UK cloud oversight: Bank of England targets AWS & TSMC

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

Bob: And I'm Bob Russell. Today: UK regulators put big cloud providers under direct financial oversight, TSMC announces three new advanced packaging plants in Taiwan, LG steps into the AI server rack business, Meta unveils a new AI model built for autonomous agents, and to close out, two things you can actually use today: the mapping platform ArcGIS with its new AI layer, and a clever open-source podcast tool called Castopod. Let's start in London.

UK regulators start policing the cloud giants that hold up British finance.

Samantha: From today, the Bank of England, along with the Prudential Regulation Authority and the Financial Conduct Authority, has formal power to directly oversee what they call Critical Third Parties. In plain terms, that means the biggest cloud and tech providers underpinning UK banks and insurers are now supervised the way banks themselves are. The named companies include Amazon Web Services, Google Cloud, Microsoft, and Oracle. It's a fundamental shift: regulators no longer just talk to the banks about their cloud contracts, they can now go straight to the cloud itself.

Bob: And the trigger for this is years of near-misses. When a hyperscaler has an outage, half the country's payment apps wobble. British regulators have been arguing for a while that the concentration of critical banking workloads on just a handful of Ay-Double-You-Es and Microsoft regions has become a systemic risk. Under the new regime, those providers face resilience testing, incident reporting duties, and scenario exercises overseen by the Bank of England directly. If they fall short, the regulator can order changes. That's a serious escalation from the current model, where the responsibility sat entirely with each bank's procurement team.

Samantha: The European angle here is significant. The UK is moving ahead of Brussels on formalising this, but the European Union's own Digital Operational Resilience Act, DORA, is running on a parallel track for the eurozone. So for any European bank using the same American cloud providers, expect two overlapping regimes asking similar questions. And for the cloud giants themselves, this is the first time British financial regulators can knock on their door and demand answers, without going through a bank as intermediary. That's a meaningful new lever.

Bob: The bottom line: if you use online banking in the UK, this is designed to make outages rarer and recoveries faster. And it's a template other regulators will be watching closely.

TSMC plans three more advanced packaging plants to feed the AI chip boom.

Samantha: Taiwan Semiconductor Manufacturing has confirmed plans to build three additional advanced packaging facilities in the second phase of the Chiayi Science Park. The site is roughly ninety hectares, and TSMC wants to turn it into a dedicated advanced packaging cluster. This is the less glamorous cousin of chip fabrication, but arguably just as important right now. Advanced packaging is how you stack and connect multiple chips into a single powerful module, which is exactly what Nvidia and others need for their AI accelerators.

Bob: The shortage bottleneck for AI chips has quietly moved from raw manufacturing to this packaging step. TSMC's specific technique, known in the industry as CoWoS, has been sold out for well over a year. Every AI GPU that Nvidia ships depends on it. Three new plants signals TSMC believes demand isn't just a spike, it's the new normal. And Taiwan gets to keep the most valuable part of the AI supply chain firmly at home, even as fabrication itself spreads to Arizona and Germany.

Samantha: For Europe, this matters in two ways. First, the German TSMC fab in Dresden, backed by heavy state subsidies, does not do this advanced packaging. So even chips made in Europe still travel back to Taiwan for the final assembly. Second, it reinforces just how concentrated one link of the chain remains. European industrial policy has focused on fabs, but the packaging gap is now the more urgent strategic question. Industry groups have been pointing out for months that without packaging capacity, European fab investments deliver only part of the resilience story.

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.

LG jumps into the AI server rack business, taking on Taiwan's incumbents.

Bob: LG Electronics is stepping into an unexpected market: AI server racks. The company has been named as a new supplier for Nvidia's next-generation AI computing platform, called Vera Rubin. What LG brings to the table isn't obvious at first glance. It's a home appliance maker. But that turns out to be the point. Modern AI racks generate enormous heat, and cooling them is one of the hardest engineering problems in data centres today. LG has decades of experience with refrigeration, air handling, and thermal engineering from fridges and air conditioners.

Samantha: This diversifies a supply chain that has been almost entirely dominated by Taiwanese firms like Foxconn and Quanta. Nvidia has been pushing to broaden its supplier base, partly to reduce concentration risk, partly to keep pricing competitive. And for LG, this is a significant strategic pivot. The consumer appliance market is mature and margins are thin. Selling high-density cooling systems to hyperscalers, on the other hand, is a rapidly growing business with much better economics.

Bob: For European data centre operators, more competition on rack hardware is genuinely useful. Europe is building AI infrastructure quickly, but energy and cooling constraints are the binding limits, especially in the Nordics and Ireland. A supplier with genuine expertise in thermal management, and enough scale to matter, could ease some of those bottlenecks. It also underlines how the AI boom is pulling in players you would never have expected five years ago. A fridge company is now selling to Nvidia. That's the strangest and most telling detail of this whole story.

Meta launches Muse Spark 1.1, an AI model built to run autonomous agents.

Samantha: Meta has unveiled Muse Spark 1.1, an AI model aimed specifically at running autonomous agents. Meaning, software that can carry out multi-step tasks on your behalf, use tools, interact with computer interfaces, and coordinate what Meta calls sub-agents. The pitch is that instead of asking a chatbot to write you a plan, you hand the task to Muse Spark and it actually executes it: booking things, navigating apps, writing and testing code, checking its own work.

Bob: The technical claim here is that it handles long-running tasks better than existing models. That's been the weak spot for the agent category all year. Models like OpenAI's and Anthropic's have shown promise, but they still get lost in tasks that require dozens of steps. Meta is also pricing this aggressively, positioning it as a more accessible option for developers building agent-based applications. That's a competitive move against Anthropic and OpenAI, which have been dominating the developer market for tool-using models.

Samantha: For European businesses, the interesting question isn't the model itself, it's the regulatory picture around it. Under the European Union's AI Act, autonomous agents that can take real-world actions raise a whole set of accountability questions. Who's responsible when an agent books the wrong thing, or moves money incorrectly? The rules on high-risk AI systems are being interpreted right now, and agent frameworks like Muse Spark are exactly what regulators had in mind when they debated those clauses. So European developers can use it, but they'll need to think carefully about audit trails and human oversight.

Bob: The broader shift is that we're moving from AI that answers questions to AI that does things. That transition is arguably bigger than the chatbot boom itself, because it changes what software fundamentally is. And Meta pricing it low suggests they want to commoditise the market before it fully forms.

The world's biggest mapping event puts AI at the centre of ArcGIS.

Samantha: To close out, two things you can actually try today. First, if you work with maps or spatial data in any form, the Esri User Conference opened today in San Diego, and the headline is a major expansion of AI features inside ArcGIS. ArcGIS is the dominant professional mapping platform globally, used by everyone from city planners to logistics companies to emergency services. The new AI layer is designed to automate a lot of the tedious spatial analysis work: classifying satellite imagery, spotting patterns in geographic data, and generating summaries you'd normally build by hand.

Bob: What makes this a practical find rather than just industry news is that Esri offers free access tiers, including ArcGIS Online for personal use and a substantial student and non-profit programme. So if you've been curious about doing anything with maps, plotting your own data, analysing neighbourhood patterns, visualising travel routes, this is a good week to try it. The AI features that were previously buried inside enterprise contracts are being rolled into the broader platform. Users report that the new natural language interface lets you type questions like "show me flood risk near this postcode" and get an actual map back, rather than having to build queries manually.

Samantha: For European users specifically, ArcGIS Online works with Ordnance Survey data in the UK, and integrates well with national mapping agencies across France, Germany, the Netherlands and the Nordics. It runs in the browser, no installation needed, and the free tier is genuinely usable for small projects. If you've ever wanted to make a proper map of, say, cycling routes in your city, or track something over time on a map, this is the platform professionals actually use. And it's now much less intimidating for beginners thanks to the new AI assistance.

Bob: One caveat: the deeper analytical features still require a paid subscription. But for anyone dipping a toe in spatial data for the first time, the free tier and the new AI helpers make ArcGIS worth an afternoon of exploration.

Castopod is the open-source podcasting platform that keeps your show in your own hands.

Samantha: And the second find today, staying non-AI and useful: Castopod. It's a free, open-source platform for hosting your own podcast, developed by a French team. If you've ever thought about starting a podcast, or you already run one and dislike being locked into Spotify or Apple's ecosystem, Castopod is worth a serious look. You install it on your own web hosting, and it handles the entire feed, statistics, episode management, and even social features through the Fediverse, meaning listeners can follow your show from Mastodon.

Bob: The interesting thing about Castopod is that it treats podcasts the way podcasts were originally designed: as open RSS feeds you control, not as content trapped inside a platform. It supports what's called Podcasting 2.0, a set of open standards including chapters, transcripts, and even listener payments in some formats. Users highlight how straightforward the setup is compared to running your own WordPress site with plugins bolted on. And because it's open-source, there's no monthly hosting fee beyond whatever you pay for your own server.

Samantha: For European creators, the appeal is very concrete: your audience data stays with you, not with a US platform, which matters increasingly for GDPR compliance if you're running a professional show. It works with all the major podcast directories, so listeners on Spotify, Apple Podcasts, or Pocket Casts can still find you normally. Reviewers describe it as the closest thing the podcast world has to what WordPress was for blogging: a proper self-hosted foundation that puts you in control.

Samantha: Today we covered: UK regulators taking direct oversight of the big cloud providers, TSMC expanding advanced chip packaging in Taiwan, LG jumping into the AI server rack market for Nvidia, Meta launching its Muse Spark agent model, and two things worth trying yourself: ArcGIS with its new AI features, and the open-source podcasting platform Castopod.

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.