Indian Companies Shift to Chinese LLMs as U.S. Token Costs Rise

Indian companies are turning to Chinese large language models to limit artificial-intelligence spending. Models from DeepSeek, Alibaba and Moonshot AI are being used by consumer-technology startups and larger enterprises, according to reporting by Nikkei.

Puneet Kumar, CEO at Mirae Asset Venture Investments India, said several startups he met since mid-2025 rely on open-weight Chinese models that reduce costs by an order of magnitude. These models allow users to download and modify parameters locally, unlike proprietary U.S. systems.

Pricing comparisons illustrate the gap. DeepSeek models accessed through Microsoft in southern India charge 19 cents to $1.74 per million input tokens and 51 cents to $5.40 per million output tokens. Moonshot's Kimi ranges up to 95 cents for input and $4 for output.

OpenAI's GPT 5.5 series lists $5 to $12 for input and $30 to $54 for output.

Vidya Madhavan, founder of dating app Schmooze, said she switched to Alibaba's Qwen models after comparing performance and cost. She described the move as a way to achieve similar results at lower expense. Nikhil Narendran, a partner at law firm Trilegal, said token bills are becoming unsustainable for Indian users.

He noted that locally hosted open-weight models keep data inside India, though security concerns remain. Usage of Chinese models on the Open Router marketplace more than doubled to 25 trillion tokens in the final week of June, exceeding U.S. model usage by 78 percent.

Companies including Coinbase, DoorDash and Airbnb have publicly reported adopting Chinese models.