AI Shift: Alibaba’s Qwen Surges Past Meta & Google as Stripe Bets Big on OpenRouter

Alibaba’s Qwen models just crossed three billion downloads. That officially makes them the most popular open-weight AI models in the world, outpacing both Meta and Google. Last week in episode 142, we talked about Alibaba’s lean model. Today, you're seeing the result. The center of gravity in open-source AI just shifted east. Here’s what else is moving. First, Stripe just spent over seven billion dollars to acquire OpenRouter. That’s a unified API for more than four hundred different AI models. It’s a massive consolidation play. Stripe isn't just processing payments anymore; it's positioning itself as the universal toolkit for developers building with AI. Second, the money keeps pouring into niche infrastructure. An embodied AI company you’ve likely never heard of, Mifeng Technology, just closed a funding round in the hundreds of millions of yuan.

It’s led by China Telecom. They’re building the data pipeline for AI that interacts with the physical world. This is not about chatbots. This is about robots. And third, regulation is no longer a future concept. The EU AI Act officially came into effect on August second. The consequences are already here. On August fourteenth, Anthropic announced its Claude models will now embed text watermarks to comply. They're not alone. The era of unlabeled, untraceable AI-generated content is ending, at least in Europe. Finally, the model wars continue. Google DeepMind just launched Gemini 3.7 Flash, which it’s calling its most intelligent model for coding and agentic tasks. It's another move in the endless chess game for performance leadership, but the board itself is starting to look different.

Okay, let’s go deeper on two of those points, because they represent fundamental shifts. First, that Alibaba number. Three billion downloads isn't just a vanity metric. It's a signal that the global developer community is coalescing around a non-Western, open-weight model family. For years, the default assumption was that leadership would come from a handful of labs in California. That assumption is now broken. Here’s the thing. When a model is open, you can download it, modify it, and run it on your own hardware. You aren't beholden to an API, a usage policy, or a pricing structure set by a single company. This decouples innovation from the control of a few big platforms. Alibaba’s success proves there's a massive global appetite for that freedom.

It also means that the most widely-used AI tools are now evolving in a decentralized way that's much harder to track and control, for better or for worse. The question is no longer just "who has the best model?" It’s now "who controls the most vibrant ecosystem?" Today, the answer to that second question is Alibaba. The second major shift is that funding for Mifeng Technology. Embodied AI. It sounds abstract, but it's brutally physical. For an AI to learn how to operate a robot arm or navigate a warehouse, it needs data of the physical world. Not text from the internet. It needs hours and hours of video, tactile sensor readings, and force feedback. That’s what Mifeng is building. Their "MEgo" product line includes things like grippers and head-mounted devices specifically for collecting this kind of data.

Their stated goal is to produce tens of millions of hours of physical interaction data. This is the new gold rush. While everyone else was scraping the web for text, companies like Mifeng started building the digital mines for the physical world. This funding round isn't just another bet on AI; it's a bet that the next frontier for large models lies COMPLETELY outside the internet. It’s a bet on data that can only be generated by touching, seeing, and moving through actual reality. So what's the through-line here? For all the talk of bigger models and higher benchmark scores, the real movement is happening at the edges. Not in fundamental breakthroughs—the research notes a distinct lack of those right now. The action is in deployment, infrastructure, and regulation.

The industry is getting less abstract. It's getting real. It’s running on hardware all over the world, it's starting to touch physical objects, and it's finally being forced to obey human laws.