AI Signal Watch: Alibaba Unleashes a 2.4T Parameter Model, Gemini Hits 1B Users

Alibaba just released the weights for a 2.4 trillion parameter model. It’s now one of the largest open models on the planet, and it changes the game for what you can build without asking a lab for permission. In episode 138, we saw Meta’s Muse Glimmer escalate the race for capable local models. Today, a Chinese tech giant just reset the board on sheer scale. That’s the lead. Here are the other signals moving the market. First, Google’s Gemini AI assistant just crossed one billion monthly active users. Sundar Pichai called it their "fastest growing product ever." That makes Gemini the fourteenth product at Google to hit the billion-user mark, and it did it faster than any of the others. The adoption curve here is vertical. Second, the money is still moving. A Swedish AI coding startup called Lovable just raised four hundred million dollars in a Series C, pushing its valuation to thirteen-point-three billion dollars.

For context, they are doing five hundred million in annualized revenue. This isn’t a bet on future tech. It’s a bet on a real, scaling business. Third, a different kind of money move. A firm called Thrive Holdings just raised two billion dollars specifically to acquire AI-focused service firms. This is the first major signal of consolidation in the services layer—the consultants and integrators who actually get AI working inside big companies. And finally, two strategic moves from the big labs. Anthropic just locked in a TWENTY-year compute deal with Riot Platforms. That’s a nine-point-one billion dollar commitment for one hundred ninety-one megawatts of power through 2048. They are securing their ability to train models for decades. And Google just reorganized DeepMind, moving Demis Hassabis up to Chief Scientist of Alphabet to focus squarely on AGI, while promoting Koray Kavukcuoglu to run the Gemini model teams.

Okay, let's go back to that Alibaba model. It's called Qwen3.8-2.4T-A95B. The name is a mouthful, but the number that matters is 2.4 trillion. That’s the number of parameters. Releasing the weights means anyone with enough hardware can now download, inspect, and run a model of that scale. This isn't just another seventy-billion parameter model you can fine-tune. This is a direct shot across the bow at the closed, proprietary frontier models from the major US labs. Alibaba is claiming performance that rivals models like Fable 5. Now, here's the catch. These claims are from Alibaba. The model's actual performance and its adoption outside of China are still completely unverified. We don't know how it truly stacks up on independent benchmarks. But that almost doesn't matter. The signal is what's important. A top-tier Chinese technology company has decided that its strategic advantage lies in arming the open-source community with a frontier-scale model.

This move is designed to build an entire ecosystem around their architecture, creating a powerful alternative to the Western AI stack. The era of US-only dominance in open-weight models is OVER. Now let's follow the money, because it tells the other half of the story. That four-hundred-million-dollar check for Lovable is more than just a big number. It's validation of a very specific strategy. Lovable isn't trying to build a general-purpose AI. It's an AI coding assistant, and it's absolutely dominating its niche. The company hosts sixty million projects. It attracts nine hundred million visitors a month. And as of June, it was booking half a billion dollars in annualized revenue. Here’s what’s different: Lovable built its own proprietary AI model but also lets users access other frontier models through its platform. They have deep technical sophistication AND a massive user base.

This is the blueprint for a successful vertical AI company. They solved a specific, high-value problem and built a product so good that developers are willing to pay for it at scale. Then you have Thrive Holdings raising two billion dollars. They aren't building a product at all. They're buying the companies that build AI solutions for everyone else. This is the classic pick-and-shovel play. As every company on Earth scrambles to implement AI, a huge market emerges for the service firms that can do the complex integration work. Thrive is betting that this services layer is about to consolidate, and they've raised a war chest to be the one doing the consolidating. So you have two massive flows of capital. One is going to hyper-focused, high-growth products like Lovable. The other is going to the infrastructure and services layer that supports the entire ecosystem.

The easy money for building general foundation models is gone. The smart money is now moving downstream. The entire landscape is bifurcating. On one side, you have massive, state-backed models like Qwen being released into the open, aiming to create a global standard. On the other, you have hyper-focused, capital-intensive private companies like Lovable capturing immense value in specific markets. The question is no longer just about who has the biggest model. It's about which ecosystem wins—the open, distributed network or the closed, profitable product. Today, both of those camps just got a lot stronger.