China Built A Godlike Computer That Shouldn't Work — Unless The Multiverse Is Real

In March 2025, a machine in Hefei, China, spent a few hundred seconds on a problem that the fastest supercomputer on earth would need 6.4 billion years to finish. That's not a typo, it's on the cover of Physical Review Letters. But the man who founded Google's rival quantum program says results like these lend weight to one explanation, that the computation is spread across parallel universes. So what is this Chinese machine actually doing in those few hundred seconds? That question has been open since 1997, and it just stopped being philosophy. The machine is called Zuchongzhi 3.0, a superconducting quantum processor at the University of Science and Technology of China, 105 qubits connected by 182 couplers, every operation running above 99% accuracy. A qubit is the part you need to hold in your head for the next 10 minutes. A normal bit is a zero or a one. One or the other. Always. A qubit can sit in a blend of both at once. And every qubit you add doubles the number of possibilities the machine holds in play. By 83 qubits, you're at roughly 10 trillion trillion possibilities at once. The test Zuchongzhi beat the world with is called random circuit sampling. You push the qubits through a long random sequence of operations, a deliberate quantum scramble, then you ask the machine for samples of what falls out. Picture a set of dice that are all invisibly linked, so the rolls come out carrying a hidden pattern. The machine just rolls, and the greatest check, whether its rolls carry the statistical fingerprint that this exact circuit should produce. Anything classical that wants to imitate those rolls, has to track every possibility in the scramble at once. And with 83 qubits pushed through 32 rounds, that tracking swallows supercomputers whole. Frontier, the American supercomputer, would need billions of years to imitate what Zuchongzhi delivered in minutes. One million samples, and a million times faster than the best result Google had published only a few months earlier. The man behind it is Pan Jianwei, and his career is its own signal. He earned his PhD in Vienna under Anton Zeilinger, the physicist who later won the Nobel Prize for entanglement, then went home in 2001 to build a quantum lab in Hefei. In 2016 his team launched Micius, the world's first quantum satellite, and a year later Pan called his old professor over a quantum encrypted video line from Beijing to Vienna. A connection built so that any eavesdropper on the line would give themselves away. In China they call him the father of quantum. In 2024 London's Royal Society elected him a fellow, in the same stretch that Washington was tightening export controls on China's quantum industry. For Zu Chongzhi, his group picked random circuit sampling as the proving ground, and that choice looks like a technicality right now. Hold on to it, because it decides the end of this story. Oxford, 1997. A physicist named David Deutsch publishes a book called The Fabric of Reality, and buried in it is a challenge that reads like a riddle. Shor's algorithm, the quantum method for cracking the encryption that protects banks and governments Deutsch worked out that running it on a serious number would use something like 10 to the power 500 times more computational resources than the universe visibly contains. For scale, the entire visible universe holds about 10 to the power 80 atoms. The algorithm needs more working space than there are atoms to build it from. And yet the math says it finishes. And Deutsch is no outsider shouting at the field. He wrote the founding paper of quantum computing itself back in 1985. So he asked the question nobody has managed to bury. When the algorithm factors that number, where was the number factorized? His answer was blunt. The work gets done in parallel universes. And a working quantum computer is the multiverse showing its hand. At the time you could file all of it under thought experiments, because no machine on Earth came close to making it bite. The machine started biting in 2019. That autumn, Google announced its Sycamore processor had done a sampling task in 200 seconds that would take a supercomputer 10,000 years, and the phrase quantum supremacy went around the world. It held up for less than three years. Then three physicists in Beijing, Feng Pan, Ke Yang Chen and Pan Zhang, came forward with a classical method that pulled a million samples out of Sycamore's supposedly impossible circuit in about 15 hours on 512 ordinary GPUs and they estimated that a top supercomputer could do it in tens of seconds so the first quantum supremacy claim in history was taken apart by Chinese physicists with a room full of graphics cards Google's 10,000 year wall had come down to an afternoon it stung but it was also the best quality control the field ever Google. Because from that day on, nobody could publish a supremacy claim without asking what the physicists in Beijing would do to it. That defeat rewrote the rules for everyone. From then on, a quantum machine had two opponents. The supercomputers in front of it, and every mathematician on earth, hunting for a shortcut through its circuit. The next machines were built with that scar tissue. In December 2024, Google came back with a chip called Willow, 105 qubits. And a sampling result it said would take the world's fastest supercomputer 10 septillion years. That's a one with 25 zeros, and the universe is around 14 billion years old, so headlines did what headlines do. But the announcement carried a sentence that physicists are still arguing about. Hartmut Neven, the founder of Google Quantum AI, wrote that this mind-boggling number exceeds known time scales in physics, and that it lends credence to the notion that quantum computation occurs in many parallel universes, in line with the idea that we live in a multiverse, a prediction first made by David Deutsch. There it was, Deutsch's riddle, dormant for 27 years, printed as the official commentary on a Google product, and news outlets ran it straight. Google says its chip may prove parallel universes exist. The pushback came fast, and it came from people who love quantum computing. The astrophysicist Ethan Siegel wrote that Neven had conflated a Hilbert space, the mathematical arena where quantum states live, with actual parallel universes. Scott Aaronson, one of the field's most respected theorists, put it drier. The new experiment doesn't add anything new to this old debate, he wrote. It's yet another confirmation of the predictions of quantum mechanics. And on the physics, the critics have a point, that every interpretation of quantum mechanics predicts exactly the same machine behavior, so no benchmark can crown one of them. But go back and read what Deutsch demanded. Not a prediction. He wrote that solving the equations merely predicts the algorithm will work. What he demanded was an explanation of where all that work happens. And to this day that demand is still standing there open. It's also about to get company. Because three months after Neven's sentence, China answered with hardware. That's the March 2025 result you started with, and its real message sits in the fine print. The paper places Zuchongzhi 3.0's circuit six orders of magnitude beyond Google's experiments and classical simulation cast. In plain terms, the scramble runs so deep that the published shortcuts, including the one that killed the 2019 claim, don't come anywhere near it. The same country that collapsed the first supremacy claim has now built one that holds. Where this race lands is the rest of this story. and it runs straight through your own encrypted data. So if you want the ending, this is a good moment to subscribe. First, the technicality I asked you to hold on to. To grade a machine on random circuit sampling, someone has to work out what the right answers look like, and working out those answers is the exact task that takes billions of years. Aaronson spelled out the consequence for Willow, and it lands with full force in her face. For the same reason the result would take 10 to the 25 years to simulate, It would also take 10 to the 25 years to directly verify. The teams do what they can. They run smaller versions of the circuit that a supercomputer can still confirm. They measure error rates and they extrapolate upward, which is honest work. But the flagship number itself, the billions of years, no computer on earth can mark that exam. Aronson calls it an unverifiable regime. And think about what that means for a second. The machines now produce answers that nothing in the classical universe can check. We believe them anyway, because every smaller test that could be checked came back right. So there is a box in her face delivering in minutes work the visible universe doesn't seem to have room for, and the one referee who could tell us how. A classical computer redoing the math would need billions of years to finish the job. The intelligence world stopped waiting for that argument years ago in August 2021 the NSA put out a warning in plain English adversaries may be collecting encrypted data now waiting for the day when quantum computers can decrypt it the strategy has a name harvest now decrypt later and it works like this the encrypted traffic you produce today your banking sessions your medical records diplomatic cables that still matter in 30 years can be recorded in transit and parked in storage. It costs the collector almost nothing, which is exactly why the NSA felt the need to warn about it out loud. Then one day, somewhere, an error-corrected quantum machine matures, runs Shor's algorithm, the same one from Deutsch's riddle, and the entire archive opens retroactively. Every message you have ever sent through a fiber cable is a candidate, and changing your password tomorrow changes nothing, because the copy in the archive keeps the old lock. That's why the NSA has ordered American government systems onto quantum-resistant encryption between 2030 and 2033. New codes built on math problems quantum machines have no known shortcut for. And why? 23 days after the Su Chongzhi 3.0 paper, Washington added more than 50 Chinese companies to its export blacklist, seven of them specifically for advancing China's quantum program. Another round of quantum-related additions followed in September. You don't sanction a philosophy experiment. Then, in December 2025, came the part that actually moves those deadlines. Pan's group unveiled Su Chongzhi 3.2, 107 qubits, and this one wasn't chasing a benchmark record. It demonstrated quantum error correction below the fault tolerance threshold, published as another cover paper in Physical Review Letters. Qubits are fragile. Heat, vibration, a stray photon, and the blend of zero and one collapses. Which is why every machine in this story still makes errors constantly. Error correction spreads one perfect logical qubit across many imperfect physical ones. In the strongest version the Chinese team tested, roughly a hundred physical qubits stood guard around a single protected one. And below threshold means that fixing now outruns the damage. Each time the team stepped up the strength of the code, the logical error rate dropped by a factor of 1.4 per step. A sampling benchmark breaks no codes. But an error-corrected machine that keeps improving as it grows is the road that ends at Shor's algorithm. And China just showed it's on that road, with its own architecture. A second route, parallel to Google's, toward the machine the NSA is bracing for. So go back to Hefei one more time. Few hundred seconds of runtime, on a circuit built so deep no published shortcut can touch it. Graded on an exam no computer can mark. Inside a machine whose qubits are, according to the physics everyone shares, genuinely holding all those possibilities at once. The machine works, the predictions hold, and the people who built these processors for a living still can't agree on where the computation happens. Nevin says the work is spread across parallel universes. His critics say he's confusing a mathematical space with real universes. Deutsch asked this question in 1997. Where was the number factorized? And 28 years later, nobody has answered him. The machines have only made it bigger. From 200 seconds against 10,000 years, to minutes against numbers that outrun the age of the universe. And for the first time since he asked it, the machines making the question bigger are coming out of Hefei.