Rising expenses for premium US AI services like OpenAI’s GPT and Anthropic’s Claude have prompted many global enterprises to pivot towards Chinese open-weight AI models, which offer competitive performance at a fraction of the cost.
- Chinese open-weight models operate at roughly one-fifth the cost of leading US proprietary models.
- Open-weight models now represent nearly 30% of token volume on Vercel’s AI Gateway, tripling since April.
- Major global firms, including Coinbase and leading banks, are integrating Chinese models to reduce AI spending.
What happened
Global businesses are increasingly transitioning away from premium closed-source AI platforms offered by US companies in favor of Chinese open-weight models. For example, Zhipu’s GLM-5.2, which functions at about one-fifth the cost of Anthropic’s Claude Opus 4.8, has experienced a 50-fold increase in daily token volume on Vercel's AI development platform since mid-June 2026. Another Chinese model, DeepSeek’s V4 Flash, has also grown to command over 20% of Vercel’s platform traffic, up from 15% a month prior.
This rapid adoption of open-weight AI is driven by the models’ ability to be downloaded and deployed on local hardware for free, drastically lowering operating costs compared to the pay-per-token structure of proprietary systems. Consequently, companies worldwide are reconsidering their AI expenditure strategies in response to escalating token prices charged by premium US services.
Why it matters
The growing performance and cost-effectiveness of Chinese open-weight models are reshaping global AI sourcing dynamics. According to analysts at Goldman Sachs, these models have reached a tipping point in intelligence capabilities comparable to proprietary US models, prompting a surge in enterprise adoption both domestically within China and internationally. Forecasts suggest daily token consumption for these Chinese models will increase from 350 trillion tokens in 2026 to 4,600 trillion by 2030, with over half of usage expected from users outside China.
This shift poses potential challenges to the valuation outlook for US AI labs and cloud infrastructure providers, as commoditization of large language model intelligence may lead to cost-based competition and thinning margins. However, regulatory and geopolitical barriers are expected to limit Chinese AI penetration in certain sensitive sectors, preserving some differentiation in the US market.
What to watch next
Key indicators to monitor include the continued growth trajectory of international adoption of Chinese open-weight models and the evolving cost structures for AI services globally. Enterprise decisions to integrate open-weight models alongside premium options, as seen with companies like Coinbase experimenting with mixed AI architectures, will provide further insight into market preferences and budget management strategies.
Additionally, regulatory developments in Western markets regarding AI software sourcing and geopolitical tensions could influence the accessibility and attractiveness of Chinese models abroad. The competitive landscape between US proprietary AI providers and Chinese open-source alternatives will play a crucial role in shaping the future commercial deployment and development incentives in the AI sector.