Samsung 48x Chip Profit Surge & Memory Crunch 2027

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

Bob: And I'm Bob Russell. Today: AI moves deep into the life sciences, with the Chan Zuckerberg Biohub launching a half-billion-dollar virtual biology project and Mayo Clinic detecting pancreatic cancer years before diagnosis. Then: Nvidia's push to own the entire AI stack with Nemotron 3 Nano Omni. Samsung's chip division reports a forty-eight-fold profit surge and warns of a memory crunch into 2027. The US FCC bans Chinese labs from testing American electronics. And we close with a Sony robot arm beating human pros at table tennis. Let's start with the biggest story of the day, AI in biology.

Chan Zuckerberg Biohub launches Virtual Biology Initiative with five hundred million dollars to simulate life with AI.

Samantha: This is a big one. The Chan Zuckerberg Initiative, run by Mark Zuckerberg and Priscilla Chan, is putting five hundred million dollars into a new program called the Virtual Biology Initiative. The goal is ambitious: build an AI-driven simulation platform that models virtual cells and entire biological systems. Predict disease, design drugs, and run experiments in silico, without always needing a wet lab.

Bob: And it expands the Biohub network across San Francisco, Chicago and New York. The pitch is that biology is too complex for humans to fully grasp, but maybe not too complex for large AI models. Train them on enough cellular data, and you get a kind of foundation model for life itself. Predict how a cell reacts to a drug, how a tumour evolves, how a protein folds, all before you touch a pipette.

Samantha: It fits a clear pattern. Isomorphic Labs, owned by Alphabet, is doing this for drug discovery. Recursion is doing it with phenotypic screening. And now CZI wants to build the underlying simulation layer. AI is moving from language and code straight into the life sciences, and the capital flowing in is enormous.

Bob: The interesting tension is scientific. Some biologists are sceptical that you can simulate a cell from first principles. A cell is messy, stochastic, full of feedback loops we barely understand. Others say we don't need perfect physics, we just need patterns that predict outcomes well enough to be useful. That debate will define how this money gets spent.

Samantha: And there's a second layer here, and it links straight into our next story. Mayo Clinic just published a landmark validation study where an AI model can flag pancreatic cancer up to three years before formal diagnosis, based on standard CT scans.

Bob: Pancreatic cancer is one of the deadliest, with a five-year survival rate of around twelve percent, mostly because it's caught late. The Mayo model picks up subtle changes in the pancreas that radiologists miss by eye. And critically, it works on scans originally taken for unrelated reasons, what they call opportunistic screening.

Samantha: So you go in for a kidney stone, the scan happens, and an AI quietly flags a pancreas that looks suspicious. Three years before symptoms. That's a genuine shift in what early detection means.

Bob: For Europe this is double-edged. The science is exciting, but the regulatory path is steep. The EU AI Act classifies clinical AI as high-risk. Add the Medical Device Regulation, and you get long approval cycles. European hospitals are eager, but the question is whether the validation data, the liability frameworks, and the reimbursement codes catch up in time.

Samantha: And on the CZI side, European biotech hubs in Cambridge, Munich, Leuven will watch closely. If the virtual cell platform becomes open or semi-open, European labs benefit. If it stays proprietary, it's another American AI moat in a strategic field.

Nvidia launches Nemotron 3 Nano Omni in a bid to own the entire enterprise AI stack.

Bob: Moving on. Nvidia is no longer just a chip company. With its new Nemotron 3 Nano Omni platform, it's making a serious play to own the full enterprise AI stack, from silicon up to the software businesses actually deploy.

Samantha: The logic is straightforward. Chips are a brutal cyclical business. Platforms generate recurring revenue. Amazon and Microsoft figured this out with cloud. Nvidia is essentially saying: we'll be the operating layer for enterprise AI, the place where companies build, deploy and run their models. Hardware plus models plus tooling plus services.

Bob: And data centre revenue is still the cash engine. But Nemotron 3 Nano Omni is meant to lock customers into Nvidia's world above the chip layer. Multimodal models, optimised inference, integration with enterprise data, the works. It's a defensive and offensive move at the same time.

Samantha: Defensive against whom? AMD, custom silicon from hyperscalers, and increasingly Chinese alternatives. If Nvidia's only moat is CUDA and chip performance, that erodes. If the moat becomes the entire stack, it's much harder to dislodge.

Bob: There's a competitive question for buyers too. Do you want one vendor controlling chips, models and deployment tools? Or do you want to mix and match? Some enterprises will love the simplicity. Others will see vendor lock-in and start hedging.

Samantha: For European companies this matters because sovereignty keeps coming up. If Nvidia becomes the default AI operating layer, European AI strategy becomes American AI strategy by default. Mistral, Aleph Alpha, Silo AI, the whole European model ecosystem, they need infrastructure partners that aren't all in one camp.

Bob: Brussels will watch this through a competition lens too. A vertically integrated AI giant raises the same questions cloud raised a decade ago. Expect scrutiny, eventually.

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.

Samsung's chip arm posts a forty-eight-fold profit jump and warns the memory crunch will worsen into 2027.

Bob: Samsung Electronics' semiconductor division just posted a forty-eight-fold increase in operating income last quarter. Record profit of fifty-three point seven trillion won, that's about thirty-six point two six billion US dollars. Driven almost entirely by AI data centre demand and high-bandwidth memory.

Samantha: And the warning attached is the headline within the headline. Samsung says the global memory chip shortage will get worse next year. Customers are already placing orders for 2027. Meta, Alphabet and the other hyperscalers are essentially booking capacity years ahead.

Bob: High-bandwidth memory is the bottleneck of the AI build-out. You can have all the GPUs you want, without HBM stacks feeding them, they sit idle. Samsung, SK hynix and Micron are the three names that matter, and pricing power has shifted hard in their favour.

Samantha: A forty-eight-fold profit jump is the kind of number that makes you double-check the decimal point. It tells you how concentrated the AI infrastructure spend has become, and how much of it ends up in memory.

Bob: It also signals risk. If AI capex slows, this profit profile reverses fast. Memory has always been cyclical. The bet right now is that demand stays structurally elevated because models keep growing.

Samantha: For Europe the practical effect is squeeze. European hyperscalers and sovereign cloud projects, OVHcloud, the Gaia-X-aligned providers, will pay more and wait longer. Telcos building edge AI, automotive companies running on-device models, all face the same supply constraint.

Bob: And the European Chips Act was always more about logic than memory. There's no real European HBM player. So this shortage exposes a strategic gap that no amount of subsidy can close quickly.

The US FCC unanimously votes to ban Chinese labs from testing electronics destined for America.

Samantha: The US Federal Communications Commission has unanimously approved a proposal to bar Chinese laboratories from testing and certifying electronic devices sold in the United States. Smartphones, cameras, computers, the lot. Roughly seventy-five percent of US electronics are currently tested in China.

Bob: New rules will streamline approval for devices tested in US labs or in reciprocal countries. The framing is integrity, security and reciprocity. The subtext is: another layer of decoupling, this time at the certification level.

Samantha: This is a quieter but very real escalation. Tariffs and export controls grab the headlines, but testing and certification is plumbing. Move that plumbing, and supply chains have to physically reorganise.

Bob: And it has knock-on costs. Setting up testing capacity outside China takes time and capital. Smaller manufacturers, especially consumer electronics brands with thin margins, will feel this in 2026 and 2027.

Samantha: It also raises a precedent question. If the US can route certification by political alignment, others will too. China can mirror the rule. The EU might be tempted in specific sectors. Suddenly compliance is fragmented along bloc lines.

Bob: For Europe, the options are uncomfortable. Align with Washington and accept higher costs. Stay neutral and risk losing US market access for European brands tested in mixed labs. Or invest seriously in domestic testing infrastructure, which is the slow expensive option.

Samantha: European notified bodies under the Radio Equipment Directive already do a lot of this work. The question is scale. If American demand pivots to reciprocal countries, Europe could either capture that work or watch it go to other partners.

Bob: It's the kind of policy story that doesn't trend on social media, but ends up shaping where factories and labs sit for the next decade.

OMODA and JAECOO secure one thousand orders for AI-powered intelligent police robots in smart cities.

Samantha: At the 2026 Chery International Business Summit, OMODA and JAECOO's AiMOGA Robotics unit launched full commercial deployment of its Intelligent Police Robots. One thousand intentional orders, one hundred units already delivered. They're aimed at urban roads, traffic hubs and public governance.

Bob: The robots are built on intelligent vehicle technology. Real-time traffic guidance, illegal parking identification, basically extending what's inside an autonomous car into a roaming public-space platform. Cameras, sensors, AI perception, on wheels.

Samantha: A thousand orders is significant because it's commercial, not pilot. Robotics in public space has been stuck at demo phase for years. This crosses into operational deployment, with Chinese cities clearly comfortable buying at scale.

Bob: And the framing matters. These are described as assisting frontline staff, not replacing officers. Traffic guidance, identifying violations, optimising flow. Whether that line holds in practice is another question.

Samantha: There's a real ethical layer. A roaming sensor platform with police branding is a surveillance device by design. Facial recognition, licence plate capture, behaviour analysis, the technical capability is all there even if the stated function is narrower.

Bob: For European cities the AI Act is the immediate filter. Real-time biometric identification in public spaces is heavily restricted. So a one-to-one import of this product is unlikely. But stripped-down versions, traffic-focused only, could find a market.

Samantha: And European robotics firms, in Germany, the Nordics, the Netherlands, will see this as a competitive signal. If Chinese vendors industrialise public-sector robotics first, European players risk being locked out of their own municipal markets later.

Bob: Municipal procurement in Europe is slow, fragmented, and privacy-conscious. That's a feature, not a bug, but it does mean watching others move first.

Sony's robot arm Ace beats human pros at table tennis using nine cameras and reinforcement learning.

Samantha: And to close, something genuinely fun. Sony has built a robotic arm called Ace that can challenge, and sometimes beat, elite human table tennis players. Nine cameras tracking ball spin, reinforcement learning driving the arm, and apparently very good footwork, metaphorically speaking.

Bob: Table tennis is a beautiful test case. The ball moves at insane speed, spin changes everything, and you have to predict trajectory in milliseconds. Reading spin from visual cues alone is hard even for humans. Nine cameras give the robot a perception advantage no player has.

Samantha: And reinforcement learning means it didn't get programmed move by move. It played, lost, adjusted, played again. Until it could read a topspin loop and counter it before the ball crossed the net.

Bob: Sony frames this as a milestone for machine agility in uncertain environments. Which is corporate speak for: the techniques behind this robot translate directly into manufacturing, logistics and any setting where machines need to react fast to messy reality.

Samantha: There's something charmingly human about the test, though. We measure machine intelligence by whether it can beat us at the games we love. Chess, Go, StarCraft, now table tennis.

Bob: European robotics, KUKA, ABB, the Swiss and Italian precision players, will look at the perception stack here, not the paddle. Nine-camera tracking with on-the-fly trajectory prediction is exactly what high-speed assembly lines and quality control need.

Samantha: And you can imagine the demo circuit. Trade fair in Hannover, robot arm playing visitors, queue around the block. It's marketing gold and a serious tech showcase at the same time.

Bob: I want to know its win rate against the actual world top hundred. That's the real benchmark.

Samantha: Today we covered: AI moving into virtual biology and early cancer detection, Nvidia's full-stack AI push, Samsung's record chip profits and the looming memory crunch, the FCC banning Chinese test labs, China's intelligent police robots going commercial, and Sony's table tennis robot beating the pros.

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.