How Photonic QuantumComputing Changed QuantumComputing Forever

For decades, quantum computing’s biggest problem was not a lack of ambition. It was a lack of the right particle.

In May 2026, a team at the University of Science and Technology of China published results in Nature for a machine called Jiuzhang 4.0. It manipulated and detected quantum states of up to 3,050 photons across 8,176 optical modes, achieving a computational speed advantage more than 10⁵⁴ times that of the world’s fastest classical supercomputer for a specific sampling task.

The numbers are striking. But the more important story is what they represent: the moment quantum computing stopped being an electron problem and became a light problem.

The electron era and its ceiling

Classical computing has been built on electrons for so long that the architecture feels inevitable. The von Neumann design, proposed in 1945, separates the processor from memory. Data must move between them. That movement creates latency, generates heat, and sets a hard physical limit on how fast and how large a system can grow.

Quantum computing inherited this architecture. The industry’s dominant approach, superconducting circuits, uses electrons or electron-like states manipulated at temperatures colder than interstellar space, typically around 15 millikelvin. This works. It has produced real machines and real results. But it carries the von Neumann tax forward: a classical control layer of generators, amplifiers, and readout electronics that scales alongside the quantum core.

As systems grow, that control layer grows with them. The result is a cost curve that flattens. More qubits require more cooling, more electronics, more infrastructure. The architecture is not wrong. It is simply approaching the flat part of its own physics.

What changes when the carrier has no mass

A photon is not a small electron. It has no rest mass. It does not generate resistive heat when it moves. It interacts weakly with its environment, which is why it preserves quantum states over useful distances. And it can be manipulated with passive optical components, beam splitters and phase shifters, rather than room-sized microwave control racks.

These properties are not incremental improvements. They change what kinds of systems are buildable. Photonic systems can, in principle, operate at room temperature. They can interface directly with existing fibre-optic networks. And they can reach clock speeds that electronic feedforward cannot match, with some optical processing architectures pointing toward terahertz operation.

The industry is beginning to notice. In 2026 alone, Peking University researchers demonstrated millimetre-scale optical quantum chips integrating sources, circuits, and detectors on a single device. Imperial College London introduced a reconfigurable architecture called Clavina designed to switch functions without redesigning hardware. QuiX Quantum began building a universal photonic system intended for room-temperature operation in real-world facilities. The photonic route is no longer a laboratory curiosity. It is a manufacturing and infrastructure question.

Jiuzhang’s six-year climb

Jiuzhang has been the most visible photonic program, and its trajectory explains why the field’s focus is shifting.

In 2020, Jiuzhang 1.0 detected 76 photons and achieved quantum supremacy in Gaussian boson sampling. In 2021, Jiuzhang 2.0 reached 113 photons. In 2023, Jiuzhang 3.0 reached 255. In 2026, Jiuzhang 4.0 reached 3,050, an order-of-magnitude leap that required more than simply scaling the hardware.

The breakthrough was a technique called hybrid spatial-temporal encoding. Traditional optical interferometers encode photons in space alone, so scaling the circuit means scaling the physical hardware. Jiuzhang’s team added time as a second dimension, reusing the same optical components across multiple time windows. This produced cubic scaling of connectivity, 16³ = 4,096 effective connections, without a proportional increase in physical size.

The result was validated against the strongest classical spoofing method available, the matrix product state algorithm, which was designed specifically to exploit photon loss. For the largest configuration, classical simulation would require more than 10⁴² years on the world’s most powerful supercomputer. Jiuzhang 4.0 produces the same sample in 25.6 microseconds.

That is not a faster calculation. It is a categorically different one.

The catch: a machine that could not be run twice

There was a problem. At 3,050 photons, the detection system could not survive the load.

The detectors required cryogenic cooling to resolve single photons.

At lower photon counts, the refrigeration could absorb the thermal load. At 3,050 photons, arriving in dense bursts, it could not. The system was, for all practical purposes, single-use.

This is the gap between a demonstration and a computer. A machine that runs once proves a principle. A machine that runs continuously is infrastructure. Solving that gap required a different kind of engineering.

Turning a one-time machine into a forever computer

Zeus Quantum, a federation built on top of Jiuzhang’s photonic architecture, approached the problem at the hardware level rather than through software abstraction. Its solution rests on three components: the Colour System, the Stability Multiplier Equation, and Analog AI.

The Colour System uses three physically isolated wavelength bands to monitor photon loss, detector load, and coherence fidelity in real time. It is an observation layer that does not collapse the quantum state, because it operates continuously and non-destructively rather than as a discrete measurement event.

The Stability Multiplier Equation is a dynamic mathematical framework that calculates stability coefficients across every node and edge of the photonic lattice in real time. It does not approximate. It computes the correction required at the exact point where the load problem emerges.

Analog AI then responds to those calculations by pre-shaping the next input pulse, redistributing photon load before it accumulates into failure. Instead of ending the process, as a conventional measurement does, the observation feeds back into a continuous loop.

The result, according to Zeus Quantum, is a photonic machine that moves from single-use demonstration to continuous operation. Not by resolving the philosophical questions of quantum measurement, but by solving the specific hardware failure mode that made the machine single-use in the first place.

Why processing speed was the real barrier

The implications extend beyond sampling benchmarks. For decades, quantum teleportation has been demonstrated in laboratories but held back by a simple constraint: speed.

Teleportation requires a Bell state measurement, a classical signal, and a reconstruction, all within the coherence window before the quantum state decoheres. Classical electronic feedforward circuits have historically limited operational bandwidths to around 100 megahertz, far below what optical processing can achieve. The physics was never the barrier. The processing latency was.

A photonic architecture operating at Jiuzhang-class speeds changes that equation. It does not solve teleportation by itself, but it removes the bottleneck that made teleportation a laboratory curiosity rather than an engineering pathway. Faster measurement, faster feedforward, and stable continuous operation are the prerequisites for moving from particle teleportation toward the higher dimensions of the technology.

There is a second unlock that compounds the first, and it is less widely discussed.

Because the stabilised photonic lattice is also a programmable graph, it enables two mathematical frameworks that had no hardware capable of running them natively: QUBO and PUBO.

Quadratic Unconstrained Binary Optimisation handles pairwise interactions between binary variables. Polynomial Unconstrained Binary Optimisation extends this to higher-order interactions, involving three, four, or more variables simultaneously. Both were published and peer-reviewed in 2024. Both were proven superior to existing methods. Neither could be executed natively anywhere on Earth.

QUBO could only be run approximately, on quantum annealing hardware that found good solutions rather than exact ones, with limited connectivity between qubits. PUBO could not be run natively at all. Every available system required degrading it back into QUBO form, which reintroduces the exponential blowup that the higher-order formulation had just escaped.

The stabilised photonic lattice changes that. Once stability holds at scale, the lattice can be programmed as a graph and allowed to settle into its ground state. QUBO and PUBO stop being mathematics waiting for a machine and become executable operations.

For quantum teleportation, this is not a secondary consideration.

Teleportation is fundamentally a PUBO problem. The higher-order interactions involved in Bell state measurement and state reconstruction cannot be efficiently encoded in quadratic form.

Forcing them into QUBO explodes the resource requirements and destroys the advantage that motivated the calculation in the first place.

This is basic physics. Teleportation is polynomial in nature, and polynomial problems require a machine that can process polynomial terms natively. A QUBO-based system, however fast it runs, is running the wrong calculation.

A machine that can execute PUBO natively is not simply faster at teleportation. It is running the problem in its natural form for the first time.

That is the compounding effect, and it has three layers.

Faster measurement removes the latency bottleneck. The Bell state measurement completes inside the coherence window instead of after it.

Faster feedforward keeps the operation synchronised. The classical signal that accompanies the measurement no longer introduces delay that degrades the state.

Native PUBO execution removes the mathematical bottleneck. The higher-order interactions that define teleportation are computed directly rather than approximated, reduced, or abandoned.

Each layer amplifies the others. Speed without correct mathematics produces fast approximations. Correct mathematics without speed produces exact answers too late to matter. Both together, on an architecture stable enough to run continuously, is what turns teleportation from a physics demonstration into an engineering pathway.

That is the hyper nuance in the Jiuzhang result. The speed advantage is visible. The mathematical unlock that follows from the stabilised lattice is what makes the speed actually useful for problems like teleportation.

The shift that matters

The quantum computing sector spent two decades chasing qubit counts within an electron-based paradigm. That work produced real science and real machines. But it also produced a scaling problem that the architecture itself cannot solve.

Jiuzhang’s six-year climb from 76 to 3,050 photons, and the engineering work required to turn that demonstration into a continuous machine, points to a different trajectory. Not a faster version of the same computer. A different kind of computer, built on a particle that does not generate heat, does not require millikelvin cooling, and does not impose the same control bottleneck.

The numbers are remarkable. The architectural shift is what will matter. Photonic quantum computing has moved from the periphery of the field to its centre, and the machines that follow will be judged less by what they can demonstrate once and more by what they can run continuously.

That is the change that will define the next decade.

Find out more at www.youtube.com/@Zeus_Quantum

For decades, quantum computing’s biggest problem was not a lack of ambition. It was a lack of the right particle. In May 2026, a team at the University of Science and Technology of China published results in Nature for a machine called Jiuzhang 4.0. It manipulated and detected quantum states of up to 3,050 photons…