• 3050 Photons! Jiuzhang 4.0 Emerges, as China‘s Quantum Computing Once Again Leaves the World Behind

    The news hit the scientific community like a thunderclap. While most of the world was sleeping, a team of researchers in Hefei, China, quietly published a paper in Nature that just turned the entire field of quantum computing upside down. They call it Jiuzhang 4.0. And it’s terrifyingly fast.

    Let me put it in numbers that you can actually wrap your head around.

    Jiuzhang 4.0 can manipulate and detect up to 3,050 photons simultaneously. To give you some context, its predecessor, Jiuzhang 3.0 released just three years ago, could only handle 255. That is not just an upgrade; it is a seismic leap of more than ten times in raw quantum scale.

    But here’s where it gets insane. When tasked with solving a specific mathematical problem called Gaussian boson sampling, Jiuzhang 4.0 generates a single sample in just 25 microseconds. For those of you counting at home, that is 25 millionths of a second. A blink of an eye? That takes about 350,000 microseconds. This thing works thousands of times faster than a blink.

    How fast is the world’s most powerful supercomputer at doing the same thing? That machine is the United States’ El Capitan, a $600 million beast currently sitting at the Lawrence Livermore National Laboratory. To crack the same calculation using the best classical algorithms available, El Capitan would need over 10 to the power of 42 years.

    Let me write that out for you: 10,000,000,000,000,000,000,000,000,000,000,000,000,000,000 years.

    The age of the universe, by the way, is only about 13.8 billion years. To put it in perspective, El Capitan would need to run for longer than the entire history of the cosmos 10 tredecillion times over to finish what Jiuzhang 4.0 does before you can even say “quantum supremacy.” The speed advantage here is 10 to the 54th power. That is a one followed by 54 zeros.

    If you are a fan of Chinese internet slang, you have probably seen a meme lately that just says “click” and then shows a screenshot of this quantum advantage. That is basically the Chinese netizens‘ way of saying, “Checkmate. We just ended the game.”

    Now, as someone watching this from abroad, you might be asking a very reasonable question. Is this just a marketing stunt? Didn’t Google claim quantum supremacy back in 2019 with their Sycamore processor? And didn’t classical algorithms catch up to them pretty quickly?

    That is exactly the right concern to raise. And it is precisely why Jiuzhang 4.0 is such a big deal.

    The history of quantum computing claims has been a bit of a cat-and-mouse game. A team announces a huge breakthrough, and then within months, some clever mathematicians on the classical side come up with a better algorithm that shrinks the gap. This has happened repeatedly. When Google’s Sycamore claimed quantum advantage, a Chinese team actually showed up later with a smarter classical algorithm that could match its performance. When earlier versions of the Jiuzhang series made headlines, classical algorithms using something called matrix product states clawed their way back into the competition.

    The fear in the quantum community has always been that the gap might not be real. Maybe the quantum computers are just noisy, inefficient machines that only look impressive because nobody has bothered to write the right classical software yet.

    Here is where Jiuzhang 4.0 shuts that entire debate down.

    The research team took the fight directly to the classical side. They benchmarked their results against every known state-of-the-art classical simulation method, including the very matrix product state algorithms that had threatened previous records. The conclusion? The classical algorithms get absolutely obliterated. Even the most sophisticated matrix product state simulations running on El Capitan still need 10 to the 42 years. There is no catching up this time. The quantum advantage is not just large; it is astronomically, cosmically, almost comically enormous.

    Jonathan Lavoie, a researcher at the Canadian quantum startup Xanadu, which previously held a GBS record of 219 photons, told New Scientist that the results are, without a doubt, an impressive technical achievement. Chris Langer at Quantinuum, a company that has previously demonstrated quantum advantage with a completely different type of quantum computer architecture, also weighed in, saying he thinks it is important that quantum systems can prove they are not simulable by classical machines.

    The international community is not dismissing this. They are impressed. And perhaps a little nervous.

    So how did the Chinese team pull this off? What is the secret sauce that took them from 255 photons to over 3,000?

    The answer lies in a clever piece of engineering that solved one of the biggest headaches in photonic quantum computing: photon loss.

    Think of it like this. In a photonic quantum computer, you are shooting individual particles of light through a massive maze of mirrors, beam splitters, and delay loops. Every time a photon bounces off something, there is a chance it just disappears. In the early days of photonic quantum computing, increasing the scale of the machine meant adding more physical components. More mirrors, more splitters, more loops. But each new component is another place where a photon can get lost. This was the dead end that everyone was running into. You could not scale up the machine without scaling up the loss, which completely killed the performance.

    The Jiuzhang 4.0 team did something radically different. Instead of just adding more stuff, they redesigned the architecture from the ground up. They created something called a spatiotemporal hybrid encoding circuit. What that fancy phrase means is that they let their photons interfere with each other in both space and time simultaneously. By spreading each photon across thousands of temporal and spatial modes, they achieved a cubic scaling of connectivity. The network became vastly more connected without needing to build a proportionally larger physical device. This is the breakthrough that allowed them to achieve 92 percent source efficiency and an overall system efficiency of 51 percent.

    And then there is the sheer scale. The system operates in a Hilbert space of dimension 10 to the 2,461. To put that number in context, the observable universe contains about 10 to the 80 atoms. This machine is processing information in a mathematical space that is so unfathomably vast that you cannot even begin to visualize it. Classical computers do not go there. They simply cannot.

    Of course, the skeptic will still say that Jiuzhang 4.0 is a specialized machine. It is not a general-purpose quantum computer. It cannot run Shor‘s algorithm to crack RSA encryption, and it cannot simulate complex molecules for drug discovery. It does one very specific thing: Gaussian boson sampling. That is true. But that narrow focus is exactly what makes this achievement so significant.

    Gaussian boson sampling is not just some random academic puzzle cooked up to look impressive. It has real-world applications. In the short term, it can be used for graph theory calculations and image recognition. But more importantly, it is directly related to generating bosonic error-correcting codes and large-scale quantum entangled cluster states. These are the building blocks of fault-tolerant universal quantum computers. You cannot build a quantum computer that works in the real world without solving the error correction problem first. And Jiuzhang 4.0 just demonstrated a path to generating the kind of massive quantum states you would need for that.

    According to the team, this puts them on track toward constructing what they call trillion-quantum-mode three-dimensional cluster states. That is the hardware foundation for future fault-tolerant photonic quantum computing. In other words, the specialized machine of today is laying the groundwork for the general-purpose machine of tomorrow.

    Now, let us zoom out and look at the bigger picture. Because as a foreign observer, what is perhaps most striking here is not just the machine itself, but the trajectory.

    This is not China‘s first rodeo. The Jiuzhang series has been on a relentless march since 2020, when the original version with 76 photons first demonstrated quantum advantage in an optical system. That made China the second country in the world to achieve the milestone, and the first to do it using photons. Then came Jiuzhang 2.0 with 113 photons, and Jiuzhang 3.0 with 255. Each iteration widened the lead. Meanwhile, the same research team, led by Pan Jianwei, Lu Chaoyang, Zhang Qiang, and Liu Naile, also built the Zuchongzhi series of superconducting quantum computers. In 2021, Zuchongzhi 2.0 with 66 qubits outperformed Google‘s Sycamore, which had 53. China became the only country in the world to achieve quantum computing supremacy in two completely different technical routes at the same time. Photonic and superconducting.

    Let that sink in. The United States, for all its investment, for all its Silicon Valley billionaires and national labs, cannot claim that. Google has a superconducting chip. IBM has a superconducting chip. But nobody in the West has demonstrated quantum advantage in photonic computing at anywhere near this scale. Xanadu held the previous record for Gaussian boson sampling. That record was 219 photons. Jiuzhang 4.0 just blew past that with 3,050. That is not competition. That is domination.

    A recent analysis suggested that China is now about five to eight years ahead of the United States in the race for quantum supremacy. Looking at the numbers coming out of Hefei, that estimate might actually be conservative. The West is not just trailing. It is getting lapped.

    And what is the rest of the world doing about it? Europe is scrambling. The European Union has poured billions into its Quantum Technology Flagship program, but progress has been incremental at best. Japan declared 2025 the year of quantum industrialization, but their breakthroughs remain largely theoretical. South Korea is throwing money at the problem, but they are years behind. The United States, despite having the world‘s largest economy and the most aggressive private sector investment in quantum computing, seems to have placed most of its bets on superconducting qubits. That is a viable path, but it is a path that is now running in parallel with China, not ahead of it. And in the photonic lane, China is essentially running alone.

    Perhaps the most chilling detail in the whole Jiuzhang 4.0 announcement is not the numbers. It is the quiet confidence in how the team talks about the future. They are not celebrating this as a final victory. They are treating it as a step along a planned roadmap. A roadmap that leads to universal fault-tolerant quantum computing. They have already solved the photon loss problem that everyone thought was the killer. They have already demonstrated a path to scaling up that does not require an explosion in hardware complexity. And they are already talking about the next milestone: constructing those trillion-quantum-mode cluster states.

    What happens when they get there? What happens when China not only builds the fastest specialized quantum computer in the world, but also the first general-purpose, error-corrected quantum computer that can actually do useful work?

    The implications stretch far beyond academic bragging rights. Quantum computing threatens to break most of the encryption that secures the global internet. It promises to revolutionize materials science by simulating molecular interactions that classical computers cannot touch. It could transform logistics, drug discovery, financial modeling, and artificial intelligence. Whoever gets there first does not just win a trophy. They get to set the standards. They get to define the protocols. They get to decide how this new form of computing integrates into the global economy. And they get to do it with technologies that may be subject to export controls and intellectual property restrictions that leave the rest of the world dependent on them.

    American politicians have been warning about the quantum race for years. The National Quantum Initiative Act was signed into law back in 2018. But watching Jiuzhang 4.0 hum through its calculations in 25 microseconds, it is hard to escape the feeling that the warnings may have come too late. The lead China has built in photonic quantum computing is not something that can be closed by throwing more money at the problem. It took years of focused research, brilliant engineering, and a level of patience that the Western venture capital model rarely allows for. This is not a sprint. It is a marathon. And China just lapped the field.

    As the news spreads across the global tech media, the reactions have been a mix of awe, grudging respect, and barely concealed panic. Mainstream outlets in the United States and Europe are trying to frame the story as “another step forward” or “a significant but narrow achievement.” But the subtext is unmistakable. They know what this means. They know that controlling 3,050 photons is not an incremental improvement. It is a statement. A statement that in the race to build the computers of the future, the finish line may already be in sight. And the people crossing it first are not in Silicon Valley. They are not in Boston. They are not in Cambridge or Zurich or Tokyo. They are in Hefei, China, standing in front of a tabletop of lasers, mirrors, and fiber optics, watching a machine do in 25 microseconds what the most powerful supercomputer on Earth could not finish if it ran until the heat death of the universe.

    We are not living in a world where quantum computing is a distant future technology anymore. Jiuzhang 4.0 just proved that. The future is here. It is blinking 3,050 times per calculation. And it speaks Chinese.

  • Masayoshi Son’s $100 Billion Hail Mary: The Company That Hasn’t Even Opened Is Already Rushing to IPO

    If you had to pick the single craziest story of the 2026 global capital markets, this one would be a top contender:

    A company that hasn’t even officially launched yet is already being prepped for an IPO.

    And the price tag the boss has in mind? $100 billion.

    This is not a joke. On April 30, 2026, multiple outlets reported that SoftBank is setting up a new company in the U.S. called Roze AI—an artificial intelligence and robotics firm. The plan is to take it public as early as the second half of this year, at a valuation of around $100 billion.

    The market, predictably, lost its mind. Some people admire Masayoshi Son’s boldness. Others think he’s back to painting castles in the sky. Plenty of people started whispering about WeWork again—because that debacle, after all, started in a pretty similar way.

    The man at the center of this story is 67-year-old Masayoshi Son. He might be the most polarizing figure in global investing. A 100millionbetonYahooanda100millionbetonYahooanda60 million bet on Alibaba both turned into legendary returns. But the WeWork and Uber disasters also gave him the title of “world’s biggest investment loss” for a while.

    Now he’s going all-in on artificial intelligence. And Roze AI? It’s the riskiest card in that bet.


    1. The AI Obsession: What Exactly Is Son Betting On?

    Honestly, if you don’t understand Masayoshi Son the person, you won’t understand why Roze AI is being born this way.

    In investment circles, Son has a nickname: “Madman.” It’s not an insult—it’s more a recognition that his way of placing bets is beyond what any normal person would consider sane.

    In 2017, he shocked the world with the Vision Fund, a 100billionwarchest.Hedidnthavethatkindofcashlyingaround.Hebuiltitusingamixofpreferreddebtandequity,basicallyengineeringitintoexistence.In2018,hewentfurther,pledgingpartofSoftBanksAlibabastaketoraise100billionwarchest.Hedidnthavethatkindofcashlyingaround.Hebuiltitusingamixofpreferreddebtandequity,basicallyengineeringitintoexistence.In2018,hewentfurther,pledgingpartofSoftBanksAlibabastaketoraise8 billion in cash.

    For Son, debt isn’t a risk—it’s an amplifier.

    After the WeWork fiasco, most people assumed he’d dial things back. He didn’t.

    In 2025, Son was back with a vengeance. This time, his target was the entire AI value chain. Chips, algorithms, data centers, robotics—he wanted a piece of everything.

    Just look at the moves: he loaded up on Nvidia, scaling from 1billionto1billionto3 billion. At the same time, he poured money into OpenAI, completing a staggering $40 billion investment by the end of 2025.

    But the real head-scratcher came next.

    In November 2025, SoftBank sold its entire Nvidia stake for 5.83billion.NvidiawasliterallythehottestAIstockontheplanet.Sonsoldhisshovelmakershares,andalmostatthesametime,heannouncedanadditional5.83billion.NvidiawasliterallythehottestAIstockontheplanet.Sonsoldhisshovelmakershares,andalmostatthesametime,heannouncedanadditional22.5 billion bet on OpenAI through the Vision Fund.

    From “selling shovels” to “digging for gold” yourself—that’s Son’s AI philosophy. He doesn’t just want to own shares in chip companies. He wants to own an AI empire.


    2. Dropping $34.6 Billion… and the Numbers Are Getting Scary

    But empires aren’t cheap.

    In February 2026, SoftBank disclosed a number that made a lot of people’s eyes water: the company had already invested 34.6billioninOpenAI,withanother34.6billioninOpenAI,withanother30 billion committed.

    34.6billionplus34.6billionplus30 billion. That’s $64.6 billion in total. To put that in perspective, it’s nearly half of SoftBank’s entire net worth, riding on a single bet.

    Where’s all this money coming from?

    The answer is simple: sell what you can, pledge what you can’t.

    To raise cash, Son not only dumped Nvidia, he also leveraged Arm’s equity. Sources say SoftBank even discussed selling part of its Intel position—it bought nearly 2billionworthlastAugustat2billionworthlastAugustat23 a share, and now the stock is at 96,makingthatstakeworthcloseto96,makingthatstakeworthcloseto10 billion. It’s ready to be cashed out anytime.

    But it’s still not enough.

    In March 2026, S&P downgraded SoftBank’s credit outlook from “stable” to “negative.” Their reasoning: the massive OpenAI investment could seriously weaken SoftBank’s liquidity, portfolio quality, and financial strength.

    CreditSights, a research firm, estimates that SoftBank is staring at a 32billionliquiditygap,includingbondsmaturingoverthenexttwoyearsandalreadysignedacquisitiondealslikethe32billionliquiditygap,includingbondsmaturingoverthenexttwoyearsandalreadysignedacquisitiondealslikethe5.4 billion purchase of ABB’s industrial robotics division.

    SoftBank has even resorted to borrowing against its OpenAI shares. Reports say the company is seeking a $10 billion loan, using its OpenAI equity as collateral.

    Bluntly put, Son is borrowing ammunition to keep fighting. He’s pledging assets that haven’t even gone public yet, to fund investments in companies that also haven’t gone public yet.

    If this chess game works, it’s a masterclass in capital engineering. If it doesn’t… hasn’t the WeWork lesson been brutal enough?


    3. So What Exactly Is Roze AI?

    Once you understand SoftBank’s financial squeeze, the rush to IPO Roze AI makes sense. It’s Plan B to generate cash—fast.

    Originally, Open AI’s own IPO was supposed to be the big event that would let SoftBank deleverage. But OpenAI is facing fierce competition from Google, Anthropic, and others, and doubts about its growth outlook have made the IPO timetable very uncertain.

    Plan A is shaky, so you need a Plan B.

    Enter Roze AI.

    So what does this company actually do? In simple terms: it uses robots to build data centers.

    The business has three layers. One: planning, constructing, and running data centers. Two: deploying autonomous robots to help build server farms—automation in infrastructure. Three: bundling SoftBank’s existing assets like land and data centers under one roof.

    SoftBank has already kicked off a hyperscale data center project in Ohio. Notably, the electricity for this project is backed by a $33 billion gas plant funded by the Japanese government. Son’s bet is that he can replicate this model across other U.S. states.

    On top of that, SoftBank plans to fold the recently acquired ABB Robotics into Roze AI. That gives the company both an “infrastructure” narrative and a “robotics” narrative—two of the hottest labels in today’s market, wrapped into one story.

    Sounds compelling, right?


    4. A $100 Billion Valuation… Seriously?

    But a good story doesn’t automatically mean a reasonable valuation.

    How big is 100billion?Forcomparison:Intelsmarketcapisaround100billion?Forcomparison:Intelsmarketcapisaround200 billion. Equinix, the world’s largest data center operator, is worth about $80 billion. A company that hasn’t even opened its doors yet is supposed to sail past Equinix’s entire market value? That’s not ambition—that’s fantasy.

    And there’s a key assumption baked into Roze AI’s business model: the data center boom must continue to expand rapidly and continuously. If the U.S. data center construction fever cools off, the logic behind that $100 billion price tag collapses pretty fast.

    In fact, some people inside SoftBank are already voicing doubts. Several executives think the $100 billion target is way too “ambitious,” because it relies so heavily on aggressive expansion of data center operations. Sources have also hinted that there’s serious internal disagreement about both the IPO timeline and the valuation.

    “Too ambitious.” In corporate-speak, that’s about as polite as “unrealistic” gets.

    Also worth noting: building and running data centers is an incredibly heavy-asset business. A single hyperscale facility can cost billions of dollars, and the payback period is long. Roze AI claims it will use robots to boost efficiency, but nobody has proven you can build data centers with robots at scale, let alone do it profitably. It’s hard to slap a $100 billion price tag on a business model that hasn’t even been proved yet.

    If you look at Son’s history, this story feels familiar. He gave WeWork an astronomical valuation too. That IPO imploded spectacularly, and it nearly ended his career. He later publicly admitted it was a “stain on his career,” blaming himself for pushing forward even when his team advised against it.

    Feels like we’ve seen this movie before.


    5. The Scene Outside Isn’t Helping: Three Mega-IPOs Are Looming

    Roze AI’s plan faces another massive external challenge: the U.S. IPO market in 2026 might have to digest three historic offerings at once.

    SpaceX is reportedly targeting a $75 billion IPO. OpenAI and Anthropic are also prepping to go public. At a financing summit in April 2026, multiple investors warned that all three could hit the market in the same calendar year—something virtually unheard of in capital markets history.

    One analyst pointed out that SpaceX’s offering alone could equal nearly 10% of the average daily trading volume in the U.S. stock market. Can the market absorb that much new supply all at once? Nobody knows—and that’s before you even factor in the lockup expirations that will unleash even more shares later.

    In other words, the 2026 IPO calendar is already packed. A bunch of giants are all fighting for the same limited pool of investor money. In that environment, getting a $100 billion valuation for Roze AI becomes a whole lot harder.

    The founder of Inspired Capital put it bluntly at the summit: “These could be the three largest IPOs in history, and they could all happen in the same calendar year. The performance of some of these IPOs might actually act like a bucket of cold water on reality.”


    The Bottom Line

    Masayoshi Son once said: “AI isn’t just an investment. It’s the birth of a new species—a revolution for humanity.”

    He might be right. But the way he’s betting on AI—concentrated positions, heavy leverage, rapid monetization, and packaging assets for quick IPOs—has him walking a knife’s edge every step of the way.

    Will Roze AI successfully go public? Will it justify a $100 billion valuation? Right now, nobody has the answers. But one thing’s for sure: if this fails, Son is looking at a crisis far worse than WeWork.

    Then again, if he wins…

    Well, isn’t that just the classic story of a madman turning out to be a genius?

    Masayoshi Son’s entire career has been a pendulum swinging between those two labels. The question now is: which side does this swing land on?