Hello Betamax, Open-weight AI models aren't a charity. Every free release from a Chinese AI lab is a customer acquisition strategy dressed up as an act of generosity, and it has worked well so far. When the US briefly suspended access to Anthropic's Fable 5 in mid-June over national security concerns, it validated the open-source pitch overnight: Why build upon a model that a single government directive can take away? Z.ai didn't waste a moment, releasing its GLM-5.2 model the very next day. Our top story today looks at what's actually happening underneath that narrative, and it's not as clean a win for open-source as it first appears. Third-party inference providers are now underselling some Chinese labs on their own models, in some cases by 20 cents per million tokens. It's a strange position to be in: winning the distribution war while losing out on the economics of it. Every major Chinese AI lab will have to respond to that tension eventually, whether they want to or not. Meanwhile, our second top story is a reminder that not everyone is willing to bet on the models themselves. B Capital just closed a US$500 million fund, and it's deliberately staying away from large language models, finding them "too expensive." Instead, the investment firm is going after the picks-and-shovels layer: robotics, spacetech, and quantum. These are, indeed, two different reads on the same AI model boom: one on who profits from openness, the other on who's avoiding the fight entirely. Glenn Kaonang, journalist |
