Google's Willow 2.0 Just Solved a Problem That Took the Universe 10 Septillion Years

Añadido: 2026-05-13

[00:00:03]There is a small superconducting chip inside a refrigerator in Santa Barbara, California. The chip is roughly the size of a postage stamp. The refrigerator surrounding it is the size of a small car. And inside it, the temperature hovers near 10 micel, colder than the deepest void of interstellar space. On the morning of the benchmark run, the engineers in the control room weren't expecting anything dramatic. They had run versions of the test before. They had watched the numbers climb. They knew roughly what was coming. What they were not entirely prepared for was the size of the gap. The chip's name is Willow.

[00:00:40]It was unveiled by Google Quantum AI in December of 2024. And the result it produced. The result we're going to spend the next 20 minutes unpacking was this. A computational task that Willow completed in under five minutes would have required the fastest classical supercomput on Earth approximately 10 septillion years. That number written out is a one followed by 25 zeros. It is by any honest accounting longer than the age of the universe by a factor of roughly 700 trillion. I want to be careful here because that sentence sounds like marketing copy and it isn't.

[00:01:17]It is a measurement. And the story of how that measurement came to exist and what the 2026 successor announcements are now suggesting it means is one of the strangest stories in modern physics.

[00:01:29]Not because of what was found in the cosmos, but because of what was found in a basement. Let me start where the engineers started with the noise.

[00:01:42]If you've ever tried to record audio in a busy cafe, you understand the central problem of quantum computing. The signal you want, the conversation, the music, the meaningful information is fragile.

[00:01:55]It is constantly being drowned out by everything else. Footsteps, refrigerator hum, traffic outside. In a quantum computer, the noise is even worse because the signal itself is built out of the most delicate physical states humans have ever tried to control.

[00:02:11]Superpositions of quantum bits or cubits that exist in a probabilistic blur until measured.

[00:02:18]For decades, the dominant fear among quantum researchers was simple. The more cubits you added to a chip, the more noise you accumulated and the more errors crept in. Scaling up didn't just get harder, it got exponentially harder.

[00:02:32]Some researchers in the field quietly worried it might not work at all. That quantum computing might hit a kind of physical ceiling where any attempt to build a larger machine produced more errors than it could correct. The benchmark Willow ran in late 2024 was a direct answer to that fear. And the answer against expectation was that the ceiling moved. The test itself is called random circuit sampling. It's an unglamorous name for a strange and very specific task. You take a quantum processor, you apply a long sequence of randomized quantum operations to its cubits, and then you measure the output.

[00:03:09]The result is a probability distribution, a kind of fingerprint of how those quantum operations interfered with one another. Reproducing that fingerprint on a classical computer requires simulating the behavior of every cubit in every possible state simultaneously.

[00:03:25]The cost of that simulation grows exponentially with the number of cubits.

[00:03:29]This is why random circuit sampling has become the standard benchmark for what researchers call quantum advantage. It is not a useful task. Nobody's going to cure cancer with a random circuit sample. But it is a clean, rigorous demonstration that a quantum machine is doing something. A classical machine in any reasonable time frame simply cannot.

[00:03:49]When Willow ran the benchmark, it used 105 cubits. The result was published in a Nature paper in late 2024 and the framing was deliberate. Google's team led by Hartmoot Nevin did not say we have built the future. They said in essence, here is a measurement. Here is what our chip did. Here is what the best classical algorithm running on the most powerful supercomput we know how to build would need to do the same task.

[00:04:20]The gap is 10 septillion years. I've thought about that number for weeks now honestly. And the thing I keep coming back to is how badly our intuitions handled it. A million seconds is roughly 11 days. A billion seconds is about 32 years. A trillion seconds takes you back past the last ice age, past the emergence of modern humans. back into the plea scene. 10 septillion years is so far beyond any of those scales that the comparison stops being meaningful.

[00:04:49]It is not a long time. It is a category error.

[00:04:55]But the truly unsettling part of the willow result wasn't the speed. It was the error correction. For years, a question had haunted the field. As you add more cubits to a quantum processor, does the error rate per cubit go up or down? If it goes up, quantum computing is essentially a dead end. You will always be chasing your own tail, adding cubits to fix problems that the new cubits themselves create. If it goes down, if errors decrease as the system grows, then quantum computing has a future, a real one. Willow, for the first time at scale, demonstrated the second outcome. Google's team showed that as they encoded logical cubits using larger and larger arrays of physical cubits, the error rate dropped exponentially. The threshold theorem proposed in the uh 1990s had predicted this was possible in principle. Willow showed it happening in practice. This is the part most coverage of the chip glossed over because it isn't as cinematic as 10 septillion years. But ask any quantum engineer which result mattered more. And most will tell you quietly that it was the error correction. The benchmark was the headline. The error correction was the foundation. Pause for a moment on that because what it means is this. The road forward isn't blocked. The ceiling researchers had feared the noise wall appears to be something we can climb past slowly, expensively with enormous engineering effort. but passed.

[00:06:29]Now we come to 2026 and to the announcements that have framed this video. In the months following Willow's December 2024 debut, Google Quantum AI began publishing a steady stream of follow-up work. The successor architectures, what some have informally taken to calling Willow 2.0, 0 comma though Google's own naming has been more restrained pushed the cubit count higher and refined the error correction further public statements from Hartmoot Nevin and his team have emphasized two thresholds the next generation is targeting the first is what they call a useful beyond classical computation a task that is not only impossible for classical computers in any reasonable time but also corresponds to a real world problem materials chemistry drug discovery battery design. The random circuit sampling benchmark is a proof of principle. The next milestone is a proof of usefulness. The second is the long-term goal of a fully fault tolerant logical cubit operating below a critical error threshold for extended computations.

[00:07:32]This sounds technical and it is, but the implication is enormous. A fault tolerant logical cubit is the building block of every serious application researchers have ever proposed for quantum computing, including eventually the algorithms that could break much of the world's current encryption. We are not there yet. The 2026 statements have been careful about that. But the trajectory is now public and the trajectory is shorter than most outside observers had assumed even 3 years ago.

[00:08:02]I want to be honest about uncertainty here. Quantum computing has been 5 years away for about 30 years. There are real reasons to be skeptical of any timeline.

[00:08:12]Engineering challenges in scaling cubits, in cooling, in fabrication consistency. These are not solved problems. They are problems being attacked with measurable progress by some of the best minds in physics and engineering. Whether that progress continues at the current rate or hits an unexpected wall is genuinely unknown.

[00:08:33]Researchers in the field have argued both sides, and they argue them in good faith. What is no longer in dispute is the basic claim. Quantum computers can do specific things that classical computers cannot in any time frame the universe has so far permitted. That is not a press release. That is a measurement replicated, peer-reviewed, and published in nature.

[00:08:58]So what does this mean? Not in the abstract, not for the stock price of any company. What does it actually mean for the way we understand computation and the way computation sits inside the cosmos? Here's where I have to slow down because the temptation is to overreach.

[00:09:14]There is a particular line of speculation associated with Hartmoot Nevin himself in interviews that touches on the many worlds interpretation of quantum mechanics. The argument very roughly goes like this. When a quantum computer performs a calculation that would require more resources than the observable universe contains, where physically is that calculation happening? Nean has suggested and he frames this as speculation not as established physics that one possible interpretation is that the computation is being distributed across parallel branches of reality in something like the sense David Deutsch proposed decades ago. I want to be very clear. This is a philosophical interpretation. It is not a confirmed result. The mathematics of quantum mechanics works whether you believe in many worlds, in pilot waves, in collapse models, or in any of a halfozen other interpretations. Willow's benchmark does not prove the multiverse.

[00:10:12]It does not prove anything about the underlying ontology of reality. What it proves is that a specific physical system built by humans can produce outputs that classical models of computation cannot reproduce.

[00:10:25]But, and this is the part I find genuinely strange to sit with.

[00:10:30]The question Navan is raising is a real question. If a piece of physical hardware in California can perform in 5 minutes a calculation whose classical equivalent would take longer than the lifespan of every star ever born, then something is happening in that hardware that is not reducible to ordinary classical resources. The interpretation of what is happening is open. that something is happening is not.

[00:11:00]Let me try a different angle, a more grounded one. Imagine you have a maze, a very large maze with many possible paths. A classical computer exploring this maze has to walk down each path one at a time. It can be clever about it. It can use heristics, prune dead ends, parallelize across many processors.

[00:11:21]But fundamentally, it is a walker. It places one foot in front of another. A quantum computer exploring the same maze is doing something different. It is in a sense allowing all possible paths to interfere with one another, like ripples on the surface of a pond until the paths that lead to dead ends cancel themselves out and the paths that lead somewhere meaningful reinforce. It is not faster because it walks faster. It is faster because it is not in the ordinary sense walking. That analogy is imperfect.

[00:12:03]is not that Willow is a faster classical computer. It is that Willow is doing a fundamentally different kind of thing.

[00:12:09]The comparison is almost unfair. It is like comparing the time it takes to count grains of sand one by one versus the time it takes for a wave to know the shape of the beach.

[00:12:22]There is a quieter implication of all this that I think gets missed. For most of human history, computation was something we did with our minds, with paper, with mechanical aids. The idea that the physical universe itself might be running calculations, that every photon scattering off every surface, every electron orbiting every nucleus is in some sense performing a computation is a relatively recent intellectual move. it belongs to the late 20th century to people like Richard Fineman who in 1981 gave a lecture at MIT in which he argued that nature is not classical and that if we wanted to simulate it properly we would need to build computers that were themselves quantum. Willow is the late slow vindication of that lecture 43 years on a chip the size of a postage stamp in a refrigerator colder than space running a benchmark that has no classical equivalent. The argument Fineman made in a half empty auditorium has now been rendered as a number. And the number is 10 septillion years. I find that lineage moving in a way I don't entirely know how to articulate. Science usually doesn't deliver on its promises within the lifetimes of the people who made them. Fineman didn't live to see Willow, but the chip is in a real sense his.

[00:13:40]So where does this leave us in 2026 watching the successor announcements roll out? It leaves us in a strange interim. The era of classical only computation is not over. It will not be over for a long time. Most of the calculations the world depends on, including almost certainly the device you're watching this video on, will remain classical for the foreseeable future.

[00:14:04]Quantum machines are not replacements.

[00:14:06]They are specialists. They are extraordinarily good at a small growing class of problems and useless at most others. But the ceiling has been tested.

[00:14:14]The benchmark has been published. The error correction works. And the question now is not whether quantum computing is possible. It is what we do with it when it arrives at usefulness. Some of the implications are uncomfortable.

[00:14:27]Encryption schemes that protect global financial systems are vulnerable in principle to sufficiently large quantum computers running an algorithm Peter Shore proposed in 1994. The postquantum cryptography effort is real, ongoing, and racing the hardware. Some of the implications are extraordinary.

[00:14:47]Simulations of molecular behavior that have eluded chemistry for a century may within a decade or two become routine.

[00:14:54]Drug discovery, fertilizer synthesis, room temperature superconductors.

[00:15:00]These are the targets researchers talk about cautiously when asked what quantum computing might actually do. And some of the implications, I suspect we have not yet imagined. New tools tend to reveal questions that the old tools could not formulate. The microscope did not just confirm what biologists suspected. It showed them a kingdom of life they had not known existed.

[00:15:23]A final thought before I let you go.

[00:15:26]When the engineers in Santa Barbara watched the benchmark complete, the chip itself did nothing visibly remarkable.

[00:15:32]It sat in its refrigerator. Its temperature did not change. The walls of the building did not shake. There was no flash, no sound, no sense of ceremony.

[00:15:43]Somewhere inside that small cold object, 10 septilion years of classical computation had been bypassed in 5 minutes, and the only evidence was a probability distribution on a screen. I think about that often, the quietness of it, the way the deepest ruptures in our understanding of the universe almost never announce themselves. They arrive in graphs, in tables, in logs that nobody outside a small group of specialists will ever read. And it is only later, sometimes years later that the rest of us look up and realize the ground beneath us has shifted. Willow shifted it. The successor chips are in their own quiet way shifting it again.

[00:16:25]Whether the floor holds, whether the trajectory continues is something we will only know by watching. But the measurement is real. The number is real.

[00:16:34]And the universe, as it turns out, has fewer years available than a quantum computer needs minutes. If you've stayed with me this far, thank you. I know this one was denser than most. The next episode is going to look at something stranger and quieter and possibly older than any chip. A signal pattern in deep space radio data that has not yet been ruled out as natural. I'll see you