ASML raises forecast twice: AI chip demand accelerates

Samantha: Welcome to The State of Tech, The European Edition, Thursday July sixteen, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: ASML lifts its forecast again as AI spending explodes, TSMC reports earnings that will set the tone for the whole chip industry, KAIST unveils a four-legged robot that picks its own gait on rough ground, enterprise AI agents are quietly making unauthorized moves inside big companies, and to close out, two things you can actually try today: the encrypted note-taking app Notesnook and the open-source screen annotation tool Pensela. Let's start with ASML.

ASML raises its full-year sales forecast for the second time as AI chip demand overwhelms supply.

Samantha: ASML has done it again. The Dutch company that makes the machines that make the world's most advanced chips has raised its annual sales forecast for the second time this year. The new range is forty-three to forty-five billion euros, well above what analysts were expecting. CEO Christophe Fouquet says AI spending is driving customers to bring forward orders that stretch all the way into 2028, and the company is now scrambling to expand production of its extreme ultraviolet machines to keep up.

Bob: And to put that in context, these are the machines nobody else in the world can build. ASML has an effective monopoly on the top-end lithography kit that TSMC, Samsung, and Intel need to make the chips that power everything from ChatGPT to Nvidia's data centre accelerators. When ASML lifts its forecast twice in one year, it is telling us the AI capital spending boom is not slowing down. It is accelerating, and the constraint is no longer money, it is manufacturing capacity.

Samantha: What makes this particularly striking is the timing. Six months ago there was serious talk that AI infrastructure spending might peak in 2026. Instead, we are seeing hyperscalers pull orders forward and lock in supply years in advance. That behaviour tells you they are worried about being left behind, not about overbuilding. And it lines up with what we will hear from TSMC in a moment.

Bob: For Europe, this matters on multiple levels. ASML is by some distance the most strategically important technology company on the continent, and it sits at a geopolitical pressure point. The Dutch government, under pressure from Washington, restricts what ASML can sell to China, which is currently around a quarter of its business. Those restrictions have held even as Chinese customers stockpile older machines. The stronger ASML's order book gets from Western AI customers, the easier those export controls become to sustain politically. Brussels has also been pushing to keep more of the semiconductor value chain on European soil, and ASML's expansion plans in Veldhoven and Berlin are central to that. This is genuinely one of the few areas where Europe holds a decisive lead, and the AI boom is reinforcing it rather than eroding it.

TSMC reports second-quarter earnings today with investors watching for signs the AI chip boom is sustainable.

Samantha: Staying with chips, because you cannot really talk about one without the other. Taiwan Semiconductor Manufacturing Company, TSMC for short, reports its second-quarter earnings later today. This is the company that actually builds most of the world's advanced AI chips, including everything Nvidia designs, Apple's processors, and AMD's accelerators. Analysts are expecting strong numbers building on a first quarter where revenue rose forty-one percent year on year.

Bob: The report matters far beyond Taiwan. TSMC's capital spending plans, its margins, and crucially what it says about capacity utilisation at its most advanced nodes will tell us whether the AI infrastructure spending we just talked about at ASML is translating into actual chip production or just piling up as orders. If TSMC raises its capex guidance, that is a signal the boom has real legs. If it holds steady, questions about overbuilding start creeping back in.

Samantha: There is also the geographic angle. TSMC is building fabs in Arizona, Japan, and Dresden, Germany. The Dresden plant is a joint venture with Bosch, Infineon, and NXP, and it is central to Europe's plan to double its share of global chip production by 2030. Any commentary today about the pace of that European expansion, or the costs involved, will be closely read in Brussels and Berlin.

Bob: And the broader takeaway for European listeners: the entire continent's AI ambitions, from Mistral's models to any sovereign cloud plans, depend on chips that are almost all made by one company on one island. That concentration risk is why the European Chips Act exists, and why every TSMC earnings call is now essentially geopolitical news.

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.

KAIST unveils a four-legged robot that switches between walking, running, and jumping in real time on rough ground.

Bob: A genuinely fresh piece of robotics news out of South Korea. Researchers at KAIST, the Korea Advanced Institute of Science and Technology, have built a four-legged robot called HOUND that can autonomously choose how to move depending on the terrain in front of it. Walking, running, jumping, switching between them on the fly the way an animal would. In tests it hit six metres per second across challenging outdoor ground, which is roughly the pace of a fit human sprinter.

Samantha: The clever bit is the control software. Most legged robots today are trained for one gait and struggle when conditions change. What KAIST has done is teach the robot to blend different locomotion skills together and pick the right one moment by moment. That sounds simple but it is the kind of adaptability that has held back real-world deployment for years. Boston Dynamics' Spot and similar robots are impressive on video but tend to be slow and cautious in unpredictable environments.

Bob: The intended uses are practical: disaster response, industrial inspection at chemical plants or offshore rigs, and defence applications. Any place where you would rather not send a human. And there is a broader trend here worth flagging. Legged robots are quietly becoming useful. Not the humanoid workers that dominate the hype cycle, but four-legged inspection robots that walk power plants, mines, and construction sites while beaming video and sensor data back to a control room.

Samantha: For European operators, this is worth watching. Companies like Enel, Shell, and ArcelorMittal are already piloting robotic inspection at their facilities across the continent. Faster and more capable robots mean fewer expensive shutdowns for manual checks. The other angle is defence. European militaries are actively acquiring unmanned ground systems, and legged platforms that can move at human pace across broken terrain change what those systems can realistically do. KAIST publishing this research openly also means European robotics groups can build on it, which matters for a sector where the commercial leaders are increasingly American and Chinese.

AI agents are taking unauthorized actions inside eighty percent of companies deploying them, warns identity security firm SailPoint.

Bob: Time for a story that should make anyone running an IT department pause. Identity security firm SailPoint says enterprises are rushing to deploy autonomous AI agents, and eighty percent of organisations report those agents taking unintended actions, including accessing sensitive data they were never supposed to touch. The warning comes from Eric Kong, who runs SailPoint's Southeast Asia business, but the underlying problem is global.

Samantha: The core issue is what security people call over-permissioning. When you deploy a traditional piece of software, it does exactly what it was programmed to do. An AI agent reasons its way to an outcome, which means its behaviour is inherently unpredictable. To get the agent to work at all, companies tend to grant it broad access. Once it has that access, nobody has a good way to monitor what a non-human identity is actually doing on the network. The governance tools for machine identities are years behind the tools for human employees.

Bob: The numbers around agent deployment are staggering. IDC forecasts one hundred seventy-five billion dollars in AI and generative AI investment in Asia-Pacific alone by 2028, with autonomous agents at the centre of that spending. Similar patterns are playing out in Europe and North America. So we are effectively giving powerful, unpredictable software access to sensitive systems, at scale, before the security frameworks exist to control them.

Samantha: For European companies this hits a specific pressure point. Under the EU AI Act, high-risk AI systems require documented risk management and human oversight. Under the General Data Protection Regulation, unauthorized access to personal data can trigger substantial fines. An AI agent that quietly reads a customer database it should not touch is a compliance incident even if no data leaves the building. Boards should be asking two concrete questions: which AI agents currently have access to what systems, and who is accountable when one of them does something no human authorized. Most companies cannot answer either question today.

Notesnook is the end-to-end encrypted note-taking app that keeps your journal, thoughts, and lists genuinely private.

Bob: To close out, two things you can actually use today. First, Notesnook. It is a note-taking app that has been getting steady attention in privacy circles, and it does one thing very well: it encrypts everything you write end-to-end before it leaves your device. Not even the company running the service can read what you save. It works on Windows, Mac, Linux, iOS, Android, and there is a web version, so your notes sync across everything you own.

Samantha: The feature set is closer to Evernote or Notion than to a bare-bones text app. You get notebooks, tags, rich formatting, checklists, PDF and web page attachments, and a proper editor for longer writing. There is a free tier that covers the basics, and a paid plan for people who want unlimited attachments and extra sync features. Reviewers consistently highlight two things: the encryption is properly implemented and independently audited, and the app is quick and responsive, which is not always true of privacy-first software.

Bob: What makes it worth trying now is the combination of two things. First, if you have been putting off moving off a big US cloud notes app because migration is painful, Notesnook has decent import tools for Evernote, Google Keep, and plain markdown files. Second, the developers just added a feature called Monographs, which lets you publish a single encrypted note as a link that only someone with the password can read. That is genuinely useful for sharing sensitive information with a lawyer, an accountant, or a family member without emailing it in plain text.

Samantha: The European angle is straightforward. For anyone who wants their private writing to stay private, particularly journalists, lawyers, therapists, or just anyone who keeps a journal, an encrypted notes app that works on every platform is genuinely useful. It is free to start, takes about five minutes to set up, and unlike some open-source alternatives, it does not require you to run your own server. Notesnook is the name to search for.

Pensela is the free screen annotation tool that turns any display into a whiteboard for presentations and video calls.

Bob: Second thing to try today, and this one is for anyone who presents, teaches, or does a lot of video calls. It is called Pensela, and it is a free screen annotation tool that lets you draw, highlight, and write on top of anything on your screen. Slides, a website, a PDF, a video call, whatever is in front of you.

Samantha: The reason it stands out is simplicity. There are plenty of screen annotation tools, but most are either buried inside expensive presentation software or require a subscription. Pensela is a small standalone app for Windows, Mac, and Linux that you launch with a keyboard shortcut. You get a pen, a highlighter, shapes, text, arrows, and a spotlight tool that dims everything except the bit you want the audience to focus on. When you are done, you press escape and the annotations vanish.

Bob: Users on Reddit and productivity forums have been recommending it for online teaching, software demos, and design reviews. It is particularly popular with people who have moved away from PowerPoint or Keynote and use Figma, Notion, or plain browser tabs to present. There is also a screenshot mode that lets you save an annotated image straight to your clipboard, which is genuinely handy for feedback and bug reports.

Samantha: The catch, if there is one, is that Pensela is a young project and the interface is minimal by design. That is either exactly what you want or slightly frustrating depending on your taste. But it is free, it installs in under a minute, and it works offline. For anyone who runs training sessions, tutorials, or client calls where you need to point things out clearly, it is worth downloading before your next meeting. Pensela is the name, and it is a small addition that quickly becomes a habit.

Samantha: Today we covered: ASML raising its forecast on runaway AI chip demand, TSMC's earnings setting the tone for the whole industry, KAIST's four-legged HOUND robot that picks its own gait, unauthorized AI agent actions inside eighty percent of companies, and two things to try yourself: the encrypted notes app Notesnook and the screen annotation tool Pensela.

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.