Claude Fable 5 Cuts AI Costs: Europe's Infrastructure Gap

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

Bob: And I'm Bob Russell. Today: Anthropic unveils two new Claude models that may reset the AI race; a thirty-five billion dollar deal to build the power plants of AI; Anthropic also overtakes OpenAI as the world's most valuable AI lab; the European angle on what all this means for our cloud bills and chip supply; and to close, two things you can actually try today, the AI coding assistant Claude Code and the open-source photo organiser PhotoPrism. Let's get into it.

Anthropic launches Claude Fable 5 and Mythos 5, claiming a new state-of-the-art across nearly every benchmark.

Samantha: Anthropic dropped two new models yesterday, and this one matters. Claude Fable 5 is the broadly available model, and Claude Mythos 5 is a restricted version aimed at cybersecurity defenders and researchers. Fable 5 is, according to Anthropic, the new top performer on nearly every benchmark, especially in software engineering, knowledge work and scientific research.

Bob: And the numbers behind the numbers are interesting. Payment company Stripe says Fable 5 compressed engineering work that used to take months down to days. The price is ten dollars per million input tokens and fifty dollars per million output tokens. That is less than half what the previous Mythos Preview cost, so Anthropic is actually undercutting itself on price while pushing performance up.

Samantha: There is also a safety story baked in. Anthropic says built-in classifiers route sensitive requests around cybersecurity and bio or chemical topics over to an older model, Claude Opus 4.8. That happens in less than five percent of sessions. External red-teamers spent more than a thousand hours trying to break it, and Anthropic says they did not find a universal jailbreak.

Bob: Available right now through the API, with a phased rollout to paying subscribers running until June twenty-two.

Samantha: The bottom line for European companies: this is a meaningful drop in the cost of serious AI work. If you are running a software team in Amsterdam, Munich or Dublin, that halving of token prices flows straight into your cloud bill. And it shifts the competitive picture in Europe, where most start-ups are renting these models rather than training their own.

Bob: There is a sharper European angle too. Mythos 5 is aimed at cyber defenders, and that lands in the middle of Europe's NIS2 cybersecurity rules, which push companies to detect and respond faster. Tools that help defenders triage threats are exactly what regulators are nudging firms toward. The catch is that frontier models like these still sit on American infrastructure, so Europe gets the productivity boost but stays in the dependency it has been trying to reduce.

Samantha: And that dependency theme runs through the rest of today's episode.

Broadcom, Apollo and Blackstone launch a thirty-five billion dollar platform to build twenty gigawatts of AI computing power.

Bob: Chip designer Broadcom has teamed up with two big American investment firms, Apollo and Blackstone, to set up something they are calling the AI XPV Platform. The opening cheque is thirty-five billion dollars. The goal: roll out more than twenty gigawatts of computing capacity for top AI labs by 2028.

Samantha: Twenty gigawatts is roughly the electricity demand of a small country. The first customers named are Anthropic and OpenAI. The platform will use Broadcom's custom AI chips and networking gear to bring down the cost of training and running frontier AI models, both per token and per kilowatt-hour. The first phase is more than one gigawatt of new capacity for Anthropic, going live mid-2026 in data centres run by a company called Fluidstack.

Bob: What is shifting in the market here is who pays for the build-out. Until now, AI labs and hyperscalers funded their own data centres or leaned on Microsoft and Amazon. Now you have private equity stepping in to underwrite the physical layer of AI, treating compute capacity as a long-term infrastructure asset, like toll roads or pipelines.

Samantha: For Europe, the lens is twofold. First, none of these twenty gigawatts are confirmed for European soil, which deepens the gap with the US on AI infrastructure. Second, our grid operators are already warning about data centre demand in Ireland, the Netherlands and Frankfurt. If similar deals do land here, the conversation about power, water and grid connections becomes very real, very fast.

Bob: And it puts pressure on the European Commission's own AI gigafactories plan, which is measured in single-digit billions, not tens of billions.

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.

Anthropic overtakes OpenAI as the world's most valuable AI lab at nine hundred sixty-five billion dollars.

Bob: Anthropic has hit a valuation of nine hundred sixty-five billion dollars. That makes it, on paper, the most valuable AI lab in the world, ahead of OpenAI. To put that in perspective, that is close to a trillion dollars for a company that did not exist five years ago.

Samantha: The valuation reflects two things at once: genuine progress on the Claude models, and an enormous flood of capital chasing frontier AI. Investors are betting Anthropic's mix of safety framing and high-end performance, the same combination behind yesterday's Fable 5 launch, will keep enterprise customers paying premium prices.

Bob: It is also a signal that the AI market is no longer a one-horse race. A year ago every conversation started with OpenAI. Now buyers in banking, pharma and government are running serious bake-offs between Claude, GPT models and Google's Gemini. That competition is good for customers, including in Europe, because it pushes prices down, as we just saw with Fable 5.

Samantha: The European lens here is uncomfortable. Europe has Mistral in France, Aleph Alpha in Germany, a few promising labs in the UK. None of them are anywhere near nine hundred billion dollars in valuation. The gap with the American leaders is not just widening, it is accelerating.

Bob: And that feeds straight into the policy debate in Brussels and national capitals. If the dominant AI tools used by European hospitals, banks and ministries are built and valued in the US, the questions about sovereignty, data residency and the AI Act get sharper, not softer.

Samantha: One caveat: this is a private-market valuation, based on what investors are willing to pay in funding rounds. It is not a public stock price, and it can move quickly in either direction.

Europe's data centre boom collides with the grid as AI compute deals get bigger and faster.

Bob: Let's pull the day's three big stories together, because there is a fourth angle that does not get enough attention: the physical infrastructure side in Europe. The Broadcom-Apollo-Blackstone deal and the Anthropic valuation both point to one thing, AI is becoming an infrastructure industry, not a software industry.

Samantha: And that has direct consequences for European users, even if the headlines all say Silicon Valley. The Netherlands has paused or tightened approval for new hyperscale data centres in several regions. Ireland's grid operator has openly warned that data centres could consume close to a third of national electricity by the end of the decade. Germany is wrestling with where to put new AI campuses near Frankfurt.

Bob: When investors commit thirty-five billion dollars to twenty gigawatts of capacity, that demand has to land somewhere. If it lands in Texas or Virginia, Europe stays dependent. If it lands here, European grids, water systems and planning rules have to catch up fast.

Samantha: There is also a chip angle. Broadcom's custom AI chips are an alternative to Nvidia's. For European cloud providers like OVHcloud in France or IONOS in Germany, more choice on the chip side could lower costs over time. But they still have to buy through American supply chains.

Bob: The takeaway is that today's AI news is not really about new chatbots. It is about who owns the power plants, the chips and the buildings that AI runs on. Europe is a heavy user of these services, and a much smaller owner of the stack underneath.

Samantha: Industry groups have argued for years that European public investment in AI infrastructure needs to grow by an order of magnitude to keep pace. Today's announcements make that argument louder.

Bob: And it sets up the policy fight for the second half of this year, around the AI Act's implementation and the next EU budget cycle.

Claude Code turns your laptop into a pair-programming partner that can build whole features on its own.

Samantha: To close out, two things you can actually use today. First one is for anyone who codes, even a little. Claude Code is Anthropic's command-line tool that plugs the new Claude models straight into your terminal and your code editor. With the Fable 5 update we just talked about, it has taken a real step forward in writing, fixing and refactoring code.

Bob: What you can do with it: point it at a folder on your laptop with a project, describe what you want in plain English, and Claude Code reads your files, suggests changes, and can run commands to test them. So instead of copy-pasting snippets in and out of a chat window, you let it work directly in your project. Stripe says it compressed months of engineering into days. For a small team or a solo developer, that is a serious productivity shift.

Samantha: To try it, you install it through the standard package manager for developers, then log in with an Anthropic account. It runs on macOS, Linux and Windows through the Linux subsystem. You will need a paid Anthropic plan or API credit, so it is not free, but the entry-level subscription is about twenty dollars a month and gives you a real amount of usage.

Bob: The European angle: if you are running a small software shop in Lisbon, Tallinn or Warsaw, this is the kind of tool that lets a team of three punch like a team of ten. It also runs locally on your machine, which means your source code does not have to live in someone else's web interface, although the model itself still processes the requests in Anthropic's cloud.

Samantha: And one practical tip: start it on a small, well-tested project before letting it loose on your main codebase. Give it permission to edit one folder at a time. You want to see what it does before it touches anything you cannot easily roll back.

Bob: Worth thirty minutes of your afternoon if you have ever wished for a junior developer who never sleeps.

PhotoPrism organises decades of family photos on your own server, with no cloud subscription.

Samantha: Second find is for everyone with a phone full of photos and no plan. PhotoPrism is a free, open-source photo manager you run on your own computer or a small home server. It scans your photo library, tags faces, places and objects, and gives you a clean web interface that looks a lot like Google Photos, but without sending anything to Google.

Bob: It runs on Windows, macOS and Linux, and it works really well on a cheap mini-PC or a Raspberry Pi if you want it always on. You point it at your photo folders, walk away for a few hours while it indexes, and come back to a searchable library. You can type things like "beach 2019" or "my kids in Berlin" and it actually finds them.

Samantha: For European listeners there is a real privacy upside. Your photos stay on hardware you own, which sidesteps the whole debate about American cloud providers and data transfer. If you have been nervous about subscription photo services raising prices or scanning your library, this is a clean alternative.

Bob: Setup takes about an hour if you are comfortable following a tutorial, and there is a friendly community around it. The core version is free. There is a paid membership for some extra features, but most people will not need it.

Samantha: A small weekend project that gets you back in control of twenty years of photos. Hard to argue with that.

Samantha: Today we covered: Anthropic's launch of Claude Fable 5 and Mythos 5, the thirty-five billion dollar Broadcom, Apollo and Blackstone AI infrastructure platform, Anthropic overtaking OpenAI as the world's most valuable AI lab, the European angle on the data centre and grid squeeze, and two things to try yourself: the coding assistant Claude Code and the open-source photo manager PhotoPrism.

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.