Imagine trying to solve a puzzle so complex that even the world’s fastest supercomputer would need 2.5 billion years to finish it. Now imagine another computer doing the same task in just 200 seconds. This isn’t science fiction. It’s Jiuzhang, a groundbreaking quantum computer built in China — and it’s changing the future of computing.

Let’s explore what makes Jiuzhang so special — without needing a PhD in physics, but first, let’s understand what a quantum computer is and how it works.

Quantum computers

In a classical computer, information is carried by electrons flowing through wires, flipping tiny switches on or off to represent zeros and ones.

A “quantum computer” is a broad term for any device that uses the laws of quantum mechanics — in particular, superposition and entanglement — to process information in ways that classical machines cannot. Under the hood, all quantum computers must prepare, manipulate, and measure quantum bits (qubits), which unlike ordinary bits can exist in a superposition of 0 and 1 and can become entangled with one another., This means they can exist in a blend of both “on” and “off” at the same time — until we measure it. This extraordinary property is what gives quantum computers their power.

Within this general category, there are many different physical implementations: superconducting circuits (IBM, Google), trapped ions (IonQ, Honeywell), neutral atoms (QuEra), silicon spins (Intel, Oxford), and photons (Xanadu, USTC’s Jiuzhang). A photonic quantum computer is simply one whose qubits are carried by photons — individual particles of light — rather than by electric currents or atomic energy levels.

Photonic quantum computers

To use photons for computation, researchers start by generating individual photons on demand. Special laser setups and nonlinear crystals produce a steady stream of photons, each prepared in a known quantum state. These photons are then sent through an arrangement of mirrors, beam splitters, and phase shifters — optical components that manipulate their paths and phases. When a photon hits a beam splitter, for example, it doesn’t simply choose one path or the other; instead, it enters a superposition of traveling down both paths simultaneously. Phase shifters tweak the relative timing of these paths, altering the way different photon-paths interfere with one another.

This network of optical components is analogous to a circuit of logic gates in a classical computer. But instead of performing simple true/false operations, the photonic circuit enacts complex, reversible transformations on the quantum state of the photons. Every time two photons meet at a beam splitter, their probability amplitudes — that is, the mathematical weights describing their possible paths — interfere. The entire network of mirrors and splitters effectively computes a giant, multi-dimensional interference pattern that encodes the solution to a problem. Measuring the photons at the end of the circuit collapses their superpositions into definite outcomes. If you’ve designed the optical network correctly, the statistics of these measurement results correspond to the answer you seek.

Of course, there are practical challenges. Photons travel at the speed of light, so the control electronics must react almost instantaneously to measurement signals to reconfigure the circuit in time. Losses in fibers, imperfect mirrors, and detector inefficiencies can drop photons or introduce errors. To build large-scale machines capable of running deep, multi-step quantum algorithms, we’ll also need quantum error-correction schemes—ways of encoding each logical qubit into many physical photons so that the overall information survives mistakes.

Researchers are already tackling these hurdles by moving from bulky tabletop optics to integrated photonic chips. On these chips, beam splitters and phase shifters are tiny, on-chip waveguides and electrodes that can be tuned with voltages. Delay lines and on-chip loops can temporarily store photons, serving as short-term quantum memory. Superconducting nanowire detectors are becoming faster and more efficient, and advanced error-correcting codes that work with photons are under active development.

What Is Jiuzhang?

Jiuzhang is a photonic quantum computer that, in 2020, did something incredible: it solved a highly complex problem called Gaussian boson sampling. You can think of this as a very messy lottery draw where particles of light bounce through a maze and create a unique pattern at the end.

This type of task is so difficult for regular computers that it becomes practically impossible. Even the most powerful classical supercomputers would be stuck on it for billions of years. But Jiuzhang breezed through it in just a few minutes.

That moment is what scientists call quantum supremacy — when a quantum computer does something no normal computer can.

For Gaussian boson sampling, Jiuzhang’s specialty, the pattern of photon clicks directly reveals samples from a distribution that would be impossibly costly to calculate with a classical computer. But the same principle underlies any quantum algorithm: you prepare photons in the right initial states, guide them through a programmable network that encodes your computation, then measure them to read out the result.

Jiuzhang’s triumph was proving that we can already control and detect hundreds of photons in a single run without losing quantum coherence. It showed that large-scale photonic interferometers can be built with remarkable stability and precision. The next steps—programmability, adaptive control, quantum memory, and error correction—are all active areas of research. When those pieces come together, photons will have become not just the carriers of one special quantum task but the universal medium for solving the full spectrum of quantum-computable problems.

Jiuzhang 2.0 and 3.0: Bigger, Better, Faster

After the first success, the scientists didn’t stop. In 2021, they built Jiuzhang 2.0, which handled more photons and gave even better results. Then in 2023, they released Jiuzhang 3.0, which could track 255 photons at once — a huge leap from the original.

This version was 10 quadrillion times faster than the most powerful classical supercomputer in solving the same problem. That’s like sending a rocket to the moon in one second compared to waiting 300,000 years.

What’s the Catch?

Despite the incredible result, Jiuzhang can’t do everything. It’s amazing at one specific task (boson sampling), but it can’t yet run general-purpose software like your laptop or Google’s servers. Think of it like a Formula 1 car — unbelievably fast on a race track, but not great for grocery shopping. Still, the technology is a huge step toward making more flexible quantum computers in the future.

In principle, once you have reliable on-demand photon sources, fully programmable optical circuits, fast feed-forward control, and robust error correction, you have a universal photonic quantum computer. You could then translate any quantum algorithm — factoring large numbers, simulating complex molecules, optimizing massive logistical networks — into a sequence of optical operations. You’d simply map each algorithmic gate to a configuration of beam splitters and phase shifters, inject your stream of photons, let the light weave through the circuit, and measure the outputs.