SK Hynix record profit, $75B payout + Apple M6 iMac

Samantha: Welcome to The State of Tech, The European Edition, Wednesday October twenty-eight, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: SK Hynix posts record profits and promises a seventy-five billion dollar payout as AI demand explodes; Israeli researchers build an AI that spots brain bleeds faster than doctors; Apple unveils a new iMac with its M6 chip; Canadian regulators sound the alarm on AI model drift; and to close out, two things you can actually try today, the free diagram tool Diagrams dot net Desktop and the open-source audio workstation Ardour. Let's start with the memory chip boom.

SK Hynix posts record quarterly profit and pledges seventy-five billion dollars to shareholders as AI demand surges.

Samantha: South Korean chipmaker SK Hynix has just reported a more than seven-fold jump in quarterly operating profit, a record high, and the company is handing back around one hundred trillion won, roughly seventy-five billion dollars, to shareholders this year. That is a mix of cash dividends, share buybacks and cancellations. The driver is simple: AI infrastructure spending that keeps pushing memory chip prices higher, and SK Hynix is one of the main suppliers of the specialised memory that sits next to Nvidia's AI accelerators.

Bob: And the forward numbers are even more striking than the quarter itself. SK Hynix now says it expects demand to outstrip supply all the way through 2030. They have already locked in long-term agreements covering more than fifty percent of their total production capacity for 2027 and 2028. In plain terms, more than half of what they will make in the next few years is already sold. That is not a company bracing for a bubble, that is a company behaving as if the AI build-out is a decade-long construction project.

Samantha: For Europe, the signal matters on two levels. First, European cloud providers and AI labs rely on the same scarce supply, so booked-out capacity means tighter allocations and higher prices for anyone trying to train or serve large models from Frankfurt, Dublin or Paris. Second, it strengthens the political argument around the European Chips Act, because right now the critical memory layer of the AI stack sits almost entirely in Asia.

Bob: The bottom line is that memory, not just the headline AI processors, has become the real constraint. Former Intel chief Pat Gelsinger said as much on Tuesday at the Taiwan Deep Tech Summit, calling AI increasingly a memory workload, and arguing the broader AI cycle has more than twenty years to run. He pushed back on bubble talk, while allowing for what he called periods of disruption along the way. Taken together with SK Hynix's order book, it reads less like a spike and more like a very long plateau, which is a different investment case for European firms deciding whether to build or to rent.

Samantha: And it changes procurement conversations too. A European bank or hospital group signing a multi-year AI contract now has to assume memory scarcity is a structural cost, not a temporary one.

Israeli researchers build an AI that spots brain bleeds in seconds, outperforming doctors on CT scans.

Bob: Moving on. Researchers in Israel have presented an AI system that can detect life-threatening brain haemorrhages in seconds, reportedly outperforming physicians at spotting them on CT images. The work was shown at a CTech event, and the pitch is straightforward: the model analyses scans in real time and sends an instant alert to the clinical team, so a patient with a bleed does not sit in a queue waiting for a radiologist to open the file. The same group also described AI tools for personalised cancer care and new communication aids for ALS patients.

Samantha: The speed point is the one to hold on to. In a stroke or trauma case, every minute of delay between the scan and the decision to operate translates into brain tissue lost. If a triage model can flag a bleed the moment the scan lands on the server, the hospital can start preparing the operating theatre before a human has even read the image. The researchers frame it as a safety net, not a replacement for the radiologist, which matters for how European regulators will look at it.

Bob: And that regulatory lens is where Europe gets interesting. Under the EU AI Act, a tool like this falls into the high-risk category for medical devices, which means documented training data, human oversight, and post-market monitoring. Several European hospital networks, in Germany, the Netherlands and the Nordics, already run similar triage pilots for stroke. The Israeli work does not land in a vacuum; it joins a growing shortlist of systems that European procurement teams are now actively evaluating.

Samantha: The practical question is integration. A model that is faster than a doctor only helps if the alert reaches the right person on the right pager, inside the hospital's existing workflow. That is the unglamorous engineering part, and it is where most clinical AI projects stall. The upside, if it is done well, is a measurable drop in time-to-treatment for some of the most urgent cases a hospital sees.

Bob: It also fits a wider pattern. Diagnostic AI is quietly becoming one of the few areas where the technology has moved from demo to daily use.

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.

Apple unveils a new iMac powered by the M6 chip, promising a major leap in on-device AI.

Bob: Apple has announced a new iMac built around its M6 chip, with pre-orders starting today and shipping from the sixth of November. The headline is performance, especially for AI tasks. The M6 has a faster processor and graphics engine, plus what Apple calls neural accelerators in each core, meaning more AI work can happen directly on the machine instead of in the cloud. It runs macOS 27 and the next generation of Apple's own AI features, including the updated Siri. The design is the familiar thin all-in-one, now in seven colours.

Samantha: The interesting shift here is not the colours, it is where the AI runs. By pushing more capability into the chip itself, Apple is betting that families, freelancers and small businesses would rather have their photo edits, document summaries and voice interactions handled on the desktop in front of them, rather than sent to a data centre. For European users that is a quietly important privacy argument, because data that never leaves the machine is data that does not have to be governed by a cross-border transfer agreement.

Bob: It also lands in a specific competitive moment. Microsoft and its PC partners have spent the past year marketing AI laptops with dedicated neural chips, and Google is pushing Gemini deeper into Chromebooks and Android. Apple's answer is to keep the AI story tied to its own silicon and its own operating system, from phone to tablet to desktop. For a small European design studio, that means a credible option to run creative AI tools without an ongoing cloud bill, and without the governance headache of sending client work to a third-party model.

Samantha: The caveat is pricing and timing. Apple has not yet signalled how aggressively it will push the new Siri features into European languages, and historically some of the more advanced AI functions arrive in English first. So the hardware will be on desks in November, but the full software story for European users may land in stages.

Canadian regulators warn that AI models in healthcare quietly degrade as the real world changes.

Bob: Our fourth story takes us to Canada, where physician regulators are raising the alarm about something called AI model drift. The idea is simple and uncomfortable: an AI system trained on past data gradually gets worse as the world it is trying to describe changes around it. Researchers showed this with an opioid-risk prediction model from Alberta. It was trained before the pandemic, and during COVID-19 its performance declined significantly, because prescribing patterns, patient behaviour and overdose risks had all shifted. The findings were presented at the Canadian Network of Agencies for Regulation conference.

Samantha: And this is one of those stories that sounds technical but is actually about trust. A hospital that bought a predictive tool three years ago may still be using it today, assuming it works as advertised, when in fact the ground beneath the model has moved. The Canadian regulators are arguing that oversight cannot stop at the moment of deployment; it has to be continuous, with regular checks on whether the model is still accurate in current conditions. That is a very different compliance model from the one most hospitals and insurers run today.

Bob: For Europe, this maps directly onto the AI Act's requirement for post-market monitoring of high-risk AI. The Canadian evidence gives European regulators and health systems a concrete example to point at, with real numbers, when they ask vendors how they plan to detect drift. It also pushes back against the idea that an AI certification is a one-and-done exercise.

Samantha: And it has a broader read-across. Model drift is not unique to healthcare. Credit scoring, fraud detection, energy demand forecasting, all of these use AI trained on historical patterns that keep shifting. A bank in Milan or an energy utility in Rotterdam faces the same quiet decay problem. The lesson from Alberta is that nobody notices until someone goes looking, which is exactly why the regulators want continuous checks baked in from day one.

Bob: The uncomfortable truth is that drift monitoring costs money and nobody budgets for it upfront. This story is a useful nudge to change that.

Diagrams dot net Desktop turns any laptop into a free, offline workspace for flowcharts and architecture diagrams.

Samantha: To close out, two things you can actually try today. The first is Diagrams dot net Desktop, sometimes still referred to by its old name draw. It is a free, open-source diagramming app that runs entirely on your own machine, with no account and no cloud sync unless you explicitly turn it on. You can build flowcharts, org charts, network diagrams, floor plans, wireframes, and the kind of boxes-and-arrows architecture sketches that normally live in expensive subscription tools.

Bob: The reason to mention it today is that many teams default to paid cloud diagramming services without realising there is a free desktop version that produces the same file formats. It runs on Windows, Mac and Linux, and the files are plain XML, so you can commit them to a shared folder or a code repository without lock-in. Reviewers often highlight how quickly it opens and how little it gets in your way, which is unusual for this category of software.

Samantha: A couple of practical notes. The desktop version is maintained as a standalone download, separate from the browser version, which keeps your diagrams off any server by default. It can export to PNG, PDF and SVG, so a diagram you draw in the morning can be pasted into a slide deck in the afternoon. And because it is open source, it is widely used inside organisations that cannot send internal architecture diagrams to third-party clouds.

Bob: For European teams working under strict data handling rules, that offline-by-default behaviour is the quiet selling point. You can sketch a sensitive system layout without first negotiating with your compliance officer about where the file lives. It takes about five minutes to install and open your first blank canvas, and the keyboard shortcuts will feel familiar to anyone who has used a mainstream office suite.

Samantha: Not an AI tool, deliberately. Just a solid, boring piece of software that does one thing well, which is sometimes the most useful kind of pick.

Ardour turns any computer into a free, professional-grade multitrack audio studio for podcasts and music.

Bob: The second pick is for anyone who records audio, whether that is a podcast, a band rehearsal or a voice-over. Ardour is a free, open-source digital audio workstation that runs on Windows, Mac and Linux. It handles unlimited tracks, non-destructive editing, mixing with professional-grade plugins, and multi-channel recording from an external interface. People who use it tend to describe it as the closest free equivalent to the big paid studios, without the subscription fee attached.

Samantha: What makes it worth trying today, rather than just bookmarking, is how approachable it has become. Earlier versions had a reputation for a steep learning curve, but the project has invested heavily in a cleaner interface and better defaults. A podcaster can open it, create a two-track project for host and guest, drop in noise reduction, and export a finished episode without ever opening a browser. Everything stays on the local disk, which also means no upload times and no file size caps.

Bob: The project is funded through optional user contributions, which is why the downloaded builds on their website carry a suggested payment. The source code and the Linux builds are entirely free, and most Linux distributions include it in their standard software catalogue. For European users who want a studio that is not tied to a US cloud service for backups or licensing, that independence is part of the appeal.

Samantha: One practical tip reviewers often share: pair Ardour with a cheap USB audio interface and a decent microphone, and you have a recording setup good enough for professional output, for well under the cost of a single year of a competing subscription. It is overbuilt for a casual voice memo, but exactly right for someone who wants to grow into serious audio work without changing tools later.

Samantha: Today we covered: SK Hynix's record profit and seventy-five billion dollar shareholder payout on the back of AI demand; an Israeli AI that detects brain bleeds faster than doctors; Apple's new M6 iMac pushing AI onto the desktop; Canadian regulators warning about AI model drift in healthcare; and two things to try yourself: the free offline diagramming app Diagrams dot net Desktop, and the free open-source audio studio Ardour.

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.