Alibaba’s Qwen turned chip restrictions into a multilingual AI advantage

In 2024, Qwen emerged against a difficult backdrop. U.S. export controls had blocked China from Nvidia’s most advanced chips since 2022, while the AI race rewarded ever-larger amounts of computing power. Many observers thought those restrictions would doom Chinese AI, yet Qwen variants began appearing near the top of public benchmarks.

Alibaba’s decisive move was to make multilingual capability a competitive wedge rather than chase only the English-centric frontier. Qwen was built to handle major languages outside most AI training data, including Burmese, Bengali and Urdu, and engineers could access it from almost anywhere. That fit a Chinese market in which local companies needed stable alternatives after U.S. providers cut off API access.

The result was not absolute parity: analyst Dylan Patel said Qwen remained below GPT-4, but close enough to raise eyebrows and often stronger in formal mathematics and multilingual operations. A Qwen-based model topped Hugging Face’s Open LLM leaderboard; Alibaba said its Tongyi Qianwen models had 90,000 corporate clients at launch and later claimed 2.2 million corporate users. Xiaomi also integrated Qwen into a recipe product and mobile assistant.

A Qwen-based model topped Hugging Face’s Open LLM leaderboard, while Alibaba reported 90,000 clients at launch and 2.2 million corporate users.

Export controls made brute-force access to frontier chips a strategic vulnerability.

Chinese companies needed a model with stable domestic access after U.S. API cutoffs and government blocks.

Multilingual and low-resource language performance offered differentiation even before Qwen matched GPT-4.

When a constraint blocks scale, choose an underserved capability—such as low-resource language support—where customers value usefulness before perfect frontier-model parity.

Qwen’s breakthrough did not end the gap with U.S. frontier models, but it changed Alibaba’s position from a Chinese incumbent adapting AI into a credible global model contender. Its multilingual strengths and access model gave developers a reason to choose it on usefulness and availability, not just benchmark rank; the company’s claimed adoption and Xiaomi integrations showed that the wedge could travel into products.