The Company Behind the GLM Models Just Raised 5 Billion Dollars in a Day
Z.AI, formerly Zhipu, raised about 5 billion dollars through a discounted share placement and zero coupon convertible bonds. Sixty percent goes into the next generation of GLM models and the computers to train them.
Z.AI, the Chinese lab formerly called Zhipu and the maker of the GLM family of models, raised roughly 5 billion dollars on Monday in a single coordinated move. About 2 billion came from placing 21.97 million new shares at 714 Hong Kong dollars each, roughly ten percent below Friday’s close, which is the discount you offer when you want the money quickly. The other 3 billion came from convertible bonds due in September 2027.
Those bonds are worth a sentence of translation. A convertible bond is a loan that the lender can later swap for shares instead of asking for the money back. These pay no interest at all, and were priced to yield between minus 0.5 percent and zero, meaning buyers are accepting slightly less than they put in, in cash terms, in exchange for the right to convert at 892.50 Hong Kong dollars, about 25 percent above what the share buyers just paid. In plain terms: investors gave up interest entirely because they expect the share price to go up more than that. The company says about 60 percent of the proceeds go into research on the next generation of GLM foundation models plus the training and inference computers that requires, 15 percent into expansion and possible acquisitions, and 25 percent into general working capital. It is Z.AI’s second large raise in two months.
What is behind this
GLM matters to people outside China more than most Chinese model families, because Z.AI has consistently released open weight versions, models whose files you can download and run yourself. Alongside DeepSeek, Qwen and Kimi, that has made the open end of the market largely a Chinese affair, which in turn is why a European developer who wants a capable model without an American API key usually ends up running Chinese weights. So a war chest aimed at the next GLM generation is not a distant story. The raise also says something about where the cost sits: 60 percent of 5 billion dollars going to models and the machines to train them is a reminder that the expensive part of AI is not ideas, it is electricity and hardware.
What this means for you: Nothing today, and possibly something concrete in a few months, in the form of another free, downloadable model that is better than the last one. If you or your company run models locally, or use a hosting provider that serves open weights, this is the pipeline that keeps that option alive and cheap. One honest caveat to keep in view: open weights and open company are different things. You can inspect the model you download, but you cannot inspect what it was trained on, and where the lab sits determines which rules it answers to. For sensitive work, judge the weights on your own tests rather than on the label.
Sources
OpenAI Puts Off Its Stock Market Debut, and Says Safety Is the Reason
Sam Altman told Fortune on 12 September that going public this year would be an ill-advised moment, pushing a listing once discussed at around a trillion dollars into 2027. Anthropic is still aiming to list before year end.
