Quantum Encryption, Yann LeCun AI, South Korea Semiconductors & Data Center Supply Chain

Samantha: Welcome to The State of Tech — The European Edition. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: a quantum computing breakthrough that could crack encryption way sooner than anyone expected, Yann LeCun calls large language models "complete BS" and raises a billion dollars to prove it, South Korea's semiconductor exports smash all records on the back of AI demand, a Chinese AI machine that sorts textiles for recycling at incredible speed, US data center construction hitting a wall because of Chinese electrical parts, and Hong Kong rolls out over a hundred robots at its brand new RoboPark. Let's start with quantum.

A Caltech breakthrough could slash the hardware needed to break modern encryption by a factor of 100.

Samantha: So Bob, this one genuinely made me sit up. Researchers at Caltech have published a theoretical design that could dramatically reduce how many qubits you need to crack the encryption that basically secures the entire internet. We're talking about going from millions of qubits down to maybe ten to twenty thousand.

Bob: And that's the number that changes everything. Because up until now, the security community has been saying, look, quantum computers that can break RSA or elliptic curve cryptography — that's decades away. You'd need millions of stable qubits. Nobody's close. But this new approach uses neutral-atom qubits combined with a much more efficient form of quantum error correction. If the theory holds up in practice, you could run Shor's algorithm — the one that factors large numbers and breaks encryption — on hardware that's within reach in the next few years, not the next few decades.

Samantha: And just to be clear, this isn't some fringe paper. This is Caltech. The implications are enormous. Think about what's encrypted today — banking, government communications, medical records, Bitcoin. All of it relies on the assumption that factoring large numbers is computationally impractical.

Bob: Bitcoin is an interesting case because there's no central authority that can just push an update. If quantum computers reach this threshold, the entire blockchain security model is in question. But honestly, internet communications and government infrastructure worry me more in the short term.

Samantha: And the thing is, even if this takes five or seven years to become practical, data that's encrypted today can be harvested now and decrypted later. That "harvest now, decrypt later" attack is already a known threat.

Bob: Which is exactly why people have been pushing for post-quantum cryptography standards. NIST finalized several last year.

Samantha: For Europe specifically, this is a wake-up call. The EU has been working on its quantum strategy, there's funding through the Quantum Flagship programme, but adoption of post-quantum standards in actual infrastructure — banks, telecoms, government systems — that's still moving slowly. This research suggests the window for a comfortable transition just got a lot smaller.

Bob: And European companies handling sensitive data across borders need to be especially proactive. If your encryption can be broken within a decade, your migration plan needs to start now, not after the next regulatory review.

---

One of the godfathers of AI says LLMs are a dead end — and he's got a billion dollars to back that claim.

Bob: Alright, next up — Yann LeCun is making waves again. The Turing Award winner, Meta's chief AI scientist, has come out swinging against large language models, calling the massive investments in them — and I'm quoting here — "complete BS."

Samantha: Which is a bold thing to say when your employer has poured tens of billions into exactly that. But LeCun has been consistent on this. He's argued for years that predicting the next word in a sequence is fundamentally not how you get to human-level intelligence. His alternative is what he calls "world models" — systems that learn abstract representations of how the world works from images, video, audio, all kinds of sensory data. Not just text.

Bob: And now he's put serious money behind that conviction. His new venture, AMI Labs, has secured over a billion dollars in funding. The idea is to build AI that can genuinely understand physical reality — predict what happens when you push a ball off a table, understand cause and effect, plan actions in the real world.

Samantha: It's a fundamentally different bet. LLMs are incredible at language tasks, clearly. But LeCun's argument is that language is just a thin surface layer of human cognition. Most of what we know about the world, we learned by interacting with it, not by reading about it.

Bob: Now, whether he's right — that's the trillion-dollar question. But the fact that investors are willing to put a billion dollars behind this vision tells you there's genuine appetite for diversification in AI research.

Samantha: And for Europe, this is actually encouraging. European AI research has traditionally been strong in areas like robotics, computer vision, and embodied intelligence — exactly the domains where world models would matter most. If the field shifts even partially in LeCun's direction, European labs and startups could find themselves better positioned than they are in the current LLM race, which is dominated by American hyperscalers with massive compute budgets.

Bob: It also aligns well with EU priorities around diverse and responsible AI. If regulators have been uncomfortable with the concentration of AI progress in a single paradigm controlled by a handful of companies, this kind of diversification is exactly what they'd want to see.

---

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.

---

South Korea just posted the biggest export month in its history — and semiconductors are the reason.

Bob: South Korea's March export figures came in and they are staggering. Total exports hit eighty-six point one billion dollars — an all-time monthly record — with semiconductor shipments alone accounting for nearly thirty-three billion of that. That's a hundred and fifty-one percent increase year on year for chips.

Samantha: A hundred and fifty-one percent. Let that sink in. Samsung and SK Hynix are essentially printing money right now because the world cannot get enough memory chips for AI servers and data centers.

Bob: The Korean trade ministry is pointing directly at AI as the driver. Global investment in AI infrastructure — the servers, the training clusters, the inference hardware — all of that requires enormous amounts of high-bandwidth memory. And Korean manufacturers dominate that market. HBM chips, which are critical for AI accelerators, are in such demand that prices have stayed elevated even as production ramps up.

Samantha: This isn't just a Korean success story, though. It reveals how dependent the entire global AI buildout is on a very concentrated supply chain. If you're building AI infrastructure anywhere in the world, you're almost certainly buying Korean memory chips.

Bob: And that's the uncomfortable truth for Europe. The European Chips Act is ambitious — billions in subsidies to build fabs on European soil. But none of those planned facilities are targeting the high-bandwidth memory segment where Korea dominates. Europe is focusing on logic chips and mature-node production, which matters, but it doesn't address the specific bottleneck that AI is creating right now.

Samantha: So European companies building AI data centers are going to keep paying premium prices and competing for allocation with everyone else in the world. That's a strategic vulnerability that won't be resolved by the Chips Act alone. Diversification of supplier relationships and possibly joint ventures with Korean firms might be more practical in the near term.

---

An AI-powered sorting machine in China is transforming how the world could handle textile waste.

Samantha: Let's talk about something different now. A Chinese company called DataBeyond has launched a machine called the Fastsort-Textile. It uses AI to sort used clothing by material composition at high speed. And at a facility in Zhangjiagang, it's already cut the proportion of unrecyclable textiles from fifty percent down to thirty.

Bob: That's a twenty percentage point improvement, which in waste management is massive. The problem with textile recycling has always been sorting. You've got cotton, polyester, nylon, blends — all mixed together. And roughly seventy percent of global textile production is synthetic. To recycle any of it effectively, you need to know exactly what it's made of, and doing that by hand is slow and expensive.

Samantha: This machine uses spectral analysis combined with AI classification to identify fabrics at speed. It's the kind of application where AI genuinely shines — pattern recognition on messy, real-world data at a scale humans simply can't match.

Bob: The fashion industry has been under pressure for years over its environmental footprint. Fast fashion generates mountains of waste, and recycling rates are abysmal — under one percent of clothing gets recycled into new clothing globally. If machines like this can dramatically improve sorting accuracy and speed, you unlock the economics of textile recycling in a way that wasn't possible before.

Samantha: Europe is actually ahead of many regions on textile waste regulation. The EU's Strategy for Sustainable Textiles is pushing for mandatory extended producer responsibility, and several member states are already implementing collection requirements. But collection without effective sorting is just moving the problem around. This kind of AI-powered sorting technology could be exactly what European recyclers need to make the economics work. I'd expect to see serious interest from European waste management companies in either licensing this technology or developing competing solutions.

---

Almost half of planned US data centers could face delays this year — because of a dependence on Chinese electrical equipment.

Bob: Alright, another story that reveals just how fragile the AI infrastructure buildout really is. Reports are coming in that nearly half of the data centers planned for construction in the US this year could face significant delays or outright cancellations. The reason? A critical shortage of electrical equipment — transformers, switchgear, batteries — much of which is imported from China.

Samantha: This is ironic on multiple levels. The US is trying to reduce technological dependence on China, pouring billions into domestic AI capabilities, but the physical infrastructure to power those AI systems relies on Chinese-manufactured components that nobody else is producing in sufficient quantities.

Bob: A massive facility planned for OpenAI in Texas is among those affected. The lead times for large power transformers have stretched to two or three years in some cases. Domestic US manufacturers simply don't have the capacity, and tariff uncertainties are making the import situation even more complicated.

Samantha: This is a bottleneck that almost nobody was talking about two years ago. Everyone focused on GPU supply, on chip fabrication. But you can have all the GPUs in the world — if you can't power the building they sit in, you've got nothing.

Bob: And the lesson for Europe is staring us right in the face. European data center construction is accelerating too — Dublin, Amsterdam, Frankfurt, the Nordics. If the same electrical components are in short supply globally, Europe will hit the same wall. The smart move is to start mapping these dependencies now, investing in domestic manufacturing of power infrastructure, and locking in supply contracts before the crunch gets worse. We've been warned.

---

Hong Kong launches RoboPark with over a hundred robots, signaling the physical AI era is here.

Samantha: And finally, Hong Kong has just wrapped up its InnoEX 2026 fair with a brand new showcase called RoboPark. Over a hundred robots on display — healthcare robots, industrial automation, entertainment bots. And notably, four of the world's top five humanoid robot makers were there.

Bob: This is Hong Kong making a very deliberate statement about positioning itself as a global hub for physical AI and robotics. The emphasis wasn't just on cute demo robots — although there were plenty of those. It was on commercial-ready systems. Robots that can work in warehouses, assist in surgeries, handle logistics. The event also featured what they're calling the "low-altitude economy" — drones and autonomous aerial vehicles for delivery and inspection.

Samantha: The humanoid robot segment is genuinely heating up. Companies from China, Japan, the US — they're all racing to get general-purpose humanoid robots into commercial deployment. And the pace of improvement, largely driven by AI advances in perception and manipulation, has been remarkable over the past eighteen months.

Bob: What's changed is the AI layer. The hardware for robots has been decent for years. What was missing was the intelligence to operate in unstructured environments. With better vision models, better reinforcement learning, better language interfaces — robots are finally becoming useful outside of highly controlled factory settings.

Samantha: Europe has an incredible legacy in robotics — think KUKA, ABB, companies that built the backbone of industrial automation. But the new wave of AI-native robotics is moving fast, and a lot of the energy and investment is in Asia and the US. Events like RoboPark are a signal. If European robotics companies don't integrate the latest AI capabilities quickly, they risk losing ground in what could be the most important hardware market of the next decade.

Bob: And it's not just about competition. Europe's aging population means demand for service robots — in eldercare, healthcare, logistics — is going to surge. Getting this right is an economic and social imperative.

---

Samantha: Alright, that's our show. Today we covered: a quantum breakthrough threatening encryption timelines, LeCun's billion-dollar bet against LLMs, South Korea's record-smashing chip exports, AI-powered textile recycling in China, US data center delays from Chinese parts shortages, and Hong Kong's new RoboPark showcase.

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.