Moonshot AI rents Nvidia chips in Europe; Micron pauses fab

Samantha: Welcome to The State of Tech, The European Edition, Friday November six, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: a Chinese AI lab reportedly taps Nvidia's top chips through data centres in Germany and the Netherlands, Micron freezes a four and a half billion dollar chip plant in Taiwan, Anthropic speeds up while its own CEO calls for a slowdown, and tech stocks tumble on fears of an AI crackdown in Washington. Plus two things to try yourself: the free photo enlarger Upscayl and the disk space mapper WinDirStat. Let's start with that chip route through Europe.

Moonshot AI reportedly uses Nvidia's newest chips by renting data centre capacity in Germany and the Netherlands.

Samantha: Moonshot AI, the Beijing-based AI lab, has reportedly found a way to use Nvidia's most advanced AI chips without importing a single one. Instead of shipping the hardware to China, the company is renting computing power from data centres in Germany and the Netherlands. The chips stay in Europe, and the work gets done remotely. According to the reports, the arrangement officially started on Thursday. It involves Nvidia's Blackwell generation, the newest top-end chips the company makes for training and running large AI models.

Bob: And that's the crux. American export controls are designed to limit shipments of advanced chips to China. Leasing capacity in a server hall in Germany or the Netherlands works differently. The hardware never crosses the border. Only the instructions and the results travel back and forth. So the chips physically stay put, while a company in Beijing effectively gets to use them. That's why the reports describe this as circumventing the restrictions. Whether any rules were actually broken, the reports don't say, and that distinction matters.

Samantha: It also shows how hard technology sanctions are to enforce once computing power is sold almost like electricity. You don't need to own a chip to use it, you need a contract with someone who does. And demand is enormous. Labs in the US, in China and in Europe are all chasing the same limited supply of top-end hardware. If the most advanced chips are hard to get at home, renting them abroad becomes the obvious workaround. This is a race for infrastructure just as much as a race for better models.

Bob: For Europe, this puts data centre operators in an awkward spotlight. Facilities in Germany and the Netherlands suddenly look like a key link in the global AI supply chain, and not only for European customers. That invites questions about who exactly is renting this capacity, and whether renting compute should be treated the same way as exporting a chip. European operators could find themselves caught between American pressure and lucrative foreign demand, with regulators on both sides of the Atlantic looking at the same server halls.

Samantha: The bottom line: the reports raise a basic question about how effective American export controls really are. If compute can be rented across borders, controlling physical shipments only covers part of the picture. For European companies that rent out capacity, there's a commercial opportunity here, but also a political risk. And for European AI developers competing for the same scarce chips, it's a reminder that they're not the only ones knocking on the doors of those data centres.

Micron halts a four and a half billion dollar memory chip plant in Taiwan over a labour dispute.

Bob: Micron, the American memory chip maker, has hit the brakes on a four and a half billion dollar plant in Taiwan. Construction at the site in Taoyuan has largely stopped. The reason isn't technology or demand, it's people. A drawn-out labour dispute has made the company wary, in particular the possibility that a potential agreement with a local union could be overturned. Micron points to those uncertainties as the reason for the pause. And Taiwan is a crucial production base for the components that go into AI systems, so this reaches far beyond one construction site.

Samantha: Memory chips are the less glamorous half of computing. The processor does the thinking, and memory holds everything it's working on. Every AI server needs huge amounts of it. The market is already short on memory, and a stalled plant means less new supply coming online at exactly the wrong moment. The concern is that this halt makes the existing shortage worse, with knock-on effects for every company that depends on these components, from cloud providers to the makers of AI hardware.

Bob: What stands out is the cause. Chip projects usually stall over subsidies, export rules or energy prices. Here it's an unresolved labour conflict, and the fear that a deal on the table might not hold. For a company planning a multi-billion dollar factory, that uncertainty was apparently enough to stop building. It's a reminder that the global chip supply chain doesn't only run on clever engineering and government money. It also depends on stable agreements with the people who build and run the factories.

Samantha: For Europe, the effect is indirect but real. European tech companies and the wider AI industry buy memory on the same global market as everyone else. Less supply generally means longer waits and higher prices, and those costs can eventually trickle down to businesses and consumers. It also feeds the ongoing European debate about depending on a handful of production hubs on the other side of the world. Every disruption in Asia, whatever its cause, becomes another data point in that discussion.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

Anthropic is developing AI twice as fast as a year ago, while its CEO urges a slowdown.

Bob: Anthropic is building its AI systems roughly twice as fast as it did a year ago. That's the finding of an independent monitoring group, in an analysis published on Thursday. The awkward part is the contrast. Anthropic's CEO, Dario Amodei, has publicly argued that AI is improving too quickly, and that the pace should slow down to give safety measures time to catch up. So the company's own trajectory is now being held up against its boss's words.

Samantha: You have a CEO saying, in effect, ease off the accelerator, while an outside count suggests his own company is pressing harder than before. The analysis highlights that tension between rapid progress and the growing debate about responsible development inside the leading labs. And it reflects a dynamic that critics of the industry often point to: individual labs may worry about the overall pace, but none of them wants to fall behind the competition.

Bob: The analysis points to Anthropic's internal development capabilities as one factor behind the faster pace. In other words, the company has built up the capacity to move quicker, and that capacity doesn't switch off just because the CEO argues for restraint in public. For outside observers, the key question is whether safety work can keep up when development itself is speeding up.

Samantha: Safety research is moving too. Also on Thursday, researchers announced a new method for controlling how AI systems behave, aimed at making them more predictable and keeping them doing what they were designed to do. Details are still limited, but it's described as a significant step for AI alignment. Whether methods like that can be developed and adopted as fast as the systems they're meant to control is exactly what the Anthropic findings put in question.

Bob: For Europe, this adds urgency to the discussion on AI safety and governance. European debates about AI rules often come down to how far companies can be trusted to balance speed and safety on their own. When an independent analysis shows that even a lab known for its safety message is accelerating, policymakers get a sharper picture of the competitive pressure these companies face, and of the limits of voluntary restraint.

Tech stocks lose trillions as investors fear an AI crackdown after the US midterm elections.

Samantha: Wall Street has turned nervous about artificial intelligence, and politics is the trigger. After Democratic victories in the American midterm elections, which we covered yesterday, investors dumped tech stocks. The tech-heavy Nasdaq index fell sharply on Thursday, and the selloff has wiped out trillions of dollars in market value. The fear is that a new political balance in Washington could mean a much tougher climate for the AI industry and for the data centres that power it.

Bob: Investors are reacting to three specific risks. The first is moratoriums, temporary bans on building new data centres. The second is stricter rules for the AI industry more broadly. The third is the possible loss of tax incentives that have made building data centres more attractive. Data centres are the factories of the AI boom, so anything that slows their construction hits the whole chain of companies that build, supply and depend on them. That's why the selling spread so widely.

Samantha: The reports also point to widespread skepticism about the future of the AI trade itself. Enormous amounts of money have gone into AI on the expectation of fast, uninterrupted growth, and this week shows how sensitive that bet is to anything that might slow it down. It's worth being precise, though. The selloff is driven by fears of what could happen. So far these are prospects and possibilities, not adopted laws or confirmed policy changes.

Bob: For Europe, ripple effects are possible. European markets open in a few hours, and investor sentiment rarely stops at the Atlantic. Beyond the trading floor, there's a longer-term question. If the mood around AI sours, that could affect investor appetite for AI infrastructure projects across the continent as well. How much of this nervousness carries over, and for how long, remains to be seen.

Upscayl is a free, open-source app that uses AI to enlarge and sharpen small photos.

Samantha: To close out, two things you can actually use today. First up: Upscayl, a free, open-source desktop app that makes small, blurry images bigger and sharper with AI. You drag in a photo, pick an enhancement style, choose where to save the result, and the app enlarges the image several times over. It's built for exactly those moments when the only copy of a picture is a tiny thumbnail, an old phone snap, or a compressed image someone sent you.

Bob: The key detail is that it runs on your own computer. The images are processed locally, not uploaded to an online service, which matters if you're working with family photos or anything you'd rather not send to a stranger's server. Upscayl is available for Windows, Mac and Linux, and you can download it from the project's own website. It also offers a batch mode, so you can point it at a whole folder instead of feeding in photos one at a time.

Samantha: There's a hardware catch. Upscayl relies on your graphics card to do the heavy lifting, so it works best on a reasonably modern machine. On older laptops it can be slow, and some very basic graphics chips aren't supported at all. Reviewers also note that it offers several AI models: some are tuned for photos, others for drawings and digital art, and picking the right one makes a visible difference.

Bob: One honest limitation: the AI doesn't recover lost detail, it invents plausible detail. That can look great on a landscape, but faces and text sometimes come out subtly wrong. So it's fine for a print on the wall or a sharper wallpaper, and not something to rely on where accuracy counts. Getting started takes about ten minutes: download, install, drag in a photo, and compare the result side by side with the original.

Samantha: Practical tip: start with one photo you know well, so you can spot where the AI got creative. And keep your original files. Upscayl saves a new copy rather than overwriting the source, but it's worth checking your settings before running a big folder through it.

WinDirStat shows Windows users exactly what is eating up their hard drive, as a colourful map.

Bob: The second one has no AI at all: WinDirStat, a free, open-source tool for Windows that shows you where all your disk space went. You point it at a drive, it scans everything, and then it draws a map of coloured blocks. Every block is a file, and the bigger the block, the more space that file takes. Colours show the file type, so videos, photos and program files each get their own shade. A forgotten folder of old videos jumps out immediately.

Samantha: The app combines that map with a plain list of folders, sorted by size, so you can drill down from the drive to the exact folder causing the problem. Click a block and it shows you which file it is and where it lives. From there you can open the folder or delete the file directly. For anyone whose laptop keeps warning about low storage, that turns a vague annoyance into a concrete list of things to clean up.

Bob: A word of caution: the app will happily show you system files and program folders too, and deleting those can break things. Users generally recommend sticking to your own documents, downloads and media, and leaving anything inside the Windows or Program Files folders alone. A full scan of a large drive can take a few minutes, so it's a good job to start before making a coffee.

Samantha: The limitation to know: WinDirStat is Windows only, so Mac and Linux users will need a different tool. For Windows users it's free, needs no account, and a first cleanup takes roughly fifteen minutes. Check your downloads folder first. That's typically where installers and old files pile up unnoticed, and where the biggest blocks tend to hide.

Samantha: Today we covered: a Chinese AI lab reaching Nvidia's top chips through European data centres, Micron freezing its Taiwan plant over a labour dispute, Anthropic speeding up despite its CEO's call for a slowdown, the tech selloff over fears of an AI crackdown, and two things to try yourself: Upscayl and WinDirStat.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.