, and that 20 was the figure most people would call realistic.
The distance between "several years" and "20 years" is the most informative thing in the field right now, and the market has already reacted to it once. Quantum stocks fell roughly 40% the day Huang gave that answer, CNBC reported. Two months later he softened the remark. A sector that can lose 40% of its value on one executive's offhand estimate of a date is a sector priced on belief about timelines rather than on delivered capability — and the executive in question sells the classical hardware that quantum would eventually be competing with for the same data center floor.
The remote-access model is less a strategic choice than a consequence of thermodynamics. Quantum hardware occupies a great deal of space and functions only at extreme cold. MIT's account is that heat introduces errors in qubits, so the systems are housed in refrigeration units held just above absolute zero, at −459 degrees Fahrenheit. No home thermostat reaches that. Presenting the constraint as an access strategy is legitimate — mainframes and then GPUs both became broadly usable at the moment someone else agreed to host them — but the honest framing is that "quantum for everyone" is a distribution promise, not a physics one.
The physics itself is further along and duller than the public impression of it. Most people have either never heard of quantum computing or file it under science fiction and the distant future. In practice qubits have been demonstrated to work: a classical bit is always a zero or a one, while a qubit before measurement can hold a superposition of both states at once, as the Caltech Science Exchange explains it. IBM's position is that large quantum machines will solve certain hard problems many times faster than today's classical computers, turning calculations that would take conventional machines millennia into minutes or hours. What remains unsolved is running enough qubits, reliably and cheaply enough, that the technology is not restricted to governments and rich institutions.
Dzurak calls the destination "utility scale": the point at which quantum computers generate more value than they cost to own and operate. That definition rewards a second reading. It can be satisfied by making the machines far more capable, or by making them far cheaper to run, and only one of those is the breakthrough the sector has been funded to deliver. Co-locating with AI data centers, which is Diraq's stated route to lower cost and lower energy use, works on the second variable.
The case for pairing quantum with AI rests on something that has already happened. Over the past few years AI agents acting on people's behalf have spread into research universities and Fortune 500 companies, and those agents can be placed inside the same computing system as quantum processors. Microsoft is working along these lines and calls the arrangement "scientist in the loop": the agents make recommendations, the researchers make the decisions. The Quantum Insider describes teams of agents analyzing large bodies of information, proposing hypotheses, optimizing experiments and testing theories under human supervision.
The most concrete result cited for the combination does not involve agents at all. Researchers at the University of Toronto and Insilico Medicine, in work published in the journal Biotechnology, combined quantum computing, generative AI and classical computational methods to design molecules targeting KRAS, a protein linked to cancer and previously considered undruggable. Igor Stagljar, a co-investigator and professor of biochemistry and molecular genetics at the Donnelly Centre in the University of Toronto's Temerty Faculty of Medicine, believes methods of this kind can cut several years off the preclinical stage of drug development.
That result is the strongest evidence on offer, which is why what it leaves out matters. Three methods were used together, and nothing in the account separates their contributions. Generative chemistry has been producing candidate molecules against hard targets for years on entirely classical hardware, so the question that decides whether this is a quantum milestone or an AI milestone with a quantum component is precisely the one the write-up does not answer. Stagljar's estimate is also a forecast about a pipeline rather than a measurement of one: molecules designed this way still have to survive everything that comes after design, which is where candidates against undruggable targets have historically died.
Dzurak's own position is that the most interesting applications have not been thought of yet, since every large technological shift begins as an idea in someone's head. The applications people have thought of lean academic. Google's research division points to gravity as an example of a problem quantum processors might open up, alongside time crystals, quantum chaos, chemistry and complex physical theories — work reaching toward the unified account of the universe that eluded Einstein and the generations of physicists after him. Valuable, and unlikely to be what pays for the data centers.
Which leaves encryption, the one application nobody has to imagine. The concern is that these machines will break the cryptography protecting bank accounts and nuclear codes. Dzurak's answer is that quantum computers will indeed attack certain classes of ciphers, and that codes resistant to quantum attack already exist, built specifically so quantum machines cannot break them. Both statements are true, and neither addresses the timing problem security experts keep raising: harvest now, decrypt later. Encrypted traffic and archives are being accumulated today by people who intend to open them once the hardware catches up.
So the field is running two clocks that are not synchronized. The commercial clock is the one Dzurak is managing — several years to returns, machines in data centers, cost per useful result, an ever-widening pool of people and companies with computing power at hand. The adversarial clock started earlier and is indifferent to all of it, because reading stored data does not require utility scale, or affordability, or availability to a researcher with a laptop. It requires one machine that works, once. If Huang's twenty years is nearer the mark than Dzurak's several, that is a reprieve for the drug pipelines and the opposite for anyone whose encrypted data is being copied this year, because it means the collection window is two decades wide and currently open.