Google Robotics, AGIBOT Embodied AI, xAI Speech-to-Text & Taiwan Chips

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

Bob: And I'm Bob Russell. Today: Google's new robotics brain for industrial machines, AGIBOT declaring this the year embodied AI goes mainstream, xAI launching a speech-to-text API to challenge the incumbents, Taiwan's stubborn grip on AI hardware as geopolitics bite, Belgium firing the starting gun on NIS2 cybersecurity audits, and Meta cutting eight thousand jobs while betting big on AI agents. Let's start with Google and the robots that can finally read a gauge.

Robots that can actually look at a pressure gauge and understand what it means.

Samantha: So Google has unveiled something called Gemini Robotics-ER 1.6. The ER stands for embodied reasoning, and honestly Bob, this feels like one of those quiet announcements that ends up mattering more than the flashy ones.

Bob: It really does. The headline capability is spatial understanding and multi-view perception. In plain English, a robot can now look at an industrial environment from several angles, work out where things are in three dimensions, and crucially, read instruments. Gauges, dials, displays. The sort of stuff a human inspector does on a factory floor.

Samantha: And that's the smart move, right? They're not promising you a robot butler.

Bob: No, and that's refreshing. The pitch is reliability in boring but high-value tasks. Checking pipelines, monitoring production lines, doing the rounds in a warehouse. Multi-step workflows, not just one repetitive motion. That's the real unlock.

Samantha: Because a lot of industrial AI has been stuck at the demo stage. Impressive video, terrible real-world performance.

Bob: Exactly the problem this is trying to solve. If a robot can reliably do an inspection round and flag anomalies, that's genuine productivity. Not hype.

Samantha: The bottom line here is that we're watching robots move from doing to understanding. A robot arm that welds the same seam a thousand times is useful. A robot that can walk into a room it's never seen and figure out what's wrong with the machinery, that's a completely different category.

Bob: And Google has the cloud infrastructure and the Gemini model family to back it up. They're not a robotics company in the traditional sense, but they are becoming the brain vendor for robotics companies.

Samantha: For Europe this lands in a very specific spot. Germany, Northern Italy, Czechia, Poland, these are manufacturing economies staring at demographic decline and labour shortages that aren't going away.

Bob: And European industrial robotics is strong. KUKA, ABB, Universal Robots in Denmark. The question is whether they partner with Google's reasoning layer or try to build their own. There's a real strategic decision there.

Samantha: Do you let an American company be the brain inside your European robot?

Bob: That's the sovereignty question, isn't it. And it echoes all the debates we've had about cloud infrastructure. Europe wants to benefit from the technology but not depend on it. Easier said than done when the capability gap is real.

Samantha: For European factory operators though, this is mostly good news. Better inspection, safer workplaces, less downtime. Just keep one eye on where the intelligence actually lives.

Bob: And speaking of embodied AI going live, our second story fits right in. AGIBOT is calling 2026 Deployment Year One.

A robotics company is declaring this the year embodied AI stops being a demo.

Samantha: AGIBOT held its 2026 Partner Conference and the language was bold. Deployment Year One. Four new robotic platforms. A new architecture they're calling One Robotic Body, Three Intelligences.

Bob: Which is a mouthful, but the idea is layered intelligence. Perception, reasoning, and action as separate but integrated systems running on a single physical platform. Foundational models underneath, application models on top.

Samantha: And they're talking about an open ecosystem. Partners, developers, third-party applications. That's the model that worked for smartphones. Whether it works for robots is the big bet.

Bob: The strategic shift here is important. For years embodied AI has been trapped in research labs and isolated pilot projects. What AGIBOT is saying is, we're done with that phase, we're going to scale. Retail, hospitality, logistics, services.

Samantha: Which is either very ambitious or very premature.

Bob: Probably both. But you only get to define the category if you move first. And coming right after Google's announcement, you can see a pattern. Multiple players are convinced this is the year embodied AI tips into production.

Samantha: I want to be careful here because we've been told robotics is about to explode roughly every two years since 2015. What's different now is that the underlying AI models are actually capable of the reasoning these tasks require.

Bob: And the cost curve has come down. Sensors, compute, actuators. A humanoid robot that cost half a million five years ago costs a fraction of that now.

Samantha: For European industries facing real labour gaps, this offers genuine relief. But the flip side is the job displacement conversation, and Europe tends to take that conversation more seriously than most regions.

Bob: Works councils, unions, social partners. The deployment of embodied AI in Europe will look different from deployment in China or the US. Slower, more negotiated, more focused on augmentation rather than replacement.

Samantha: Which could be a competitive disadvantage or a competitive advantage depending on how you look at it. Robots that work alongside humans might actually be more productive than robots that replace them.

Bob: And European robotics firms need to decide fast whether they're players or buyers in this market.

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.

xAI is moving beyond chatbots and into infrastructure.

Bob: xAI has launched Grok Speech to Text, an API for developers. Over 25 languages, real-time streaming, speaker diarization, which is the ability to tell who's talking in a multi-person conversation.

Samantha: And this isn't them entering the market cold. The technology already runs Grok Voice, Tesla's in-car assistant, and Starlink customer support. So they've been battle-testing it.

Bob: That's the real story. Musk's AI play has been positioned as a competitor to OpenAI and Anthropic at the model level. But an API product like this says, we also want to be the plumbing. Compete with Deepgram, compete with Google's Speech API, compete with Whisper.

Samantha: And they're leaning on pricing. Described as competitive, which in this market usually means aggressive.

Bob: The interesting angle is the 25-plus languages. Speech-to-text has historically been strong in English and mediocre in everything else. If Grok STT actually delivers high-quality transcription in Dutch, Polish, Greek, Portuguese, that changes the market.

Samantha: Every voice assistant, every meeting tool, every call centre system, every accessibility feature, they all run on speech-to-text. It's infrastructure nobody sees until it breaks.

Bob: And the quality difference between a good transcription engine and a mediocre one is the difference between a product that works and a product you abandon after a week.

Samantha: Europe has 24 official EU languages and dozens of widely spoken regional languages. Any speech technology that takes that seriously has a huge market here.

Bob: The caveat is regulatory. GDPR, the AI Act, sector-specific rules in healthcare and finance. European developers will want to know where the audio gets processed, how it's stored, what happens to the training data.

Samantha: And xAI's track record on data governance is, let's say, not the most reassuring in the industry.

Bob: That's putting it diplomatically. But if the product is good and the price is right, European developers will test it. Pragmatism usually wins over principle when you're shipping a product.

Taiwan is still the single point of failure in global AI.

Samantha: A new report reminds us that TSMC remains the absolute cornerstone of the AI hardware supply chain. The US has more data centres, but the chips inside them are overwhelmingly made in Taiwan.

Bob: And TSMC just reported a 40.6 percent year-on-year revenue increase in Q1 2026. AI demand is the engine. But they're warning about component price hikes driven by geopolitical instability.

Samantha: Helium. You wouldn't think that's the bottleneck in advanced semiconductors, but it absolutely is. It's used in chip fabrication and in data centre cooling. And Middle East conflicts are squeezing supply.

Bob: This is the fragility nobody talks about. Everyone obsesses over the Taiwan Strait scenario, which is legitimate, but the supply chain has a dozen other chokepoints. Helium from Qatar, rare earths from China, specialised chemicals from Japan.

Samantha: We talk about AI as if it's software that floats in the cloud. It's not. It's silicon made in one country, using gases from another, on machines built in a third, powered by electricity from a fourth.

Bob: Every step of that chain is a policy risk. And unlike software, you can't just fork it and move it to a different data centre.

Samantha: The EU Chips Act, passed in 2023, aims to double Europe's share of global chip production to 20 percent by 2030. We are behind schedule.

Bob: Intel's Magdeburg project has had delays, TSMC's Dresden fab is progressing but slowly, and the cutting-edge nodes, the three-nanometre and two-nanometre processes that matter for AI, those aren't coming to Europe any time soon.

Samantha: So Europe's digital sovereignty story has a hardware gap at the centre of it.

Bob: And with helium supply wobbling and Taiwan's geopolitical exposure unchanged, the urgency is only going up. This isn't a five-year problem, it's a right-now problem.

Belgium just became the first EU country to start NIS2 enforcement.

Samantha: Today, April 18th, Belgium's NIS2 conformity assessment deadline kicked in. Essential entities, that's energy, transport, healthcare, digital infrastructure and more, must now complete formal cybersecurity audits.

Bob: And this is the start of a cascade. Other EU member states are expected to implement similar deadlines through the rest of the year. Belgium went first, but the others are coming.

Samantha: For listeners who haven't followed this closely, NIS2 is the EU's updated cybersecurity directive. It mandates risk management, a 24-hour early warning after an incident, a 72-hour formal report, and personal liability for management. Your CEO can be held accountable.

Bob: Which changes the conversation in boardrooms. Cybersecurity used to be something the IT director worried about. Now it's something the chief executive worries about because their own liability is on the line.

Samantha: Documented risk assessments, tested incident-response plans, supply chain reviews. None of this is optional anymore. And the fines are substantial.

Bob: The key takeaway here is that a lot of organisations have been treating NIS2 as a paper exercise. That's about to stop working. Audits mean auditors show up and ask to see the evidence.

Samantha: Every essential entity in the EU, and that's a huge category, needs to be able to demonstrate compliance. That means spending. Tools, processes, people.

Bob: And there's a talent shortage. European cybersecurity professionals are already in short supply, and NIS2 is going to intensify demand sharply.

Samantha: Expect salaries to go up and expect a lot of consultancy invoices over the next twelve months.

Bob: It's also a moment where European cybersecurity vendors have a genuine home-market advantage. A European buyer audited under European rules might reasonably prefer a European supplier.

Samantha: Now before we wrap up, one more story, and it's a slightly lighter one but with real implications. Meta is cutting jobs again.

Eight thousand Meta employees, roughly 10 percent of the company, are being let go, and AI is the explicit reason.

Bob: Starting May 20th, Meta is planning its next round of layoffs. About 8,000 people. And the company is openly saying this is about creating a leaner organisation with fewer management layers and more AI-assisted employees.

Samantha: What's striking is how direct they're being about it. Usually companies dress these things up. Meta is saying, we're restructuring around AI, we need fewer people, and engineers are moving into a new Applied AI group building agents for coding and complex tasks.

Bob: That detail about fewer management layers is really interesting. The assumption has always been that AI comes for the low-skill jobs first. What we're actually seeing at Meta and elsewhere is that AI is coming for coordination work. Project managers, team leads, middle managers.

Samantha: Because an AI agent can summarise status, schedule meetings, track projects, draft updates. That's a lot of middle management work right there.

Bob: The fun part, or maybe the unsettling part, is that the people most vocal about AI hype are increasingly the ones being made redundant by it.

Samantha: For a long time the debate has been abstract. Will AI take jobs, won't it. Now we have a named company, a named number, and a named reason. Eight thousand people at Meta because of AI adoption.

Bob: And Meta isn't struggling financially. This isn't cost-cutting in a downturn. It's strategic restructuring in a profitable company. Which is arguably more telling.

Samantha: Meta has thousands of employees in Europe, in Dublin, London, Paris, and elsewhere. European layoffs are legally more complex than American ones. Works councils, collective consultation, notice periods.

Bob: Which means the European impact will unfold more slowly and with more negotiation. Not necessarily fewer job losses, but the process is different.

Samantha: And it puts pressure on European governments to think seriously about reskilling. If AI is genuinely displacing white-collar work, the policy response needs to be bigger than a training voucher.

Bob: The wow factor here is really the speed. Two years ago generative AI was a novelty. Now it's explicitly cited as the reason 8,000 people are being made redundant at one of the biggest tech companies in the world.

Samantha: Makes you wonder what the number looks like in 2027.

Bob: That's the question that should keep policymakers awake at night.

Samantha: Today we covered: Google's robotics reasoning model, AGIBOT's deployment year for embodied AI, xAI's new speech-to-text API, Taiwan's critical role in AI hardware, Belgium starting NIS2 enforcement, and Meta's eight thousand AI-driven layoffs.

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.