China’s open AI advantage may not last forever

Chinese open-source AI models like Qwen, DeepSeek, and Kimi are gaining popularity among Indian startups as a cost-effective alternative to American models. Indian companies are switching to these Chinese models to reduce AI costs by an order of magnitude, according to a report in July 2026. This shift is driven by five reinforcing factors: cost, prestige, commoditization, capital, and infrastructure.

China is expected to start releasing its frontier open-weight models around late 2028. These models, such as DeepSeek's R1 model trained for $294,000, are more affordable than those developed by American companies like OpenAI or Anthropic. Chinese AI companies also benefit from prestige, as demonstrated by DeepSeek's release in January 2025, which caused a significant drop in U.S. tech stocks.

Additionally, China's open-weight models help commoditize the market, making it harder for American labs to charge premium prices for their proprietary weights.

China's strategy relies on capital and infrastructure, with the government subsidizing strategic sectors and driving AI adoption through energy, cloud, and physical infrastructure. Alibaba's cloud revenue increased by 34% year-on-year due to providing Qwen for free. However, three conditions must be met before China restricts access to its frontier models: consolidation, lock-in, and saturation.

China is likely to adopt a graduated restriction approach, starting with API access to frontier models and releasing weights after a six-month embargo. Smaller models will remain free as an on-ramp. Preferential access for World Artificial Intelligence Cooperation Organization (WAICO) members will also be considered. India should leverage the open ecosystem while preparing for potential disruptions in the future.