A physicist at Oxford argues quantum computing faces a ceiling imposed by physics itself — not engineering. The field just received its highest honor. The test is already underway.
In March 2026, two things happened in quantum computing that belong in the same sentence. The ACM awarded its 2025 Turing Award — the Nobel Prize of computing — to Charles Bennett and Gilles Brassard for establishing the foundations of quantum information science. It was the first time in the award's sixty-year history that quantum physics was recognized. Days later, Atom Computing and Cisco announced a partnership to build a distributed quantum network exceeding one thousand qubits. And somewhere between the celebration and the construction, a physicist at Oxford published a paper arguing that one thousand qubits might be all there is.
The Paper
Tim Palmer's "Rational Quantum Mechanics" appeared in the Proceedings of the National Academy of Sciences this month. The argument is deceptively simple. Standard quantum mechanics assumes Hilbert space — the mathematical arena where quantum states live — is continuous. Palmer argues it is discrete. If the amplitudes and phases that define a quantum state must satisfy rational-number constraints rather than spanning the full continuum of real numbers, then the information capacity of Hilbert space is finite. And if it is finite, the number of qubits that can be meaningfully entangled has a ceiling.
Palmer puts that ceiling between two hundred and four hundred qubits for current technologies, with an absolute maximum around one thousand under any physical implementation. Above that threshold, there is not enough information capacity to allocate even one bit per dimension of Hilbert space. The quantum state cannot be specified with enough precision for the computation to proceed.
The implication is severe. If Palmer is right, quantum computers will never factor a 2,048-bit RSA integer — not because of engineering limitations, but because physics does not provide enough room.
Where We Are
The scaling landscape shows an industry approaching the threshold from below — fast enough that the ceiling, if it exists, should become visible within years.
Quantinuum's Helios processor recently achieved ninety-four error-protected logical qubits from ninety-eight physical qubits — a near one-to-one encoding ratio that represents beyond-break-even quantum error correction. Logical gate error rates reached roughly one error in ten thousand operations, ten to one hundred times better than the physical qubits underneath. At ten percent of Palmer's predicted ceiling, Helios is the canary.
Caltech assembled 6,100 physical neutral-atom qubits in an optical tweezer array published in Nature — the largest qubit array ever constructed. They maintained superposition for thirteen seconds with 99.98 percent single-qubit gate accuracy. But these qubits have not yet been entangled at scale. Physical qubits and logical qubits are different currencies, and the exchange rate is steep.
IBM plans to ship Kookaburra this year — a 1,386-qubit multi-chip processor connecting three dies through chip-to-chip couplers, scaling to a combined 4,158-qubit system. If Palmer's ceiling exists, Kookaburra will be the first architecture to press directly against it.
And Atom Computing's AC1000 platform already holds 1,180 qubits in a 1,225-site atomic array — the first universal gate-based system to cross one thousand. Their new partnership with Cisco, announced March 25, aims to link multiple such processors through quantum networking. The explicit goal: bypass the limits of a single system by distributing entanglement across a network.
The Contradiction
Google disagrees with Palmer — dramatically. Earlier this month, Google accelerated its estimate of Q-Day — the date when a quantum computer can break current encryption — from roughly 2035 to 2029. The revision was grounded in research by Craig Gidney showing a 2,048-bit RSA key could be broken in under a week by a quantum computer with one million noisy qubits. Google is already migrating to post-quantum cryptography on that timeline, planning deployment in Android by next year.
Palmer says the computation is physically impossible. Google says it will happen within three years. This is not a difference of degree. It is a difference about what the universe permits.
The U.S. federal government's own deadline for post-quantum migration is 2035. The NSA set 2031. Google just leapfrogged both. Either Google's engineers know something the government doesn't, or Palmer knows something Google doesn't.
The Experiment
The test is already underway, and the timeline is shorter than most scientific disputes allow. Three independent architectures are converging on the same boundary from different directions.
IBM's Kookaburra, shipping this year, will place over a thousand entangled qubits in a single coherent system — directly in the range where Palmer's ceiling should start to bind. If error correction gains plateau despite scaling to more physical qubits, that is evidence for the ceiling.
Atom Computing's distributed approach with Cisco offers a different probe. If linking multiple thousand-qubit processors through quantum networking yields no computational advantage beyond what a single sub-thousand-qubit system achieves, the ceiling holds even when the engineering route changes.
Quantinuum's trajectory from fifty-six to ninety-four logical qubits provides the third track. If the path from ninety-four to two hundred shows unexpected resistance — error rates that stop improving despite better hardware — the ceiling becomes visible at a smaller scale.
Within two to four years, these architectures will independently probe the same boundary. If all three plateau in the same range, the coincidence becomes difficult to dismiss as engineering.
What the Continuum Hides
Palmer's paper is not really about quantum computing. It is about what kind of universe we live in.
Standard quantum mechanics treats Hilbert space as continuous — an infinite-dimensional arena of real-valued amplitudes. This assumption has worked for a century. Every prediction confirmed, every experiment consistent. Palmer argues the assumption was never tested at scale, because no experiment before large-scale quantum computing could probe the difference between a continuous and a discrete Hilbert space. The two descriptions diverge only when many qubits are entangled simultaneously.
Wheeler's phrase — "it from bit" — proposed in 1990 that information is the fundamental substrate of physics. It has been a philosophical position for thirty-six years. Palmer's contribution is to make it empirical. If Hilbert space is discrete, the continuum was always the approximation, not the other way around. A century of physics was rounding off.
The Turing Award to Bennett and Brassard recognized the theoretical foundations built on the continuum assumption. BB84, quantum teleportation, quantum key distribution — these assume an infinite Hilbert space. If Palmer is right, the foundations are subtly different from what Bennett and Brassard formalized. The field's greatest honor arrives at the moment its core mathematical assumption faces its first real test.
The Prediction
The falsifiable claim is specific: no quantum computer will factor a 2,048-bit RSA integer within the next five years. Not because the engineering is hard — though it is — but because the physics does not permit it.
Google bets the opposite. Q-Day by 2029. One million noisy qubits, one week of computation, encryption broken. The infrastructure migration is already underway.
This journal documented quantum computing's emerging role as a co-processor — not a replacement for classical computing, but a partner constrained to specific problem types. The Hard Limit asks how far the partnership extends. If Palmer is right, the quantum side of the equation has a ceiling written into the structure of spacetime itself. The co-processor is powerful but bounded, and the boundary is not moving.
If he is wrong, the race to one million qubits continues, Q-Day arrives on schedule, and the $650 billion quantum investment was a bet on the right physics all along.
Either way, we are about to find out. The experiments are running. The architectures are scaling. And they are converging on the same thousand-qubit boundary — from three different directions, with three different technologies, each carrying its own assumptions about what the universe allows.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)