TSMC $100B US chip expansion & China's AI bloc

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

Bob: And I'm Bob Russell. Today: TSMC pours another hundred billion dollars into American chip fabs, China rallies twenty-nine nations behind a new AI cooperation body, Japan builds the world's first national physical-AI factory with NVIDIA, Apple Intelligence finally clears the Chinese regulator by leaning on Alibaba, and to close out, two things you can actually use today: the writing app Ghostwriter and the offline maps app OsmAnd. Let's start with Taiwan.

TSMC pledges another hundred billion dollars for American AI chip fabs as agentic AI demand explodes.

Samantha: TSMC, the Taiwanese company that makes most of the world's advanced chips, is putting another hundred billion dollars into the United States. That takes its total US commitment to two hundred sixty-five billion, and it's raising its full-year capital spending to more than sixty-four billion. The money will pay for four more advanced wafer fabs and chip packaging plants at its Arizona site, all built on two-nanometer technology or better.

Bob: And the driver here is very specific. TSMC is pointing at what it calls agentic AI, the wave of AI systems that don't just answer questions but actually take actions, book things, run software, chain tasks together. Those systems need enormous amounts of custom silicon, and Nvidia, Apple, and AMD all rely on TSMC to make it. So when TSMC talks about capacity shortages, it's really saying the demand curve for AI hardware is still bending upward, not flattening.

Samantha: The political layer is impossible to ignore. Washington has spent years pushing to bring leading-edge chip manufacturing onto American soil, and this is the most concrete answer yet. Arizona is quietly becoming a second home for the most advanced semiconductor production on the planet. What used to sit almost entirely on one island is now being duplicated, at massive expense, in the American desert.

Bob: For Europe the comparison is uncomfortable. The European Chips Act aimed to double the continent's share of global chip production by 2030, but the numbers being announced in Arizona now dwarf anything under construction in Dresden, Magdeburg, or the Netherlands. TSMC's European plant is a joint venture focused on more mature chips, not the cutting-edge two-nanometer work heading to Arizona. Industry groups have been arguing for months that Europe needs a step change in ambition if it wants a real seat at the AI hardware table.

Samantha: There's also a customer question here. European carmakers, industrial firms, and telecom operators all want access to the newest AI accelerators. If the leading edge is concentrated between Taiwan and Arizona, European buyers effectively depend on decisions made in Taipei and Washington. The bottom line: the AI boom is being wired into physical geography, and Europe is not yet on that map at the top level.

China launches the World AI Cooperation Organization with twenty-nine nations, staking a claim on global AI governance.

Bob: Twenty-nine countries have formally signed on to a new body called the World Artificial Intelligence Cooperation Organization, led by China. The idea is to institutionalise Beijing's influence over how AI is regulated, shared, and deployed, especially in developing economies. The pitch to those countries is sovereign AI on affordable terms: low-cost, open-source models, cheaper hardware, and shared infrastructure that doesn't depend on American export licences.

Samantha: And to make the point about independence, Huawei showed off a new AI computing cluster called the Atlas 950 SuperPoD, designed to work without high-end American semiconductors. That's the message underneath the diplomacy, that China can build serious AI infrastructure on its own stack. For countries in Southeast Asia, Africa, and parts of Latin America that have been struggling to get access to Nvidia chips, this is a real alternative, not just a slogan.

Bob: Compare that with the American approach, which has been about export controls, alliances, and keeping the most powerful chips inside a trusted circle. Now there are effectively two competing offers on the table. One says come with us and get access to the best hardware, but accept our restrictions. The other says come with us and get workable hardware plus fewer conditions. For a lot of governments, that second offer is genuinely attractive.

Samantha: For Europe this is uncomfortable ground. Brussels has spent years trying to position the AI Act as a global template for trustworthy AI, a kind of third way between American markets and Chinese state control. If twenty-nine countries are already signing up to a Beijing-led framework, that European ambition gets harder. European companies operating in those markets may find themselves navigating three different rulebooks at once, and diplomats will have to work harder to keep the EU model in the conversation.

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.

Japan and NVIDIA build the world's first national infrastructure for physical AI, from robots to digital twins.

Bob: Japan is doing something no other country has tried at this scale. Together with NVIDIA and a company called Noetra, backed by Japan's trade ministry, it's building what they're calling the world's first national AI infrastructure specifically for physical AI. That means AI that controls things in the real world, robots on factory floors, self-driving vehicles, digital twins of ports and power plants, industrial machines that learn.

Samantha: The scale is striking. The AI factory will run on thirteen thousand seven hundred fifty of NVIDIA's newest processors called Vera, paired with twenty-seven thousand five hundred of its next-generation Rubin graphics chips. The data centre itself is rated at a hundred forty megawatts, which is enormous, roughly the electricity draw of a small city. And the whole thing is aimed at building open, multimodal AI models that Japanese industry can plug into.

Bob: This is a very different strategy from most national AI plans, which tend to focus on chatbots and language models. Japan is betting that the real economic value of AI will come from moving atoms, not just words. Given that Japan's strengths are cars, robotics, and precision manufacturing, that's a logical bet. It's also a way of keeping domestic industry competitive against Chinese rivals that are pouring money into the same areas.

Samantha: The lesson for Europe is fairly direct. European manufacturers face the same demographic pressures as Japan, ageing populations, shrinking workforces, high labour costs. If Japanese carmakers and factories get access to shared physical-AI infrastructure at national scale, that changes the competitive playing field. German industrial groups have been calling for something similar, but nothing on this scale has been approved in Brussels or Berlin yet. A separate development from Mitsubishi Heavy Industries this week, shipping a large industrial cooling unit designed for AI data centres to the United States, hints at how quickly this hardware race is spreading.

Apple Intelligence finally clears China, powered by Alibaba's Qwen models instead of OpenAI.

Bob: Apple has cleared a hurdle it's been stuck on for more than a year. China's Cyberspace Administration has approved Apple Intelligence, the AI features built into iPhones, for use in mainland China. And critically, those features will not be running on OpenAI's models, the way they do in Europe and the United States. In China they'll run on Alibaba's Qwen models, with Baidu possibly involved for some functions.

Samantha: This is a genuinely important moment because it makes explicit something the industry has been dancing around: the AI world is splitting into two technology stacks. If you sell phones or software in China, you use Chinese AI models trained by Chinese companies. If you sell in the West, you use Western ones. Apple has essentially agreed to that split rather than lose access to hundreds of millions of Chinese customers.

Bob: For Chinese iPhone users this means the AI features will finally work, image generation, writing help, smart summaries. For Apple it means keeping a market that has been under pressure from Huawei and Xiaomi. And for Alibaba it's a significant validation, having the world's most valuable consumer electronics company running on your AI models is a serious credential.

Samantha: For Europe there's a subtler point buried in this deal. It shows that even the biggest global tech companies will bend to local AI rules if the market matters enough. That's an argument European regulators have been making about the AI Act: set high standards, and companies will comply because Europe is too big to ignore. Apple's Beijing deal proves that logic can work. It also means European buyers may end up with a third variant of Apple Intelligence over time, one shaped by European privacy and AI rules.

Ghostwriter turns any laptop into a calm, distraction-free writing space for long-form work.

Bob: Right, to close out, two things you can actually use today. The first is a writing app called Ghostwriter. It's a free, open-source, markdown-based writing environment that strips your screen down to just the words. No formatting toolbars, no notifications, no side panels. It's available for Windows and Linux, and there are similar builds for Mac users.

Samantha: The appeal is straightforward. Most modern writing tools, from Word to Google Docs to Notion, are cluttered with buttons, comments, and AI suggestions popping up as you type. Ghostwriter goes the opposite way. It gives you a dark screen, a blinking cursor, and a live word count. It uses markdown, which is a very simple way of marking headings and lists with plain text, so your document stays readable in any editor for decades to come.

Bob: Users report that the focus mode is what makes it stick. It dims everything except the sentence you're currently writing, which sounds gimmicky until you try it during a long piece of writing. Journalists, novelists, and students have picked it up for exactly that reason. And because it's open source, nothing you type gets sent anywhere, no cloud upload, no training data harvesting.

Samantha: It's also genuinely lightweight. It will run happily on an old laptop that struggles with Word. You can export directly to PDF or HTML, which covers most needs. For anyone drafting a thesis, a business proposal, or a long report over the next few weeks, it's worth twenty minutes to install and try. Fair warning, if you're wedded to real-time collaboration and live comments, this isn't that tool. It's built for solo, deep work.

Bob: One practical tip from experienced users: pair it with a simple cloud sync folder so your drafts back up automatically. That way you get the calm interface without worrying about losing your work if your laptop dies. Free, no account required, runs entirely on your own machine.

OsmAnd gives you offline maps and turn-by-turn navigation across Europe with no tracking.

Samantha: The second thing to try is a mapping app called OsmAnd. It uses the same underlying data as OpenStreetMap, which is the open, community-built alternative to Google Maps. You download entire countries or regions to your phone in advance, and then everything works offline, driving directions, walking routes, cycling, hiking trails.

Bob: This is very useful for European travel, where you cross borders constantly and mobile data can be patchy in mountains, rural areas, or on trains through tunnels. Once you've downloaded, say, France, Italy, and Switzerland, the app doesn't need any connection at all. It's available on both iPhone and Android, and there's a generous free version with paid tiers if you want unlimited downloads.

Samantha: Where OsmAnd really shines compared to Google Maps is in cycling and hiking. It shows elevation profiles, trail difficulty, mountain huts, drinking water fountains, all the details that community mappers have added over the years. Reviewers regularly rate it above the mainstream apps for anything off the main road. And because you're not streaming map tiles, your battery lasts noticeably longer on a long day out.

Bob: The privacy point matters too. OsmAnd doesn't build a profile of everywhere you've been. There's no advertising engine watching which restaurants you searched or which hotels you looked at. For anyone uncomfortable with how much location data the big platforms collect, that alone is a reason to try it.

Samantha: The learning curve is a bit steeper than Google Maps, the interface is packed with options, but once you've set up your favourite map style and downloaded your regions, it becomes second nature. A good weekend project before your next trip.

Samantha: Today we covered: TSMC's hundred-billion-dollar American chip expansion, China's new twenty-nine-nation AI cooperation body, Japan's national physical-AI factory with NVIDIA, Apple Intelligence launching in China on Alibaba's models, and two things to try yourself: the writing app Ghostwriter and the offline maps app OsmAnd.

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.