DMA Forces Google to Open Android to AI Rivals

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

Bob: And I'm Bob Russell. Today: Brussels forces Google to open up Android and share search data with AI rivals, Google's flagship AI model slips its deadline for a third time, a French-backed non-profit builds a public AI infrastructure for everyone, Hyundai moves to full ownership of Boston Dynamics, and to close out, two things you can actually use today, the offline knowledge app Kiwix Reader and the private read-later service Wallabag One. Let's start with the biggest one.

Brussels orders Google to open Android and share search data with AI rivals.

Samantha: The European Commission has come down hard on Google. Under the Digital Markets Act, Brussels is now requiring the company to open up its Android ecosystem to competing AI assistants, and to share parts of its search data with rival AI developers. We're talking about two billion Android phones worldwide, so the reach here is enormous.

Bob: And the timing matters. This lands at a moment when Google is already under pressure over the delays with its next big AI model, which we'll get to in a minute. So competitors are being handed a real opening. Until now, if you wanted to build a serious AI assistant on Android, you were essentially building on Google's terms, with Google's defaults in the way. That changes. Rivals will be able to plug in more directly, and they get access to slices of search data that were previously locked inside Google's own systems. That's the kind of raw material you need to train and refine a competitive assistant.

Samantha: The Commission frames this as a competition remedy, not a punishment. The point is to make sure the AI market that grows on top of smartphones is not simply a repeat of the search market, where one company ended up with almost everything. And this is binding, not a proposal. Google has to comply.

Bob: For European users, the practical effect could be significant over the next year or two. Imagine picking your default AI assistant on a new Android phone the same way you pick a browser. That's the direction Brussels is pushing. For European AI developers, from Mistral in France to a long tail of smaller startups, the data-sharing piece is arguably even more important than the Android access. Search behaviour is one of the most valuable training signals you can get, and until now it was a moat.

Samantha: Of course, the details will decide how meaningful this really is. What counts as a fair share of search data, how it's anonymised, how quickly Google actually rolls out the technical changes. Analysts point out that similar DMA orders around app stores took months to translate into anything users could feel. But the signal to the market is loud. Brussels is treating AI distribution as the next front in platform competition, and it's willing to move before the market is fully locked in.

Bob: It also raises the stakes for the American response. Washington has been reluctant to intervene in the AI market this early. Europe is now setting a template that regulators in the UK, Japan, and even parts of Asia will study closely. Whether you like the DMA or not, it is now shaping global AI policy in a way that goes well beyond European borders.

Google's next flagship AI model slips its deadline for the third time.

Samantha: Staying with Google, but a very different story. The company's next flagship AI model, its top tier system, has reportedly missed its release target for the third time. The internal date was July seventeenth, and it came and went. Now, according to reports, Google is considering pushing out a lighter, faster version as a stopgap, while the big model gets more work.

Bob: And investors noticed. Alphabet shares dropped around four percent on the news, which is a meaningful reaction for a company that size. There's still no official model card, no pricing, no benchmarks. In the AI world, when a top lab goes that quiet for that long on its headline product, people start filling in the blanks themselves. Some of that speculation is fair, some of it isn't, but the silence itself becomes the story.

Samantha: The competitive backdrop makes it worse. Anthropic has been shipping updates on a steady cadence. OpenAI is doing the same. And Chinese labs have been putting out capable open-weight models that are eating into the mid-tier of the market. Every month Google's flagship slips, the window for it to be the clear leader narrows. A stopgap release helps keep developers engaged, but it also tells the market that the main event isn't ready.

Bob: For European businesses, this is more than gossip about a Silicon Valley product cycle. A lot of European companies have been standardising on Google's AI stack, particularly through its cloud services. If the flagship model is delayed and underperforming when it lands, procurement teams start asking hard questions. Do we hedge with Mistral, with Anthropic through AWS, with an open-weight model we can run ourselves? Those conversations are already happening.

Samantha: And it feeds directly into the previous story. Brussels is forcing Google to open the door to rivals on Android just as Google is stumbling on its own product timeline. If you're a European AI startup, that's a rare alignment. More access to data, more access to distribution, and a market leader that looks less untouchable than it did six months ago.

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.

A French-backed non-profit is building a public AI infrastructure open to everyone.

Bob: Now to something quite different in spirit. A non-profit called Current AI is trying to build what it describes as public AI infrastructure, free and open to everyone, roughly the way the early web was. It was founded in February 2025 by Martin Tisne, and it has now raised around four hundred million dollars.

Samantha: The funding mix is interesting. The French government put in a hundred million dollars. The Ford Foundation and the MacArthur Foundation are involved, and so are some private-sector names, including DeepMind and Salesforce. That combination of public money, philanthropy, and industry is unusual, and it's a deliberate attempt to keep the project from being captured by any single sponsor.

Bob: The goal is essentially to make sure that basic AI capabilities, translation, speech recognition, useful assistants, don't stay locked behind a handful of American and Chinese providers. Current AI has already introduced a device called Suno Sutra, which delivers offline AI services in twenty-two Indian languages. That's meaningful, because a huge share of the world speaks languages that big commercial models handle poorly or not at all.

Samantha: For Europe, the French involvement is the headline. It positions Paris as a serious player in the public-interest AI conversation, alongside its support for Mistral on the commercial side. Critics note that four hundred million dollars is not a lot compared with the hundreds of billions being poured into AI infrastructure in the United States. Industry groups have argued the number would need to be far higher to shift the balance in any structural way.

Bob: That's the fair pushback. Still, the model matters as much as the money. If Current AI can show that a genuinely open, multilingual AI layer can be built and maintained outside the big commercial labs, it sets a template. Universities, public broadcasters, health services, all sorts of European institutions could plug into that layer instead of paying per query to a hyperscaler. It's early, but the direction is worth watching.

Hyundai moves to full ownership of robotics pioneer Boston Dynamics.

Samantha: Onto our non-AI story, though robotics obviously touches AI at the edges. Hyundai Motor Group is buying out SoftBank's remaining stake in Boston Dynamics and moving to full ownership. SoftBank exercised what's called a put option, and reports value the remaining ten percent stake at around three hundred twenty-five million dollars.

Bob: This closes a chapter that started back in 2021, when Hyundai bought a majority stake and SoftBank kept a slice with an eventual IPO in mind. That IPO deadline is now gone. Boston Dynamics stays private, fully inside the Hyundai group. And it comes right after the company unveiled its production version of the Atlas humanoid robot at CES, with units already committed for real deployments this year.

Samantha: The strategic logic for Hyundai is pretty clear. Car manufacturing is one of the most robot-intensive industries in the world, and the next generation of factory automation looks a lot more like a walking, flexible machine than a fixed robotic arm. Owning Boston Dynamics outright means Hyundai controls the roadmap, the pricing, and crucially, who else gets access.

Bob: That last point matters for Europe. Volkswagen, Stellantis, BMW, all of these companies are looking hard at humanoid robots for their assembly lines. If Hyundai decides to keep the best Atlas units for its own factories first, European carmakers will have to look at alternatives, from XPENG in China, Unitree also in China, or Figure and Agility Robotics in the US. Reports out today suggest the humanoid robotics industry has hit an inflection point, with verified revenue-generating deployments now confirmed in automotive, logistics, hospitality, and even a small number of home environments.

Samantha: And the gap between leading and trailing companies is widening based on actual deployment counts, not just demos. That's the shift. For years humanoid robots were a stage show. Now they're being counted on balance sheets. For European industrial policy, the question is whether the continent wants to buy those robots, or build them. Right now, the honest answer is mostly buy.

Bob: And to close out, two things you can actually use today. Both non-AI, both free, and both genuinely useful.

Kiwix Reader lets you download entire libraries of knowledge for offline use.

Samantha: The first one is called Kiwix Reader. It's a free, open-source app that lets you download huge chunks of the internet and use them completely offline. The most famous package is the full English Wikipedia, all of it, images and text, in one file you can store on your laptop or your phone. But it goes well beyond that.

Bob: You can grab offline copies of medical references, coding documentation, Project Gutenberg's book collection, language learning materials, TED talks, and educational content in dozens of languages. Once it's on your device, no internet is required. Users often describe it as carrying a small library in your pocket, which sounds dramatic until you actually try it on a long train ride through the Alps with no signal.

Samantha: It works on Windows, Mac, Linux, Android, and iOS. The app itself is a few megabytes. The content packages range from a few hundred megabytes for something small, like a medical first-aid guide, to around a hundred gigabytes if you want all of Wikipedia in full. Most people start with a slimmer version, maybe just the articles without images, which is closer to fifty gigabytes, or a specific topic package.

Bob: It's maintained by a Swiss non-profit, which is worth noting. The project was originally built to give schools in areas with limited internet access to Wikipedia, and it has since grown into a general offline knowledge tool. Reviewers praise how simple it is. You install the app, you browse a catalogue, you download what you want, and it just works. No account, no subscription, no tracking. If you travel, if you hike, or if you just want a backup that survives a bad connection, it's genuinely handy.

Samantha: There's also a practical resilience angle. Teachers, doctors working in remote areas, and journalists in restrictive countries all use Kiwix precisely because it doesn't need the open web to function. For everyday users, the appeal is simpler. Sometimes you just want to look something up on a plane without paying for onboard WiFi.

Wallabag One saves the articles you meant to read, on your own terms.

Bob: And the second one is called Wallabag One. It's a read-it-later service, so you save articles from the web to read afterwards, but this one is European, privacy-friendly, and open source. Think of it as an alternative to Pocket, which recently shut down and left a lot of people looking for a new home for their reading list.

Samantha: The core idea is simple. You install a browser extension, or use a share button on your phone, and any article you save gets stripped down to clean text and pictures. No ads, no cookie banners, no pop-ups begging you to subscribe. You can read it later on any device, and you can tag and search across everything you've saved.

Bob: The hosted version is called Wallabag One, and it's run by the French team behind the project. It costs a small annual fee, but you can also self-host the underlying open-source software for free if you have the technical inclination. It works with iOS, Android, and any modern browser. Users often mention how nice it is that articles saved years ago are still readable, even when the original web page has disappeared.

Samantha: And because the code is open, you can export everything at any time. No lock-in. For people who read a lot of longform journalism, or who are building up a personal research archive, that portability really matters. Today we covered: Brussels ordering Google to open Android and share search data, Google's flagship AI model slipping again, Current AI building an open public AI infrastructure with French backing, Hyundai taking full ownership of Boston Dynamics, and two things to try yourself: Kiwix Reader for offline knowledge, and Wallabag One for saving articles the private way.

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.