Arm ships its first chip. Robots prove they're real businesses.

Arm crossed a line this week. For 35 years, the company licensed CPU designs to others. Last week, it announced it's now selling its own chips-starting with Meta, moving into datacentres. The shift matters because it signals something deeper: the gap between ISA ownership and the full system is closing. Arm's CEO Rene Haas describes it plainly: you can't stay a platform company by staying hands-off. The CPU market is about to get more crowded, but Arm sees something the others don't-agentic AI requires token distribution at scale, which means CPU demand is about to explode, not shrink.

When robots become balance sheets

Unitree's IPO filing landed this week with numbers that changed the conversation. A humanoid robot company just proved-on paper-that the economics actually work. Average selling price dropped from $85,000 to $25,000 in two years. Gross margins rose to 59.8% despite that. That's not a science project. That's vertical integration actually paying off. The catch: 73% of their revenue still comes from research and education. Industrial deployment is only 9%. The hardware works. The market? Still early. But the fact that a company can ship 3,700 humanoids and sell 96% of them tells you something's shifted from theoretical to real.

Tools are reshaping how developers work

Claude's computer use feature launched to what Latent Space called its biggest reception ever. Meanwhile, a DevOps engineer with 20 years of experience published something worth reading: LLMs work best as a "very fast junior engineer," not as a replacement. The insight cuts through the hype. You still need domain expertise to supervise the output. You still need to know when to stop iterating and fix it yourself. But when you do know your domain? An LLM handling boilerplate and repetitive work-the Terraform imports, the configuration scaffolding-saves weeks. The pattern holds across tools: they accelerate execution for people who already know what they're doing.

At NVIDIA GTC this week, the theme was physical AI coming of age. Robots are doing real work at scale now-bin-picking, tile-cutting, delivery. Simulation matters. AI training matters. But what mattered most in the demos? Speed. A new robot demo with UR and Generalist came together in days, not months. That's the compounding effect of better models, better tools, and better hardware all moving at once. For the first time, it feels like robotics isn't waiting for the next breakthrough. It's consolidating the last one.

Three separate stories, but they're all pointing at the same thing: the gap between what's theoretically possible and what's commercially viable is closing fast. Arm's betting on it. Unitree proved it. And every developer with access to Claude is living it.