Demand is too hot! KNOWLEDGE ATLAS AI is in urgent need of computing power and has implemented a "purchase limit": the GLM programming plan will only sell 20% daily, with priority given to old users

Due to the surge in users following the release of the latest large language model GLM-4.7, KNOWLEDGE ATLAS AI has encountered a serious computing power bottleneck and has had to urgently implement throttling measures. The company announced that starting from January 23, it will significantly reduce the daily new subscription volume of its programming assistant service "GLM Coding Plan" to 20% of the original level, and promises to prioritize the access experience of existing users. The increase in users has led to frequent concurrent throttling errors and a significant decrease in response speed during peak hours on recent weekday afternoons. KNOWLEDGE ATLAS AI explained that this is a temporary resource strain caused by rapid growth. This AI company, which just completed a high-profile IPO in Hong Kong this month, is directly competing with international leaders such as OpenAI and Anthropic. Implementing throttling measures in response to sudden traffic peaks is not an isolated case in the rapidly growing AI industry. Last year, DeepSeek also limited access to its API services to manage server capacity. Computing Power Strain Forces Throttling Measures KNOWLEDGE ATLAS AI confirmed through an official WeChat statement that the throttling measures for its GLM programming assistant will officially start at 10 AM on January 23, with new quotas released daily at the same time thereafter. The company stated that the automatic renewal of existing subscription users will not be affected by this adjustment, but did not specify when the throttling measures would end. GLM Coding Plan is an AI programming assistant service that competes with Claude. KNOWLEDGE ATLAS AI explained that due to the surge in users, there have been service quality fluctuations such as response delays and concurrent errors during peak hours. The implementation of throttling aims to prioritize the computing resources and user experience of existing users. Industry Facing Capacity Challenges Implementing throttling measures to cope with the capacity pressure caused by the surge in users has become a common phenomenon in the early stages of high growth in the AI industry. Last year, DeepSeek, which became a global phenomenon, also limited open access to its API services due to tight server resources. Behind such "throttling" actions, highlights the temporary contradiction between the explosive growth in demand for AI technology applications and the speed of underlying computing power infrastructure construction. As services such as large language models and programming assistants rapidly gain popularity, stable and scalable computing power supply has become a key constraint for companies' large-scale expansion. For the market, the computing power bottleneck not only confirms the strong terminal demand but also reveals the practical challenges faced by AI companies in transitioning from technological breakthroughs to stable services in their operational processes. Effectively balancing the rapid growth of user scale with sustainable control of service quality and costs has become a core strategic issue that leading companies in the industry need to address