Big week for AI money and AI muscle. Europe's throwing billions at chip factories, a private equity style buyout just grabbed a piece of the open source stack half of AI runs on, and Google's robots can now move like they've got a body instead of just a claw. Let's get into it.
Europe wants its own AI chip factories
The European Commission opened its call for AI Gigafactory proposals this week. They're backing it with 10 billion euros in public money, about 11.4 billion dollars, and expect that to pull in enough private cash to hit 30 billion total once everything's built out. Seven sites are planned, each one hosting at least 100,000 AI chips. That's roughly four times the scale of a typical EU data center today. Bidding closes November 12 and awards land sometime around mid-2027.
Robert's take: Europe's been watching the US and China build out compute like it's a national security project, which honestly it kind of is at this point. Ten billion in public seed money to try and pull thirty billion total isn't nothing, but building a chip factory takes years and this thing doesn't even pick winners until next year. By the time chips are humming in these gigafactories, the AI landscape will look completely different. Still, better late than never. Nobody wants to be renting compute from somebody else's backyard forever.
Nscale just bought the company behind Ray
Nscale signed a deal to acquire Anyscale, the company that stewards Ray, the open source framework a huge chunk of the AI world uses to scale up training and inference jobs. Bloomberg's putting the price around 1.65 billion dollars. Anyscale's roughly 200 employees move over to Nscale, but the Anyscale brand keeps serving its existing customers on its own.
Robert's take: This is one of those deals that won't make headlines outside the AI infrastructure crowd, but it matters more than it looks. Ray is plumbing. It's the stuff that makes distributed training actually work without every company reinventing the wheel. When infrastructure companies start buying up the plumbing providers, that tells you compute and orchestration are becoming the real battleground, not just whose model scores best on a benchmark. Keep an eye on who owns the pipes, because that's where the leverage ends up sitting.
Google's robots can move their whole body now
Google DeepMind rolled out Gemini Robotics 2 this week, and the headline feature is what they're calling whole body intelligence. Instead of a humanoid robot just fumbling around with its arms while the rest of it stands there like a mannequin, this version coordinates the whole body to do more complex physical tasks.
Robert's take: This is the part of AI that finally leaves the chatbot and moves into your garage. Everybody's been impressed by language models writing emails, but the real unlock is getting a machine to walk over, bend down, and pick something up without falling on its face. Whole body coordination sounds like a small technical detail, but it's the difference between a robot that's a party trick and one you'd actually trust to load a dishwasher. We're still a ways off from that being cheap or common, but this is a real step, not just a demo reel.