AMD Buys World Labs for $8.2B + OpenAI Dots Agents

Samantha: Welcome to The State of Tech, The European Edition, Wednesday September thirty, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Trump's voluntary AI safety pact with the big tech companies, TSMC ramps up its most advanced chips as demand overheats, Australia orders an urgent cyber refresh after an AI-powered hack on its health system, a global study exposes a widening AI skills divide at work, and two things you can actually use today, the encrypted European messenger Olvid and the free open-source photo tool Nitrux Pix. Let's get into it.

AMD Writes an $8.2 Billion Cheque for Fei-Fei Li's World Labs

Samantha: Okay, this one caught me off guard. AMD is buying Fei-Fei Li's startup World Labs for 8.2 billion dollars. That's a chipmaker buying a company that builds what people call world models, AI that understands three-dimensional space rather than just text.

Bob: It's a strange pairing on paper. AMD sells silicon, and World Labs is basically a research outfit with a very famous founder. But if you squint, the logic is pretty clear. Nvidia doesn't just sell chips, it sells the whole software stack around them, and AMD has spent years trying to close that gap.

Samantha: So this is about owning the interesting workloads, not just the hardware they run on?

Bob: Partly that. If you control a leading world model, you can tune your chips for it, show customers what your hardware can do, and maybe get developers building on your platform first. It's a bit like a console maker buying a studio.

Samantha: And world models are having a moment. Robotics, simulation, games, digital twins, they all need machines that grasp how physical space works.

Bob: Which is where the real money might be, beyond chatbots. A robot needs to understand that a cup is behind a laptop and how it'll move if you nudge it. Language models are weak at that. So whoever gets there first with good spatial AI has a big head start in physical AI.

Samantha: Still, 8.2 billion is a lot to pay for a startup that isn't printing money.

Bob: It is, and closing isn't until year-end, so regulators get a look. I don't expect it to be blocked, since AMD isn't the market leader, but it's another sign of how frothy AI valuations are. Companies are paying for talent and position, not current revenue.

Samantha: There's the talent angle too. Will Fei-Fei Li and her team actually want to work inside a chip company?

Bob: That's the real risk. Acquisitions like this often stumble on culture. Researchers who valued independence can drift away within a year or two. AMD will need to give them room, or it's paying billions for a logo.

Samantha: So what do we watch for?

Bob: Two things. Whether World Labs' models stay open and available on other hardware, or become AMD-first. And whether Nvidia responds with its own deal. When one big player buys a prized startup, the others rarely sit still.

TSMC lifts 2-nanometer wafer target by twenty percent as AI demand overheats.

Samantha: Next up, Taiwan Semiconductor Manufacturing Company, or TSMC, has quietly raised its production target for the most advanced chips it makes. The company is now aiming for around one hundred twenty thousand two-nanometer wafers per month by the end of this year, roughly twenty percent more than previously planned. Two-nanometer wafers are expected to sell for about thirty thousand dollars each, a fifty percent jump on the current three-nanometer generation.

Bob: And crucially, TSMC says this is not speculative. These are locked-in orders from customers who need the chips for AI training and inference systems. That points to demand from the usual heavyweights, the American cloud giants, Nvidia, and increasingly custom silicon projects. When TSMC talks about pricing power, it is because there is essentially no alternative supplier at that node.

Samantha: For Europe this cuts two ways. On one side, European carmakers, industrial firms and the growing European AI scene rely on advanced chips that overwhelmingly come out of Taiwan. Higher prices and tighter supply mean higher costs for anyone building AI products, from a Munich robotics startup to a Paris genomics lab. On the other side, it strengthens the political case for the European Chips Act and for new fabs on European soil, even if those plants will not reach the two-nanometer frontier for years.

Bob: There is also a concentration risk story here. A single company in a single region making the world's most valuable chips is a strategic vulnerability that European policymakers keep circling back to. Nothing announced today changes that picture. If anything, it deepens it. The AI boom is being physically manufactured in one place, and everyone else is negotiating for their slice.

Samantha: Watch the second-order effects. If two-nanometer capacity is spoken for, less demanding customers get pushed onto older nodes, which tightens supply there too. That has knock-on effects for cars, medical devices and consumer electronics, exactly the industries that felt the 2021 chip crunch most sharply.

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 New Dots Never Clock Off

Bob: Let's talk about Dots, the headline launch from OpenAI's DevDay. These are always-on AI agents, each with its own cloud computer, powered by GPT-6 Astra. They run in ChatGPT, but also inside Slack and Teams.

Samantha: The always-on part is what matters. Until now, most assistants wait for you to type something. A Dot keeps working in the background. Give it a task, close your laptop, come back later, and it's still grinding away or waiting with a result.

Bob: And having its own cloud computer means it can actually do things, open apps, click around, run code, rather than just suggest what you should do.

Samantha: That's the leap. Plus there are over 4,000 plug-ins, so it can reach into the tools you already use. It's less a chatbot and more a junior colleague who never sleeps and occasionally misunderstands you.

Bob: That last bit worries me. If something is acting autonomously across my email and files, mistakes get expensive.

Samantha: They do. A wrong answer in a chat window is annoying, but a wrong action in a live system can delete something or send a message you can't recall. So permissions and audit trails matter far more than the demo suggests. Companies will want to see exactly what a Dot did and why.

Bob: OpenAI is under pressure on safety too. It's been open about models behaving in unintended ways and held back its top Astra model.

Samantha: Which makes it interesting to launch autonomous agents on that same technology in the same week. There's a tension between telling people to slow down and shipping more autonomy. I'm not saying it's hypocritical, but you can see why critics raise an eyebrow.

Bob: What about the business side? There's talk of a five hundred dollar a month plan.

Samantha: Yes, and that tells you who they're aiming at. Not casual users, but teams who'll pay real money if an agent replaces hours of grunt work. OpenAI needs to show revenue growth, and agents are the obvious way to justify premium pricing.

Bob: So does this actually change how we work?

Samantha: Possibly, but slowly. The first wave will be dull tasks like reporting, scheduling and data cleanup. If Dots handle those reliably, adoption will snowball. If they break things, trust evaporates fast.

PwC finds a widening AI skills divide across the global workforce.

Bob: Now to a study that will land on a lot of HR desks this week. Consultancy PwC has surveyed nearly fifty thousand workers across forty-eight countries and found a clear, widening divide in AI skills. The research sorts employees into four groups based on their AI proficiency, and warns that a large share of workers already feel they are falling behind more tech-savvy colleagues. PwC frames it bluntly as the risk of an AI have and have-not workforce.

Samantha: The interesting nuance is that this is not just about age or seniority. Some frontline workers are actively using AI tools every day, sometimes without their employer even knowing, while some knowledge workers are still avoiding them entirely. That creates uneven pockets of productivity inside the same company, and it makes performance reviews genuinely tricky when two people doing the same job are working with very different tools.

Bob: For European employers this collides directly with existing labour law and works council traditions. Rolling out AI training only for a favoured group can quickly become a fairness issue, and in some jurisdictions a co-determination issue. PwC's recommendation is essentially to invest deliberately in the workers with lower AI proficiency, rather than assume they will catch up on their own.

Samantha: Policymakers are watching this too. The European Commission has been clear that the AI transition needs to be inclusive, and several member states are experimenting with public AI literacy programmes. The PwC data gives them harder numbers to work with. If the gap keeps widening, it does not just hurt individual careers, it starts to distort national productivity figures.

Bob: The takeaway for listeners is fairly practical. If your employer is not offering structured AI training, the quiet reality is that colleagues who teach themselves are pulling ahead. And for managers, the question is no longer whether to train the team, but which tools to standardise on so that people are not each learning a different system in isolation.

The encrypted European messenger Olvid keeps your chats working without a phone number.

Samantha: Time to close out with two things you can actually use today. First up, a messaging app called Olvid. It is a French-built, end-to-end encrypted messenger that has quietly become one of the more interesting alternatives to the big names. What makes it stand out is that it does not require a phone number or an email address to sign up. Contacts are added by exchanging a short code in person, by QR code, or through a secure channel you already trust.

Bob: That design choice matters. Most mainstream messengers still tie your identity to your phone number, which means a leaked contact list or a SIM swap becomes a real problem. Olvid sidesteps that entirely. It has been certified by the French national cyber agency for use inside sensitive government communications, and a free version is available for individuals on iOS, Android, Windows, macOS and Linux.

Samantha: Practically, it takes about five minutes to set up. You install the app, create a profile with just a name, and then invite the people you actually want to talk to. Group chats, voice and video calls all work, and messages sync across your devices once you link them. It is not going to replace the group chat with your extended family overnight, but for work conversations, journalists, or anyone who simply does not want their number tied to yet another platform, it is a genuinely serious option.

Bob: One honest caveat: because it does not use phone numbers, you will not find friends automatically. That is the whole point, but it does mean the network you build on Olvid is one you assemble deliberately. For a lot of people that is a feature, not a bug. The app is free for personal use, with paid tiers for business features, and everything runs on servers based in Europe.

Samantha: If you have been meaning to move at least some of your conversations off the big American platforms, this is a low-friction place to start. Install it on one device, invite two or three people, and see how it feels.

The free open-source tool Nitrux Pix cleans up messy photo libraries on your own laptop.

Bob: And to round things off, something for anyone drowning in photos. Nitrux Pix is a free, open-source image manager and light editor that runs entirely on your own computer. Point it at a folder of holiday pictures, and it will help you organise them. Nothing gets uploaded to a cloud service in the process.

Samantha: The appeal is that it fits between two extremes. On one side you have the big cloud photo services, which are convenient but hand your entire visual life to a company overseas. On the other you have heavy professional tools built for photographers who want to grade every image. Nitrux Pix sits in the middle. It is for the person with fifteen thousand phone snaps who just wants to find the good ones and delete the rest.

Bob: Users often praise how quickly it chews through large folders, and reviewers highlight it as a real time saver. It runs on Linux, and because it is open source the code can be inspected by anyone. There are no ads, no accounts and no subscription.

Samantha: A realistic afternoon project: pick one messy folder, let Nitrux Pix analyse it, and spend twenty minutes going through its suggestions. Most people find they can free up several gigabytes of storage on the first pass and finally see the photos that actually matter.

Samantha: Today we covered: AMD Writes an $8.2 Billion Cheque, TSMC lifts 2-nanometer wafer target by, OpenAI's New Dots Never Clock Off, PwC finds a widening AI skills, The encrypted European messenger Olvid keeps, en The free open-source tool Nitrux Pix.

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.