Zhipu AI caps GLM coding plan sign-ups after demand surge

Zhipu AI, a Beijing-based AI firm, is limiting new sign-ups for its GLM Coding Plan due to performance issues caused by a surge in demand, the company announced on January 21.

The restrictions cap new user sign-ups at 20% of existing levels to prioritize current users amid a computing capacity crunch.

Zhipu’s global operations head said the limits are to free up resources and denied targeted rate limiting.

The company cited the strain to a fivefold increase in traffic after launching its latest model, GLM 4.7, last month.

Industry analysts note that Chinese AI firms face infrastructure constraints, especially in serving inference requests at scale, which hampers their ability to expand globally.

🔗 Source: South China Morning Post

🧠 Food for thought

Implications, context, and why it matters.

The severity of Zhipu’s compute crunch remains unclear without infrastructure details

To judge whether the pause on new sign-ups for the GLM Coding Plan is a short-lived disruption or a longer limit, the scale of Zhipu AI’s inference compute (the chips and servers used to run its AI model for user requests) needs to be spelled out.
Its hardware sourcing also needs to be pinned down, including how much it depends on sanctioned Nvidia chips, domestic substitutes, or black-market gear, since some Chinese AI companies are considering buying Nvidia’s high-performance H200 chips from the black market.
More clarity should be provided on the split between cloud partners and on-premise data centers, plus how the recent Hong Kong IPO proceeds will be directed toward computing infrastructure, to judge how the bottleneck will be addressed.

Chinese AI’s global push could create opportunities for APAC compute providers

International GPU (graphics processing unit) cloud providers can pick up spillover demand from Chinese AI firms with global users when domestic infrastructure runs hot.
APAC operators with ready Nvidia H100 or AMD MI300X capacity in hubs such as Singapore or Hong Kong can move fastest, since some providers publish on-demand GPU instance pricing 1.
Inference-optimization software vendors can also benefit by helping Chinese AI companies cut compute cost per user, which can ease strain on existing hardware while improving uptime.