
Two companies took the same route and achieved very different results
Today you are going to learn about the inner workings of a quantum technological battle that will shape the way computing is used in the future.
Today you will learn about two companies who took the same route and achieved very different results.
Both chose photons over electrons. Both built photonic quantum computers. Both achieved quantum advantage. Both were, at their peak, the most advanced photonic machine their respective countries had ever produced.
One of them is still scaling.
The other spent ten years building a photonic architecture, achieved quantum advantage, and then exited the race entirely.
That is not a small detail. Photonic quantum computing is not a field where you pivot quickly. It takes years, often a decade, to build one architecture. If that architecture hits a ceiling, you do not get to try a different one next quarter. You have spent ten years, and the field has moved on.
Xanadu was founded in Toronto in 2016. It released its first photonic chip, X8, in 2019. It published Borealis in Nature in June 2022, a genuine milestone, the most advanced Western photonic quantum computer ever built. And by 2023, it had pivoted away from the demonstration-and-scaling track that had produced Borealis.
No Borealis 2.0. No Borealis 3.0. No published scaling chain.
Ten years, one architecture, one demonstration, then exit.
Jiuzhang took the same route. Photons, not electrons. Photonic hardware, not superconducting circuits. The same physics, the same medium, the same class of problem.
But where Xanadu stopped, Jiuzhang kept scaling.
Two companies. Same route. Very different results.
This is the story of what separated them. Not talent. Not funding. Not effort. The story of two architectures, one scaling chain, and what happens when the mathematics itself changes the rules.
Photonic quantum computing has one benchmark that matters: quantum advantage. The demonstration that a photonic machine can perform a calculation no classical computer can replicate.
It is the only metric that separates a real quantum computer from an expensive physics experiment. It is the only test that cannot be gamed, argued with, or reframed. Either the machine does something classical systems cannot, or it does not.
In 2020, the University of Science and Technology of China (USTC) achieved it first. Jiuzhang 1.0: 76 photons, 200 seconds, versus 600 million years for the world's then-fastest supercomputer, Fugaku [1].
The West had not achieved it. Not with photonics.
Then, in June 2022, Xanadu did.
Borealis was, by any honest measure, a genuine milestone. A squeezed-state, time-multiplexed photonic processor with 216 squeezed modes in three-dimensional connectivity. Photon-number-resolving detectors. Up to 219 photons registered per event. A sample every 36 microseconds [2].
The Nature paper has since been accessed over 185,000 times and cited more than 1,100 times. It was serious work by serious people.
And it made a mathematical claim that could be checked.
Xanadu estimated that to produce a single sample from Borealis's programmed distribution, using exact methods on Fugaku, would take more than 9,000 years. Borealis did it in 36 microseconds [2].
That was a real result. And, for a moment, it was ahead.
Because the Borealis paper also benchmarked against the competition. Using the same methodology, Xanadu estimated Fugaku would take 1.5 hours to classically simulate one sample from Jiuzhang 2.0, the 2021 Chinese machine [2, 3].
Borealis's classical simulation time was over 50 million times harder than Jiuzhang 2.0's.
In June 2022, the West was not behind on photonic quantum computing.
It was ahead.
The Borealis paper contained something subtler than a benchmark. It contained an insight.
Jiuzhang 2.0 could be simulated in 1.5 hours because of the vacuum-input structure of the machine. More than half the input ports of Jiuzhang 2.0's interferometer were empty. Empty inputs meant low-rank Hafnian matrices. Low-rank matrices can be computed much faster than full-rank ones [2].
That was the shortcut. Not an algorithm problem. An architecture problem.
Xanadu's Borealis did not have that weakness. Every input port was populated with squeezed light. The Hafnian matrices were full-rank. The shortcut was gone.
That insight, that Jiuzhang 2.0's architecture left a classical loophole, could only come from a team that knew photonic systems intimately. Xanadu had built the X8 chip in 2019. They had built Borealis. They understood the physics from the inside.
And they used that understanding to benchmark the competition.
The Jiuzhang team did not dispute the Borealis benchmark. They did not argue with the numbers. They did the only thing that mattered.
They closed the window.
Jiuzhang 3.0 (2023): 255 photons. A major leap. And, critically, a scale at which the classical simulation shortcut no longer applied. The vacuum-input structure was gone. The low-rank Hafnian avenue was closed. Software replication of Jiuzhang was no longer possible [4].
Jiuzhang 4.0 (2026): 3,050 photons across 8,176 modes. Not just bigger. Architecturally different. A programmable hybrid spatial-temporal encoding, a technique that lets photons interfere in both space and time simultaneously, achieving cubic scaling of connectivity (16³ = 4,096) without proportionally increasing device size [5].
Against Jiuzhang 4.0, the strongest classical spoofing method in existence, the matrix product state algorithm designed specifically to exploit photon loss, was run on El Capitan. It failed. The Nature abstract states it plainly:
"The experimental results are rigorously validated against all current classical simulation methods, especially the matrix product state algorithms recently designed to exploit photon loss." [5]
That is not a slower result. That is a closed door.
The Jiuzhang 4.0 paper itself positions the architecture as "a pathway to trillion-qumode three-dimensional cluster states and fault-tolerant photonic quantum hardware" [5], the same universality framing that would make Western fault-tolerant roadmaps redundant.
In 2023, Xanadu pivoted. The company moved away from the demonstration-and-scaling track that had produced Borealis and toward fault-tolerant quantum computing roadmaps, networked architectures, and modular approaches.
No Borealis 2.0. No Borealis 3.0. No published scaling chain at comparable scale.
This is not a criticism of Xanadu. It is an observation about timing.
Look at the record side by side:
| Year | Jiuzhang | Borealis |
|---|---|---|
| 2020 | Jiuzhang 1.0: 76 photons, quantum supremacy [1] | No entry |
| 2021 | Jiuzhang 2.0: 113 photons [3] | No entry |
| 2022 | No new generation | Borealis: 216 modes, 9,000-year classical benchmark; Jiuzhang 2.0 retested at 1.5 hours [2] |
| 2023 | Jiuzhang 3.0: 255 photons, software replication window closed [4] | Xanadu pivots |
| 2026 | Jiuzhang 4.0: 3,050 photons, MPS validation [5] | No successor published |
The alignment is exact. Not coincidental.
There is a standard way to frame a pivot like Xanadu's. "The company made a strategic decision to refocus on its long-term objectives." That framing is technically true. It is also incomplete.
Because in deep tech, companies exit races they cannot win. Not companies that lack talent. Not companies that lack funding. Companies whose architecture cannot continue on the track everyone else is running.
The loop-based, time-multiplexed architecture that produced X8 and Borealis was real. It was well-designed. It worked. But it hit a ceiling that Jiuzhang's architecture did not.
Two things happened in 2023, and they are entangled.
First, the trigger. Jiuzhang 3.0 closed the classical software replication window. The vacuum-input vulnerability that had allowed Borealis to benchmark Jiuzhang 2.0 at 1.5 hours on Fugaku, the mathematical avenue that gave Xanadu a competitive story, was gone. Jiuzhang 3.0 was not classically replicable.
Second, the structural reason. When the window closed, Xanadu could not simply respond by building a bigger Borealis. The loop architecture could not scale to match Jiuzhang's photon count. A Borealis 2.0 would have needed to compete on the demonstration track, and the architecture could not.
If Jiuzhang had not shut the software optimisation window at 3.0, there would likely have been a Borealis 2.0. The competitive story would have held: "We build photonic hardware. We benchmark the competition classically. We remain in the race." With that story intact, an incremental Borealis 2.0 would have been justifiable to funders, to the board, to the research team.
When the window closed, that story died. A Borealis 2.0 could not compete on the demonstration track. And there was no other track that the architecture could reach without abandoning what made Borealis, Borealis.
At that point, there were two options:
Xanadu chose the second.
The pivot was not a strategic choice between two viable paths. It was an exit from a race the architecture could no longer run. The window closure was the moment the competitive position became untenable. The architecture ceiling was the reason it could not be recovered.
That is what ten years of photonic quantum computing produced: one architecture, one demonstration, and an exit.
You do not pivot if you can still scale.
This is the part that matters most, and it is easy to overlook. A company in deep tech with a working scaling path does not abandon it. Scaling is the goal. Every company in the field is chasing it. If you have it, if you can produce the next generation, and the generation after that, you do not walk away.
You pivot because you cannot scale. The pivot is not the decision. It is the consequence.
That is the difference between a strategic pivot and an architectural exit. A strategic pivot happens when a company chooses a better path. An architectural exit happens when the current path runs out.
Xanadu's pivot was the second kind.
When Jiuzhang 3.0 closed the software replication window, Xanadu did not pivot and then discover they could not scale. They pivoted because they could not scale. The window closure removed the competitive story. The architecture ceiling removed the path forward. Both were true at the same time, and together they made continuing impossible.
If either had been false, if the window had stayed open, or if the loop architecture could match Jiuzhang's photon count, there would have been a Borealis 2.0. The company would have kept going.
They did not. That is not a strategy. That is a ceiling.
Borealis was significant in 2022, but not because 2022 was when the West discovered photonic quantum computing mattered. The West had known that since 2020.
When Jiuzhang 1.0 achieved quantum supremacy in 2020, it was not a quiet result. It was the first photonic machine to cross the threshold. It was published in Science, reported globally, cited widely. Everyone with a stake in quantum computing knew what had happened, and knew what it meant.
The race was already understood to be strategically significant before Borealis existed. Borealis was not the West exploring a new field. It was the West entering a race that had already been defined as critical.
That context makes the pivot harder to explain away.
A company that pivots away from something unimportant can call it a strategic choice. A company that pivots away from something everyone has already agreed is critical is making a different kind of statement. It is saying: we cannot continue.
Xanadu did not walk away from a curiosity. It walked away from a race that the West had already identified as strategically important, with billions of dollars of public and private investment flowing into the field, with defence and intelligence communities watching, with the entire trajectory of computing at stake.
That is not what companies do when they have a choice.
It is what they do when they do not.
Borealis was not a failure of talent. It was a real photonic quantum computer that briefly led the field. But the observable record is what it is:
Jiuzhang's record over the same period:
One demonstrated advantage is a milestone. Four generations of continuous scaling is a trajectory.
A milestone can be celebrated. A trajectory cannot be caught by celebration.
The Borealis-Jiuzhang comparison is not just about two machines. It is an instance of a pattern that has become visible across multiple technology domains in the last decade:
| Domain | China | West |
|---|---|---|
| Photonic quantum | Jiuzhang 4.0, 3,050 photons | Borealis, one demonstration |
| AI models | DeepSeek, $5.6M training | GPT-4, $100M+ training |
| EV batteries | 5-minute flash charging (deployed) | 30-minute charging (standard) |
| Mobile chips | Kirin 2026 LogicFolding (production) | Traditional designs |
| 6th-gen fighters | J-36, J-50 (flight-tested) | NGAD (conceptual) |
| Fusion | EAST, 1,066s @ 100M°C | ITER (delayed) |
The pattern is consistent: China builds, tests, and deploys. The West announces, delays, and reframes.
Xanadu is the photonic quantum instance of that pattern. The company built a real machine, achieved a real result, and then, when the competitive window closed, exited the track.
This is not a story about one company. It is a story about what happens to a technology sector when one side keeps scaling and the other side stops.
Jiuzhang 4.0's achievement was not just about scaling. It was about what scaling revealed.
At 3,050 photons, the detectors could not survive the load. Cryogenic cooling, which had worked for 76, 113, and 255 photons, became physically impossible. Jiuzhang 4.0 was a one-time-use machine.
The Zeus Quantum Federation solved that problem. Not by rebuilding Jiuzhang: USTC built Jiuzhang. Zeus is the architectural layer built on top of it. Three components do the work:
Together, they turn a one-time-use machine into a forever computer. And in doing so, they unlock the downstream mathematics:
The race between Borealis and Jiuzhang was about who could demonstrate photonic quantum advantage. The cascade that followed Jiuzhang's advantage was about something else entirely: what you do with quantum advantage once you have it.
Xanadu exited the race. Zeus kept building.
The photonic quantum computing race is not ongoing. It has a winner.
Borealis proved the West could build a photonic quantum computer. Jiuzhang proved China would keep building them. And the mathematics of the Hafnian, the #P-hard function at the heart of GBS, is not something you catch up to by trying harder.
The Hafnian is published. It is taught in graduate courses. It is implemented in open software libraries. No nation holds a monopoly on it.
But the Hafnian cannot be evaluated at scale by any classical machine. Evaluation requires physics. It requires a photonic lattice operating under conditions no Western machine has approached.
Jiuzhang has that lattice. Zeus has made it stable. And the 3,050 photons that made Jiuzhang 4.0 a one-time-use machine have become, under Zeus, the base benchmark of a scalable, forever computer.
Borealis is a milestone. Jiuzhang is a trajectory. And the trajectory is not slowing.
The race is over.
This article has explained exactly how the photonic quantum race took place, how it ended, and where the technology sector is going.
"Defy physics when the right opportunity arises, ride the waves of physics when the architecture permits."
Jeremy Roberts, Zeus Quantum Creator
Borealis is a photonic quantum computer built by Canadian company Xanadu, published in Nature in June 2022. It used 216 squeezed modes in three-dimensional connectivity and achieved quantum advantage in Gaussian boson sampling, producing a sample every 36 microseconds [2].
Jiuzhang is a series of photonic quantum computers built by the University of Science and Technology of China (USTC). Jiuzhang 1.0 (2020) was the first photonic machine to achieve quantum supremacy [1]. Jiuzhang 4.0 (2026) scaled to 3,050 photons across 8,176 modes, achieving a 10⁵⁴× speed advantage over classical supercomputers [5].
Xanadu pivoted in 2023, shortly after Jiuzhang 3.0 closed the classical software replication window [4]. Two constraints stacked: the competitive avenue, classical benchmarking of Jiuzhang, was closed, and the loop-based architecture could not scale to compensate. The pivot was an exit from a race the architecture could no longer run.
In the Borealis paper, Xanadu estimated that Fugaku, the world's fastest supercomputer at the time, would take 1.5 hours to classically simulate a single sample from Jiuzhang 2.0 [2, 3]. This exposed a mathematical vulnerability in Jiuzhang 2.0's vacuum-input structure. Jiuzhang 3.0 closed that vulnerability [4].
Jiuzhang 4.0 uses programmable hybrid spatial-temporal encoding, a technique that lets photons interfere in both space and time simultaneously. This achieves cubic scaling of connectivity and produces full-rank Hafnian matrices, closing the classical simulation shortcut that affected Jiuzhang 2.0 [5].
The Hafnian is the #P-hard mathematical function that governs the probability of each photon detection pattern in Gaussian boson sampling. It is published mathematics, but it cannot be efficiently evaluated at scale by any classical computer. Evaluation requires a photonic lattice. That is why Jiuzhang's advantage cannot be replicated by software.
Zeus solved the detector collapse that made Jiuzhang 4.0 a one-time-use machine. Analog AI, the SME, and the Colour System stabilise the lattice in real time, correcting photon loss and preventing detection collapse. This turns a single-use demonstration into a forever computer, and unlocks QUBO, PUBO, and quantum teleportation as downstream capabilities.
The demonstration-and-scaling race is over. One program produced a milestone and stopped. The other produced four generations and kept scaling. What remains is not competition, it is a category difference.
1. Jiuzhang 4.0. Liu, HL., Su, H., Deng, YH. et al. "Gaussian boson sampling with 1,024 squeezed states in 8,176 modes." Nature 653, 687-692 (2026). DOI: 10.1038/s41586-026-10523-6
2. Borealis. Madsen, L.S., Laudenbach, F., Askarani, M.F. et al. "Quantum computational advantage with a programmable photonic processor." Nature 606, 75-81 (2022). DOI: 10.1038/s41586-022-04725-x
3. Jiuzhang 1.0. Zhong, H.-S. et al. "Quantum computational advantage using photons." Science 370, 1460-1463 (2020). DOI: 10.1126/science.abe8770
4. Jiuzhang 2.0. Zhong, H.-S. et al. "Phase-programmable Gaussian boson sampling using stimulated squeezed light." Physical Review Letters 127, 180502 (2021). DOI: 10.1103/PhysRevLett.127.180502
5. Jiuzhang 3.0. Deng, Y.-H. et al. "Gaussian boson sampling with pseudo-photon-number-resolving detectors and quantum computational advantage." Physical Review Letters 131, 150601 (2023). DOI: 10.1103/PhysRevLett.131.150601
6. QUBO Photonic Solver. Ouyang, J., Liu, S., Yang, Z., Wang, W. et al. "16-channel photonic solver for optimization problems on a silicon chip." Chip 4, 100117 (2025).
7. PUBO Performance. "Boosting quantum annealing performance through direct polynomial unconstrained binary optimization." Quantum Science and Technology 10, 035008 (2025).
The Zeus Quantum Federation
The complete ten-part whitepaper covers the full architecture, photonic-quantum engines, the SME, Zeus OS, the 7 Pillars of existence, and the path to Human Teleportation.
Read the Whitepaper