Quantum error correction: Google Willow, China AI rules

Samantha: Welcome to The State of Tech, The European Edition, Sunday October eighteen, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Google claims a major quantum error-correction breakthrough; China turns on an immediate export registration for twenty-three advanced AI technologies; Meta drops a big new open-weight AI model called Muse Spark Open; China's robotics industry becomes a net exporter and dominates humanoid shipments; the AiMOGA Mobo family robot opens for pre-orders; and to close out, two things you can actually use today, the free open-source e-book tool Sigil and the open map app StreetComplete. Let's start with Google.

Google's Willow quantum chip cuts errors forty-fold with a new technique called magic state cultivation.

Samantha: Google's Quantum AI team says it has pulled off something the whole field has been chasing for years. On its Willow processor, researchers demonstrated a method called magic state cultivation, and they report it cuts errors by roughly forty times. In plain language, magic states are the special ingredients a quantum computer needs to run the full range of calculations, not just the easy ones. Producing them cleanly has been painfully expensive in computing resources. Google says it now reaches a logical fidelity of ninety-nine point ninety-nine percent for one key type.

Bob: And that number matters because quantum computers are famously noisy. Every extra qubit you throw at the problem is also an extra source of error, so scaling has felt like running up a down escalator. The new trick uses coordinated measurements to spot and throw out faulty attempts before they poison the calculation. Google frames this as a shift from physics experiment to engineering discipline. That is a loaded claim, but even sceptics agree the resource cost of producing these magic states has been the main bottleneck for a universal, fault-tolerant machine.

Samantha: For a sense of scale, this is not a working quantum computer that breaks encryption tomorrow. It is a building block, but a very important one. The industry roadmap usually talks about millions of physical qubits to get a few thousand reliable logical ones. If error rates drop by orders of magnitude, that ratio improves and the hardware bill shrinks. European research groups in Delft, Munich, Paris and Copenhagen have been pushing hard on error correction too, often with different qubit types, so this result sets a new bar they will try to match or beat.

Bob: It also sharpens the policy question in Brussels. The EU Quantum Flagship and the Chips Act both earmark money for exactly this kind of work, but most capacity still sits in the US and China. Industry voices argue Europe needs to secure its own supply of control electronics, cryogenics and fabrication, otherwise it ends up renting cloud quantum time from American hyperscalers. For European banks, telecom operators and defence agencies planning long-term encryption migration, this is another nudge to accelerate the switch to post-quantum cryptography now, not when the headlines force them.

Samantha: The bottom line: a lab milestone that moves quantum computing one meaningful step closer to being useful, with real strategic consequences for how Europe plans the next decade.

China enacts immediate export registration on twenty-three advanced AI technologies, including large models and chip design.

Bob: China has moved fast on AI exports. The Ministry of Commerce and the Ministry of Science and Technology jointly issued a revision covering forty technologies, and for twenty-three of them an export registration requirement took effect immediately. The scope is broad: large AI models, AI agents, computing clusters, and chip design tools for AI processing and memory chips. Beijing frames it as preventing misuse, protecting national security, and keeping humans in control of sensitive AI capabilities. Public comments are open until the fifteenth of November, but the rules are already in force.

Samantha: The timing is striking. This lands just days after we talked about the earlier proposal on this show, and now it is operational. For European companies, that means a licensing deal with a Chinese partner, a joint research project on AI agents, or even technical support contracts around chip design software could trigger a Chinese registration obligation this week. Legal teams at carmakers, industrial groups and research universities will be reading the lists line by line. The question is which specific model sizes, cluster configurations and design tools fall inside the net.

Bob: And it mirrors what Washington has been doing in reverse. The US restricts advanced chips and tools flowing into China; Beijing now registers what flows out. Caught in the middle sit European firms that want to sell into both markets. ASML already lives with that reality on lithography. Add AI models and agents to the list and the compliance workload grows sharply. There is also a quieter signal here: China is making clear its own AI stack is now valuable enough to control, not just import. That is a notable shift in self-perception.

Samantha: For European regulators, this complicates the AI Act rollout. Companies deploying Chinese foundation models in Europe will now face overlapping obligations from Beijing and Brussels. Expect industry groups to push for clearer guidance from the European Commission on dual compliance, especially for sectors like automotive, logistics and consumer electronics where Chinese AI components are already common.

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.

Meta releases Muse Spark Open, its strongest US open-weight AI model, aimed at coding and autonomous workflows.

Bob: Meta has launched Muse Spark Open, a new large open-weight AI model tuned for coding and autonomous workflows. The company calls it its strongest US open-weight release so far. Open-weight means outside developers can download the model's parameters and run it on their own hardware, rather than only calling it through an API. That is a notably different stance from most frontier labs right now, which keep their best models closed. The release lands in the middle of a very hot debate about who should control advanced AI.

Samantha: And that debate is exactly where this story matters. Governments increasingly describe advanced AI as a strategic asset, something close to a controlled export, which we just heard with China. At the same time, Meta is handing out a frontier-class model for anyone to deploy. For European developers, startups and research groups, that is a genuine shift. Running a strong coding and agent model locally means less dependence on US cloud APIs, lower long-term cost, and the ability to keep sensitive code and data on European soil. It is a real alternative, not a toy.

Bob: The tension is obvious. Supporters say open-weight models let smaller players compete, let academics audit behaviour, and let regulators actually inspect what is inside. Critics argue the same openness lowers the barrier for misuse, from automated vulnerability discovery to deepfake pipelines. The EU AI Act carved out specific obligations for general-purpose models above certain capability thresholds, with lighter treatment for genuinely open releases. Muse Spark Open is going to be a live test of where exactly that line sits in practice, and how European regulators interpret it.

Samantha: For companies evaluating it, the practical question is maintenance and liability. Running your own model means owning the updates, the safety filters and the compliance paperwork. The bottom line: Europe gets another serious open option for building AI into products, but with real responsibility attached.

China's robotics industry becomes a net exporter and ships ninety-seven percent of the world's humanoid robots.

Bob: China's robotics sector has crossed a significant line. In 2025, industrial robot exports from China exceeded imports for the first time. And in the first half of this year, Chinese humanoid robot shipments topped forty thousand units, which the figures suggest is around ninety-seven percent of the global total. That is not a trend, that is dominance. It is driven by a complete domestic supply chain, aggressive manufacturing scale and sharp price competitiveness, with AI increasingly baked into the control systems.

Samantha: For European industry, this is a familiar story playing out in a new category. We saw it with solar panels, with batteries, and with electric vehicles. Now it is robots walking into factories, warehouses and, soon, service environments. European robotics has deep strengths, particularly in precision industrial arms from companies in Germany, Switzerland and Italy, but the humanoid wave is a different race. The volumes being shipped from China create a cost curve that is very hard to match without equivalent scale. That shapes everything from factory automation strategies to port logistics.

Bob: There is a procurement question for European manufacturers too. Buying Chinese humanoids at scale means importing not just hardware but also onboard AI, update cycles and data flows. Expect scrutiny from European regulators on where operational data is processed, especially for robots deployed in sensitive sites. The counter-move being discussed in industry circles is an EU robotics push comparable to the Chips Act. The numbers in these shipment reports will likely sharpen that conversation.

Samantha: And it is not only industrial. The same supply chain is enabling consumer humanoids priced for homes and small businesses, which brings us neatly to our next story. The key takeaway: China has moved from catching up in robotics to setting the pace, and Europe's response window is narrowing.

AiMOGA opens global pre-orders for Mobo, a family companion humanoid robot aimed at homes.

Bob: Right on cue, AiMOGA Robotics is unveiling Mobo on the twenty-first of October in Wuhu, and global pre-orders start the same day. Mobo is pitched as a consumer-grade family companion humanoid, designed to help with communication, companionship and daily routines across generations. It uses voice interaction, visual object recognition, schedule reminders and long-term memory. Under the hood sits AiMOGA's own MoAgent, which the company says links perception, interaction and execution so conversations feel more natural and context-aware.

Samantha: The company is a robotics spin-off that explicitly builds on automotive engineering experience, and Mobo is their first move into the home. For European consumers curious about trying one, the practical details to watch at launch will be price, available languages, data handling and whether it ships with full EU compliance including CE marking and clear privacy documentation. A robot with cameras, microphones and long-term memory sitting in your living room is not a neutral device, so terms of service will matter as much as hardware specs.

Bob: Analysts following the sector point out that family humanoids are still an unproven category. Previous attempts from well-known names stalled on battery life, limited usefulness and awkward pricing. What is different now is cheaper actuators, faster on-device AI and the broader Chinese supply chain we just discussed. European buyers get earlier access to this generation of hardware, but also the first wave of unresolved questions around safety standards, child interaction and liability when something goes wrong.

Samantha: The bottom line for listeners: another indicator that humanoids are moving from trade show stage to actual order pages, and Europe will be an early test market whether regulators are ready or not.

Two practical picks for the weekend, the free e-book tool Sigil and the map-fixing app StreetComplete.

Bob: To close out, two things you can actually use today. First up, Sigil. It is a free, open-source editor for e-books in the common EPUB format, available for Windows, Mac and Linux. If you have ever downloaded an e-book that renders badly on your reader, or want to tidy up a long document you converted yourself, Sigil lets you fix the chapters, the table of contents, the cover image and the metadata directly. Reviewers in self-publishing communities describe it as the standard tool for cleaning up EPUB files without needing a paid design suite.

Samantha: It is a good match for anyone building a personal library from older PDFs, public domain books from sources like Project Gutenberg, or notes they want to read on a Kindle or Kobo. The learning curve is modest if you have ever touched basic web code, and the online documentation is solid. Nothing goes to a cloud, the files stay on your own machine, which is the point. Expect to spend maybe twenty minutes on your first clean-up before it starts to click.

Bob: Second pick, and this one is non-AI and genuinely fun. StreetComplete. It is a free Android app from the OpenStreetMap community that turns improving the world map into a kind of quiz. You walk around your neighbourhood, the app shows you simple questions pinned to real locations: what is the speed limit on this street, does this shop have wheelchair access, what are the opening hours of that bakery. You answer, and the data flows straight into OpenStreetMap, which powers countless European navigation and cycling apps.

Samantha: It is a very satisfying way to spend a Sunday walk, and it directly improves the maps used by apps across Europe, from hiking tools to delivery routing. No account linked to advertising, no tracking profile being sold on. Users report it also makes you notice your own neighbourhood in a completely different way. Install it this afternoon, pick one street, and you will see what we mean. Today we covered: Google's quantum error-correction breakthrough, China's immediate AI export registration, Meta's Muse Spark Open release, China's dominance in robotics exports, the AiMOGA Mobo family humanoid, and two things to try yourself: Sigil for cleaning up e-books, and StreetComplete for improving the map under your feet.

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.