Zhipu Caps GLM Coding Plan Sales Over Compute Shortage

Chinese large language model (LLM) developer Zhipu has announced a temporary limit on sales of its GLM Coding Plan. The company made the decision after the recent launch of GLM-4.7, its new flagship model, put heavier strain on its compute capacity.

GLM Coding Plan is a service that gives developers access to GLM for programming tasks. Since GLM-4.7 came out, the number of users has grown significantly, overloading compute resources, especially at peak hours (15:00 to 18:00 on weekdays). During that window, some users hit errors tied to concurrent-request limits, and the model ran more slowly.

Zhipu apologized to users for the inconvenience and said it has started work to expand compute capacity. To keep the service stable for existing customers, however, it decided to temporarily restrict GLM Coding Plan sales. Daily sales will be cut by 80% from the current level.

The move, though temporary, highlights a growing shortage of compute for training and running large language models. As models grow more complex and powerful, their compute needs rise sharply. That creates challenges for LLM developers in scaling infrastructure and managing costs.

The sales cap may affect developers who use GLM in their projects. It may delay development or force them to look for alternatives. On the other hand, it may also push developers to use available resources more efficiently and optimize their requests to the model.

Overall, the GLM Coding Plan situation shows intensifying competition for compute in AI. Companies working with LLMs will have to invest in expanding their infrastructure and in more efficient algorithms to meet growing demand for their products and services. Similar limits may well appear at other LLM developers until better ways to scale compute are found.

The cap on GLM Coding Plan sales is a temporary measure meant to keep the service stable for existing users. But it is also a signal of the growing compute shortage in AI and of the need for long-term solutions for scaling infrastructure.