TECH

Computing Shortage Forces Chinese AI Firms to Ration Services

Published: Apr. 21, 2026 4:49 p.m. GMT+8

00:00

00:00/00:00

Listen to this article 1x

A severe shortage of computing power is forcing major Chinese artificial intelligence model developers and cloud service providers to throttle services, ration access and hike prices as demand outstrips chip supply.

The computing power crunch underscores the growing mismatch between surging demand for AI applications — particularly resource-intensive coding assistants — and constrained semiconductor supply chains, threatening to throttle the industry’s rapid expansion.

You've accessed an article available only to subscribers

Subscribe today for just $.99.

VIEW OPTIONS

Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.

Subscribe to Save an extra $50. Introductory offer for new readers. Subscribe now.

Share this article

Open WeChat and scan the QR code

DIGEST HUB

Digest Hub

Back

Explore the story in 30 seconds

Chinese AI firms (MiniMax, Moonshot AI, Zhipu AI) face outages, rationing, price hikes due to computing power shortage.
AI agents like OpenClaw increase token consumption tens/hundreds-fold; OpenRouter usage up 7-8x YoY in April.
Chip limits: Nvidia GPU restrictions for China; TSMC 3nm fabs start 2027-2028, 2026 capex $52-56B.

AI generated, for reference only

Who’s Who

MiniMax: MiniMax, a prominent Chinese AI startup, is grappling with overloaded or suspended API services amid a severe computing power shortage driven by surging AI demand and chip supply constraints.

Moonshot AI: Moonshot AI is facing overloaded or suspended API services due to computing power shortages. Its Kimi model has repeatedly restricted user access amid surging AI demand.

Zhipu AI: Zhipu AI faces severe computing power shortages, leading to overloaded API services. On April 13, it refunded Coding Plan users after limiting sales to 20% of prior levels in January and hiking API/package prices in Feb-Mar. CEO Zhang Peng stated AI agents like OpenClaw require tens/hundreds more tokens per task.

DeepSeek: DeepSeek, a prominent Chinese AI startup, experienced a nearly 12-hour service outage in late March amid a severe computing power shortage.

ByteDance Ltd.: ByteDance Ltd.'s Doubao AI model has repeatedly restricted user access amid computing power shortages. The company disabled a phone feature for Doubao during the Spring Festival holiday to conserve chips and prevent failures. ByteDance has also imposed periodic quotas on its cloud services to manage high-frequency usage.

Alibaba Group Holding Ltd.: Alibaba Group Holding Ltd.'s Qwen unit suspended non-essential operations across departments during Spring Festival to free up computing power for AI models. Alibaba Cloud halted sales of a baseline coding package in April, leaving only a frequently sold-out premium package.

Alibaba Cloud: Alibaba Cloud suspended sales of its baseline coding package in April due to computing power shortages, offering only a premium package that frequently sells out. Alibaba's Qwen AI unit rationed resources internally during Spring Festival to prioritize AI model training.

Tencent Cloud: Tencent Cloud has imposed periodic quotas to control high-frequency usage amid severe computing power shortages affecting AI services.

Baidu: Baidu has imposed periodic quotas on its cloud services to control high-frequency usage amid severe computing power shortages affecting AI models.

CITIC Securities Co. Ltd.: CITIC Securities Co. Ltd. reported a seven- to eightfold year-on-year increase in weekly token consumption on API aggregator OpenRouter in April, heavily fueled by Chinese models.

Anthropic PBC: On April 4, the head of Anthropic PBC’s Claude Code announced the subscription service would no longer support third-party tools like OpenClaw to ensure sustainable growth. Inefficient context management in AI agent frameworks caused massive, costly token usage, making the model unprofitable.

Nvidia Corp.: Chinese companies face restrictions on accessing advanced Nvidia Corp. GPUs. Nvidia relies on TSMC for 3-nm chip fabrication, exacerbating global supply shortages as TSMC's capacity lags AI demand, with new fabs not online until 2027-2028.

Advanced Micro Devices Inc.: Advanced Micro Devices Inc. relies on TSMC for advanced 3-nanometer chip fabrication, contributing to tight global chip production capacity that lags behind surging AI demand.

Google LLC: Google LLC relies on Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) for advanced 3-nanometer chip fabrication, amid tight global chip production capacity constraining AI data center expansion. (32 words)

Taiwan Semiconductor Manufacturing Co. Ltd.: TSMC's wafer fabrication capacity is the primary bottleneck for data center expansion, with new plants taking 3+ years to build and reach full capacity. It has three 3-nm fabs under construction in Taiwan, Arizona, and Japan, starting production 2027-2028. Nvidia, AMD, and Google rely on TSMC for advanced chips. 2026 capex to surge 37% YoY to ~$56B for AI demand.

AI generated, for reference only