AMD $10B Taiwan Packaging Bet: AI Chip Supply Shifts

Samantha: Welcome to The State of Tech, The European Edition, Saturday May twenty-three, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: AMD pours ten billion dollars into Taiwan's chip ecosystem, China hands every humanoid robot an ID card, OpenAI cracks a famous maths puzzle, the US and Sweden sign a tech pact, then to close out, two things you can actually use today, a lunar robotics play that opens doors for European space firms, and a Korean parts maker chasing the Tesla Optimus market. Let's start with the biggest cheque of the day.

AMD commits ten billion dollars to Taiwan's chip ecosystem to expand AI packaging capacity.

Samantha: AMD is putting ten billion dollars into Taiwan's microchip ecosystem over the next three years. The money is aimed at advanced packaging, which is the bottleneck right now for anyone trying to build AI hardware at scale. Demand is running way ahead of supply, and AMD wants to lock in capacity.

Bob: And there's a really interesting twist here. AMD is backing a different packaging technology called elevated fan-out bridge, or EFB. Think of packaging as the way you stitch a processor and memory chips together so they behave like one super-fast brain. Up to now, almost everyone has had to queue at TSMC's door for their CoWoS technology, and there just isn't enough of it.

Samantha: So this is AMD saying, we want a second lane on the motorway. They're funding an alternative ecosystem in Taiwan, which gives them more chips to sell, sooner.

Bob: Exactly the kind of move that doesn't make headlines outside the tech press, but quietly reshapes the AI race. If you can't ship the hardware, you can't sell the cloud capacity, and you can't train the models. NVIDIA has been the runaway leader partly because it had better access to this packaging. AMD is trying to close that gap with cash.

Samantha: And it puts Taiwan even deeper at the heart of the global supply chain. Which, depending on how you read the geopolitics, is either reassuring or nerve-wracking.

Bob: For European buyers, this matters in a very practical way. European cloud providers, carmakers building AI systems, research labs, they all need these advanced chips. More packaging capacity means shorter waiting lists and, in theory, a little less price pressure. We've heard for two years now that AI chips are sold out through next year. Investments like this are how that finally starts to ease.

Samantha: There's also a strategic angle for Brussels. The EU has its own Chips Act, with big ambitions to bring more semiconductor work into Europe. But the very top of the food chain, this advanced packaging step, is still concentrated in Taiwan. Today's announcement underlines how hard it will be to shift that quickly.

Bob: And it's a reminder that the AI boom isn't just software. It's factories, clean rooms, and ten-billion-dollar bets on very specific machines.

China gives every humanoid robot an ID card and full lifecycle tracking.

Bob: China is rolling out something genuinely new. Every humanoid robot built in the country will get a unique ID card, and the state will track it from the factory floor all the way to the scrapyard. The industry ministry announced it, and there's already a platform live, covering more than a hundred companies and twenty-eight thousand robots across two hundred models.

Samantha: So it's a bit like a vehicle registration system, but for machines that walk around on two legs. Every unit traceable, every owner on file.

Bob: That's the shape of it. The official reason is safety and control. As humanoid robots start showing up in factories, warehouses, and eventually homes, Beijing wants to know who built them, who owns them, and what they're doing. It's also a soft form of industrial policy, because to get an ID you have to play by the standards Beijing sets.

Samantha: Which gives China a head start in writing the international rulebook. If the Chinese system is the only one that exists in practice, other countries end up copying bits of it whether they like it or not.

Bob: And there's a competitive edge too. Chinese manufacturers can point at this register and say, our robots are certified, tracked, accountable. That's a sales pitch in markets that are nervous about safety.

Samantha: For Europe, this lands right as the AI Act is being implemented and as conversations about robotics-specific rules are heating up. Brussels likes to set global standards, the so-called Brussels effect. But on humanoids, China just moved first with something concrete and operational.

Bob: And European robot makers exporting to China will have to comply with this ID system if they want market access. That's the same playbook Europe uses for its own rules, now applied in reverse.

Samantha: A real shift in who sets the standards for the next big category of hardware.

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.

OpenAI's reasoning model disproves a long-standing mathematical conjecture from Paul Erdős.

Bob: An OpenAI reasoning model has cracked open a genuinely hard maths problem. It disproved a central conjecture tied to Paul Erdős's planar unit distance problem, which is a famous puzzle in a corner of mathematics called discrete geometry. External mathematicians have checked the work, and one called it a milestone in AI maths.

Samantha: For people who aren't deep in this, the Erdős problem has been sitting there frustrating researchers for decades. It's the kind of question that looks deceptively simple, but generations of mathematicians have chipped away at it without fully resolving it.

Bob: And what's striking is the role the AI played. It didn't just check existing proofs, it produced a novel result, something the human researchers then verified. That's the collaborator role a lot of people have been predicting, and it's now showing up in published work.

Samantha: It's also a different flavour of AI win. We're used to hearing about chatbots writing emails or generating images. This is AI doing something at the frontier of human knowledge, in a field where you can't fake your way through, the proof either holds or it doesn't.

Bob: The bottom line is that AI is moving from impressive demo into actual scientific contributor. If models can disprove conjectures in pure maths, they can probably help in adjacent fields too, physics, chemistry, cryptography.

Samantha: Europe has a strong tradition in mathematics, from the Max Planck institutes in Germany to research centres in Paris, Cambridge and Amsterdam. This kind of breakthrough lands directly in their world. Some will see a new tool that accelerates their work, others will worry about what it means for the value of the human proof itself.

Bob: And there's a funding angle. European research budgets are tight, and if AI assistants can speed up theoretical work, that argument will be made loudly in Brussels and in national capitals.

Samantha: A small story in volume, big in implication.

The US and Sweden sign a Tech Prosperity Pact covering AI and next-generation mobile networks.

Bob: Washington and Stockholm have signed what they're calling a Tech Prosperity Pact. It deepens cooperation on AI, on 5G and the next step, 6G, on cloud, electronics, and security. There's a focus on AI for manufacturing and industrial automation, and on what they call critical infrastructure and defence use cases.

Samantha: Sweden is an interesting partner here. It's home to Ericsson, one of the few Western telecom equipment giants, and it's an EU member with a serious industrial base. So this isn't a symbolic handshake, it's two countries with real assets agreeing to align.

Bob: And the timing matters. The transatlantic relationship has been wobbly on tech, with disputes over data flows, AI rules, and trade. A bilateral pact like this is a way of locking in cooperation outside of the bigger, slower EU–US negotiations.

Samantha: It also tells you where the action is moving. Five years ago, the headline would have been 5G. Now we're already talking about 6G research, joint AI projects in factories, and security frameworks for both. The race for the next mobile standard has quietly started.

Bob: For the rest of Europe, this is a mixed signal. On one hand, an EU member strengthening ties with the US can pull European industry closer to American standards and platforms. On the other, Brussels prefers EU-wide deals, not member states going solo with Washington.

Samantha: Expect this to come up in EU corridors. Other capitals will be watching whether to follow Sweden's lead with their own bilateral pacts, or to push for a single European line.

Bob: And for Ericsson and its competitors, including Finland's Nokia, this opens specific doors in US infrastructure projects that have been hard to access. That's the most concrete commercial outcome of a deal like this.

Samantha: To close out, two things from outside the usual headline factory, both about the next wave of robotics, one literally on the moon, the other right here on Earth.

Colorado startup Lunar Outpost raises thirty million dollars to build the first permanent structures on the moon with robots.

Bob: Lunar Outpost, a robotics company based in Colorado, has raised thirty million dollars in fresh funding. Their plan is genuinely ambitious. They want to send robots to the moon to build the first permanent structures up there. A new lunar rover called Pegasus is due to be ready by the end of 2027, and the launch is pencilled in for 2028 alongside NASA's Artemis 4 crewed mission.

Samantha: So this isn't a flag-and-footprint mission. It's the slow, unglamorous work of preparing a site, before astronauts arrive long-term. Robots laying foundations, essentially.

Bob: That's the picture. Think of it as construction crews going in before the people. Excavating, levelling, possibly handling lunar dust, which is one of the nastiest engineering challenges out there because it's abrasive and gets into everything.

Samantha: It's interesting that this is now venture-fundable. Five years ago, lunar infrastructure was almost entirely the domain of national agencies. Now you have private companies raising rounds to do it commercially, banking on NASA contracts and eventually paying customers further out.

Bob: And the European angle is real, because the European Space Agency is involved in Artemis. ESA contributes parts of the Orion spacecraft and the Gateway lunar station. So when American robotics startups start showing up around Artemis missions, that's also future business for European partners and suppliers.

Samantha: For listeners who want to follow this, NASA's Artemis programme has its own public site, and Lunar Outpost regularly updates on their progress. The Pegasus rover is the one to watch as the 2028 launch window approaches.

Bob: It's the kind of story you can dip into without subscribing to anything or paying anything. Just bookmark the Artemis updates and check in every few months. You'll see the building blocks of permanent human presence on the moon being shipped, one rover at a time.

Samantha: And it gives a sense of just how fast the space economy is professionalising. Robots on the moon used to sound like science fiction. Now it's a Series B pitch deck.

South Korea's HL Mando aims to supply actuators for Tesla's Optimus humanoid robot.

Bob: HL Mando, a South Korean auto parts manufacturer, is expanding into humanoid robot components. They already make parts for Boston Dynamics' Spot, the four-legged robot you've probably seen videos of, and now they're targeting the actuator market for humanoids like Tesla's Optimus.

Samantha: Actuators are basically the muscles of a robot. They're the motors and joints that let it move. So if you're making a humanoid that walks, lifts, and balances, the actuator is one of the most critical and expensive parts.

Bob: And HL Mando is betting on a clear trend. Tesla has been loud about Optimus, and the fourth generation is the one they're aiming at. The company is also expanding manufacturing in North America to be close to that customer base.

Samantha: What makes this useful for our listeners is the broader picture. Humanoid robots are starting to look less like science projects and more like products with real supply chains. When established auto parts makers start retooling for this market, you know it's getting serious.

Bob: For anyone curious to see this in action, Boston Dynamics has a YouTube channel where you can watch Spot and their humanoid Atlas in regular updates. It's free, it's fun, and it gives you a sense of where the technology actually is today, beyond the press releases.

Samantha: And there's a European echo here. Companies like Germany's Neura Robotics and others are working on similar humanoids. They'll be watching closely to see which suppliers emerge as the standard providers.

Bob: The takeaway, if you've ever wondered when humanoid robots become real, watch the parts suppliers. That's where the early signals show up.

Samantha: Today we covered: AMD's ten-billion-dollar bet on Taiwan, China's ID cards for humanoid robots, OpenAI cracking an Erdős conjecture, the US–Sweden tech pact, Lunar Outpost's moon-building robots, and HL Mando chasing the Optimus supply chain.

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.