14 oktober 2026 · 14:45
Chip industry tops $1 trillion; US AI incident rules
Global semiconductor sales have crossed one trillion dollars a year, nearly four years earlier than most forecasters expected, driven by AI data centre demand. In this episode of The State of Tech, we unpack what the milestone means for European chip policy and look at the White House's new rule requiring AI companies to report major incidents immediately. We also cover what that could mean for the EU AI Act.
Beluister deze aflevering:
Transcript
Samantha: Welcome to The State of Tech, The European Edition, Wednesday October fourteen, 2026. I'm Samantha Lawrence.
Bob: And I'm Bob Russell. Today: the chip industry crosses a trillion dollars in annual sales, the White House orders instant reporting of major AI incidents, a liquid-cooled switch blueprint for AI data centres, Apple unveils a new smart home hub, an alleged bugging incident at OpenAI, and Australia bets on sovereign AI. To close out, two things to try today: the open-source note app Standard Notes and the free offline reader Pocket Alternatives, sorry, Wallabag's new rival. Let's start with that trillion-dollar milestone.
The global semiconductor industry crosses one trillion dollars in annual sales, years ahead of forecast.
Samantha: The chip industry has officially crossed one trillion dollars in annual sales, nearly four years earlier than almost anyone predicted. The announcement came at SEMICON West in San Francisco, where SEMI Americas president Joe Stockunas laid out the numbers. From four hundred billion dollars in 2019, the sector is now on track for one point five to one point six trillion by the end of this year. And the forecast for 2030 is already above two trillion. The driver, unsurprisingly, is the AI build-out: GPUs, high-bandwidth memory, advanced packaging.
Bob: What this really shows is how much AI infrastructure is bending the entire supply chain. The money is not spread evenly. It is piling into a narrow set of components: accelerators, memory stacks, and the packaging that glues them together. Everything else, from cars to consumer electronics, is now competing for capacity against hyperscalers writing eleven-figure cheques. That is why lead times for certain memory parts have blown out again, and why older fabs are suddenly worth upgrading rather than closing.
Samantha: And there is a geopolitical layer underneath. A trillion-dollar industry is no longer just a tech story, it is industrial policy. The United States, South Korea, Japan and Taiwan are all pumping public money into fabs. China is doing the same, with a focus on avoiding Western tools. The question is whether all this capacity actually gets absorbed, or whether we are watching a classic chip cycle inflate in slow motion. Stockunas was bullish, but the industry has been wrong about peaks before.
Bob: For Europe, the headline number is nine new facilities planned across the region. That is meaningful, especially around Dresden, Magdeburg and parts of Ireland and Italy. But most of those plants will produce mature nodes for cars and industrial customers, not the leading-edge chips feeding AI data centres. So Europe benefits from the trillion-dollar wave in volume, less so in value. The Chips Act money is real, but compared to the hundreds of billions flowing in the US and East Asia, the European contribution remains modest, and policymakers in Brussels are openly debating whether a second, larger package is needed to avoid being left on the sidelines of the AI hardware boom.
The White House orders AI companies to report major incidents the moment they happen.
Samantha: The White House has rewritten a key piece of its AI safety playbook. From now on, major AI incidents must be reported to federal authorities immediately, with no internal investigation buffer. That is a significant shift from the previous rules, which gave companies time to triage before notifying Washington. The new directive also requires AI firms to build automated monitoring that can detect and flag serious failures in real time. It applies to the most advanced systems being deployed in critical areas of daily life.
Bob: The practical effect is that AI companies now have to treat serious model failures the way financial firms treat suspicious transactions: the clock starts ticking the moment something goes wrong. That means investment in internal detection, logging, and legal pipelines that can fire within minutes. Industry groups argue this creates operational vulnerability, because reports may land before anyone understands what actually happened. Supporters counter that waiting is exactly how smaller incidents become public disasters.
Samantha: There is also a transparency question underneath this. The directive, as described, requires reporting to the government, not necessarily to the public. So users may not learn about a serious AI malfunction unless regulators choose to disclose it. Civil society groups have already said the next fight will be over whether these incident reports become part of the public record, the way aviation safety reports eventually did.
Bob: In Europe, this lands squarely in the middle of ongoing AI Act implementation. The Act already requires serious incident reporting for high-risk systems, but the deadlines are measured in days, not minutes. If Washington really is moving to real-time disclosure, expect European regulators to feel pressure to tighten their own clocks. The discussion in Brussels suggests the Commission is watching closely, though no formal change has been proposed yet.
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.
CyberTAN and Marvell unveil a liquid-cooled switch platform built for the AI data centre era.
Bob: Taiwanese network specialist CyberTAN, working with chip designer Marvell, has unveiled a new switch platform called Pallas, designed specifically for the thermal reality of AI clusters. It was shown at the OCP Global Summit, the big open hardware conference, and it is built around Marvell's latest switch silicon, which pushes network capacity to levels most data centres still cannot absorb. The twist is that the whole thing is liquid-cooled directly on the chip, not just in the surrounding rack.
Samantha: And this matters because modern AI clusters are hitting the limits of what air cooling can physically do. When a single switch dissipates more than a thousand watts, you cannot just add bigger fans. Direct-to-chip liquid cooling moves heat away at the source, which lets data centres pack more compute into the same floor space without melting. The companies claim this will push power usage effectiveness, the standard efficiency metric, toward the best levels in the industry.
Bob: The underlying story here is that networking is quietly becoming as important as the GPUs themselves. Training a large model means thousands of accelerators talking to each other constantly, and if the switch fabric is slow or runs hot, the whole cluster stalls. So we are seeing a shift where switch silicon and cooling become strategic, not just plumbing.
Samantha: For Europe, this hits a sensitive nerve. Several member states, led by the Netherlands, Ireland and Germany, have tightened rules on data centre power and water use. Operators who want to expand have to show real efficiency gains, not promises. A design like Pallas, if it delivers on its claimed numbers, becomes one of the few credible paths to building new AI capacity on the continent without breaking local energy caps. Whether European operators adopt it quickly depends on cost and on how the direct-to-chip approach integrates with existing cooling loops.
Apple launches HomeView, a new smart home hub with cameras, thermostats and a doorbell.
Bob: Apple has officially announced HomeView, a new smart home hub with a six to seven inch screen and a built-in speaker. It runs a new operating system that blends elements of iOS, tvOS and the Watch software, with an upgraded Siri and face recognition built in. Alongside the hub, Apple is launching a full accessory line: indoor and outdoor cameras, thermostats, and a video doorbell with two cameras. Partners include LG and Schneider Electric. Setup is one-tap from an iPhone, and everything is managed through the Apple Home app.
Samantha: This is Apple finally entering categories it had left to Amazon, Google and specialist brands for years. The pitch is familiar: tight integration, end-to-end encryption, and a single app to manage everything. For existing Apple users, that is genuinely useful, because the current smart home experience across platforms is a mess of accounts and apps.
Bob: The question is price and openness. Apple has not detailed European pricing yet, and its accessories historically sit well above the competition. The company is also betting that consumers will pay a premium for the privacy pitch, which is plausible but not guaranteed. Users report that existing HomeKit setups work well but remain narrower than Google or Amazon ecosystems in terms of third-party device support.
Samantha: In Europe, the privacy angle lands well. Surveys consistently show European consumers are more wary of indoor cameras and voice assistants than American buyers. Apple's end-to-end encryption claim, if it holds up under scrutiny, gives the company a real marketing edge here. The Digital Markets Act also forces Apple to allow more interoperability than before, which could ironically make HomeView more useful to non-Apple households than earlier generations of HomeKit devices.
A Saudi engineer at OpenAI is arrested for allegedly bugging an executive conference room.
Bob: Federal prosecutors in San Francisco have arrested a thirty-five-year-old OpenAI engineer, Abdulrahman Alfozan, after the company's security team found a recording device hidden in a television in an executive conference room. According to the complaint, the USB device was wrapped in toilet paper and inserted into the TV. Alfozan, a Saudi national who joined OpenAI in July 2025, was reportedly caught on camera entering the room without authorisation, and investigators say he had searched for information about the facility and meeting schedules.
Samantha: What makes this case notable is not just the alleged espionage, but what it signals about AI companies as targets. OpenAI is now in a category once reserved for defence contractors and biotech firms: a place where the information in a single meeting could be worth billions. The complaint does not allege any foreign state involvement at this stage, and prosecutors have been careful to describe the charges as individual conduct.
Bob: The broader shift is that AI labs are being forced to adopt security postures closer to intelligence agencies than tech startups. Background checks, badge systems, surveillance of internal spaces, restrictions on personal devices in meeting rooms, all of this is becoming standard. The uncomfortable part is that this changes the culture of these companies, which were built on openness and academic exchange.
Samantha: For European AI companies, from Mistral in Paris to Aleph Alpha in Germany, the lesson is uncomfortable but clear. As soon as a lab becomes commercially or strategically interesting, it becomes a target. Industry groups have been quietly discussing shared security standards for months, and an incident like this will accelerate that conversation.
Workato teams up with Australian partners to deliver sovereign AI inside enterprise workflows.
Bob: To close out, two things to try today, and before that one more story worth flagging. Workato, an enterprise automation platform, has announced a partnership with two Australian firms, Gulanga Group and Trellis Data, to deliver what they call sovereign AI for Australian businesses and government agencies. The idea is to integrate locally built large language models, vetted for data sovereignty, into Workato's existing workflow platform. Early testing is already running with Gulanga's Dukes AI product, powered by Trellis Data's Agentaus model.
Samantha: Sovereign AI is becoming one of the defining phrases of this cycle. The core promise is simple: your data and the model processing it both stay within national borders, governed by local law. For regulated sectors like finance, defence and healthcare, that is increasingly a hard requirement. Australia has been pushing this agenda hard, partly because of its exposure to both American cloud providers and Chinese technology.
Bob: Europe is further along the sovereign AI path in rhetoric, but mixed in execution. France has been loudest, with Mistral and public sector pilots, and Germany is investing through its own channels. The Australian model, embedding sovereign LLMs inside existing enterprise automation platforms, is practical rather than ideological, and that is what makes it interesting as a template. European procurement officers watching this will note that the integration layer, not the model itself, is where sovereignty actually gets enforced.
Two practical picks to close the episode: Standard Notes and Pocket Casts for calm, private daily use.
Samantha: Now, two things you can actually use today, both free to start, and both deliberately away from the AI hype. First, Standard Notes. It is an end-to-end encrypted notes app that works across Windows, Mac, Linux, iOS and Android. The free tier gives you plain text notes synced across all your devices, with the content encrypted so even the company running the service cannot read them. Users report it as one of the most stable options for people who have grown nervous about where their notes actually live.
Bob: The second pick is Pocket Casts, the podcast app originally built in Australia and now widely used across Europe. It is free on iOS, Android and the web, with a straightforward interface, smart silence trimming, and genuinely useful listening statistics. It does not push algorithmic recommendations the way the big streaming apps do, which many listeners find refreshing. There is a paid tier for extras, but the free version covers everything most people need, including sync between phone and desktop.
Samantha: Both are quick to set up. Standard Notes takes about five minutes to install and start syncing. Pocket Casts imports your existing podcast subscriptions in one step if you export them from another app. Neither requires a subscription to be genuinely useful, which is increasingly rare.
Samantha: Today we covered: the chip industry crossing a trillion dollars in annual sales, the White House ordering instant reporting of major AI incidents, CyberTAN and Marvell's liquid-cooled switch for AI data centres, Apple's new HomeView smart home hub, the alleged bugging incident at OpenAI, Workato's sovereign AI deal in Australia, and two things to try yourself: the encrypted notes app Standard Notes and the podcast app Pocket Casts.
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.