Anthropic, ChatGPT, and the New AI Regulation
Anthropic keeps buying compute, the EU classifies ChatGPT as a search engine, and new tools show where the AI market is headed.
Inhaltsverzeichnis
Today we’re looking at three major AI topics that together show pretty clearly where the market is shifting right now: more compute, more regulation, more product pressure. On top of that, there are a few product updates and a quirky map-service moment that shows that even old software can still be very much alive.
If you want to know how AI infrastructure, OpenAI/ChatGPT, and European digital policy are currently influencing one another, you’re in the right place today. And yes: the bill for the AI boom is not getting smaller. Quite the opposite.
🤖 Anthropic secures 35 billion dollars for AI compute
According to heise, Anthropic apparently wants to invest another 35 billion US dollars in AI compute. That’s not just “more servers,” but another signal that frontier models are now primarily an infrastructure business. Anyone who wants to build the best models needs not only strong research, but above all power, chips, cooling, and a lot of capital.
The Nvidia angle is also interesting: if Anthropic scales up this massively, demand for GPUs stays high — not just in the short term, but structurally. For the market, that means AI growth continues to depend heavily on hardware supply chains. For companies that want to use or train LLMs themselves, the message is clear: compute is no longer just a technical variable, but a strategic bottleneck. Welcome to the era in which compute time is almost a currency.
🔎 The EU classifies ChatGPT as a very large search engine
The European Commission is treating ChatGPT for the first time as a very large search engine under the Digital Services Act. That sounds like legal fine print, but it’s actually a pretty big deal: starting at at least 45 million monthly users in the EU, additional obligations apply. Among other things, OpenAI must provide risk assessments, transparency reports, and an ad archive — by the end of 2026.
Why does this matter? Because it reveals a new understanding of AI products: for regulators, ChatGPT is no longer “just a chatbot,” but an information and discovery service with search-engine-like reach. That could become a precedent for other AI assistants as well. At the same time, it remains unclear whether the Commission could indirectly also demand access to training data — legally, that is still disputed.
For you, this means: if you build or use AI products in Europe, you shouldn’t wait until the letter from Brussels is already in your inbox before thinking about regulation. The rules are being tightened right now.
🛠️ Tool tip of the day: Empirik for predictive IT operations
Empirik is a new startup with 21 million dollars in funding that wants to predict outages in IT infrastructures before they happen. The idea: don’t just monitor problems, but detect and mitigate them early — basically a kind of “Cursor for infrastructure,” only for ops instead of code.
This is especially interesting for companies with complex cloud and AI stacks. The more systems, agents, and services interact, the more important forecasting becomes compared with reactive firefighting. If traditional monitoring tools feel too much like a rearview mirror to you, this is exactly the category you should take a look at. #
✨ Anthropic Fable 5.1 gets cheaper and less restrictive
With Fable 5.1, Anthropic has released a new version that apparently costs fewer tokens while also triggering fewer false positives in safety filters. It’s a nice example of how model products are evolving right now: not just bigger and “smarter,” but also more efficient and more usable in everyday practice.
Why is this important? Because restrictions and costs are often the hidden brakes on real-world AI applications. A model may look impressive on paper — but if it’s too expensive in production or keeps blocking legitimate requests, it doesn’t help much. Releases like this are therefore often more relevant for product teams than the next big benchmark win. For many companies, what matters in the end is not whether a model is 0.3 points better, but whether it fits neatly into workflows. Unspectacular. And exactly for that reason, important.
🎨 Google is apparently building an AI alternative to Canva
With Google Pics, Google is pushing further into the creative software market. Instead of designing the classic drag-and-drop way, you’re supposed to get to the result more through prompting. This is a direct attack on tools like Canva and Adobe — just with a much stronger AI-first DNA.
The exciting question is not whether such a tool can exist. The question is whether it actually simplifies the workflow or just replaces the design interface with prompt management. For beginners, it naturally feels magical: “Make me a social graphic for LinkedIn” instead of hunting through five layers and three font sizes. For teams, it can still be valuable if fast iteration matters more than pixel-perfect control. Google is playing to its strengths here: distribution, integration, and AI expertise. Whether that will be enough to succeed in the creative market remains to be seen.
🗺️ MapQuest becomes a small internet star over Lake Ontario
The map service MapQuest is currently getting attention for a rather unusual reason: it refuses to rename Lake Ontario. What started as a small naming dispute quickly turned into a viral moment that sent the 30-year-old service climbing in the app rankings.
It’s quirky, but also instructive: in a time when digital products often seem interchangeable, a clear stance can suddenly generate reach. Of course, that’s not a sustainable growth model for every app — and probably not the reason you choose your map provider. But it shows how political and cultural issues can directly affect product perception, even in supposedly neutral services. And sometimes a map label is enough to create a new conversation topic. The internet never forgets. It just reorganizes.
Want to avoid missing any news? Subscribe to the newsletter