Chinese AI companies are expanding developer ecosystems with open-source models, while U.S. firms stick to closed, API-based approaches, MIT Technology Review reported. A turning point was DeepSeek’s R1 reasoning model released in January 2025. Researchers said Chinese open-weight models accounted for 17.1 percent of global AI model downloads in the year to August 2025, edging the U.S. share of 15.86 percent. Adoption is rising in the Global South despite constraints such as censorship and disputes over distillation.

Cutting-edge AI models can solve complex problems but may be inefficient for routine corporate work, a Databricks executive said. He cited invoice error detection, where top models may correct numbers rather than flagging issues for follow-up. He said advanced models can be strong at coding yet lag in data engineering. Companies are increasingly using smaller open-source models fine-tuned with reinforcement learning to lower costs. Interest in Alibaba Cloud’s Qwen is high, but regulation and compliance limit use.