Global Firms Shift to Cheaper Chinese AI Models Over US Rivals

Global businesses are increasingly moving away from expensive, proprietary American artificial intelligence software such as OpenAI's GPT and Anthropic's Claude. This shift favours cheaper Chinese open-weight models, now offering comparable performance. The change highlights a growing consideration of cost versus capability in AI deployment.

Credits: Cloudinary

Recent Vercel data indicates a shift. Zhipu's GLM-5.2, costing approximately one-fifth of Anthropic's Claude Opus 4.8, saw its daily token volume surge 50-fold in recent weeks. DeepSeek's V4 Flash, a streamlined version of its flagship V4 Pro, became the largest model by volume on the gateway, capturing over 20 per cent of traffic on a recent day, up from 15 per cent in a previous period.

Open-weight models now account for 29 per cent of token volume on Vercel's AI Gateway platform, nearly tripling their share over several months. This growth signifies a changing approach in how companies procure AI. Proprietary models require premium cloud subscriptions and charge per token. But open-weight models permit organisations to download code free and run it on local hardware. Businesses are re-evaluating tech spending due to the improving capabilities of these options.

A recent Goldman Sachs report stated Chinese open-weight models are achieving critical intelligence performance. This drives significant enterprise adoption domestically and globally. The investment bank predicted daily token consumption of these Chinese models would climb from 350 trillion this year to 4,600 trillion by 2030, with international users making up 55 per cent.

Interest in open-weight models has grown among larger global enterprises because rising AI token costs have become a significant concern, according to a UBS report published recently. The report cited an unnamed large global bank whose chief technology officer confirmed their institution commenced hosting open models from Alibaba Group Holding's Qwen series. This strategy manages token spending and balances the use of premium models like Claude. Alibaba owns the South China Morning Post.

Corporate entities in other industries are adopting similar methods. Brian Armstrong, co-founder and CEO of Coinbase, the largest cryptocurrency exchange in the United States, posted on X recently that his firm was experimenting with setting open-weight choices as default models. These options include GLM 5.2 and Chinese AI lab Moonshot's Kimi 2.7. And the strategy aims to keep AI expenditure stable, allowing internal engineers, or devs, to select various models depending on the task.

This evolving preference for Chinese open-weight models questions the future valuation of US AI labs and cloud giants. Torsten Slok, Apollo Global Management chief economist, wrote in a recent analysis that if Chinese models expand and token prices fall, anticipated cash flows for hyperscalers might prove overly optimistic. Dong Chen, Asia chief investment officer at Bank J. Safra Sarasin, commented that Chinese open-source model progress indicates LLMs' "raw intelligence" will likely become increasingly commoditised.

Mr Chen added that companies creating LLMs would need to compete strictly on cost, suggesting the standalone business could become a low margin industry. Still, many observers contended that Chinese open-weight model uptake would have a restricted effect on the US market. New York based corporate spending platform Ramp stated recently that Chinese AI adoption among American businesses remained very low, concentrated among highly AI intensive firms.

Daniel Yue, assistant professor at the Georgia Institute of Technology's Scheller College of Business, observed that Anthropic and OpenAI's brands have only strengthened recently. He noted many companies strongly preferred to engage with US providers through direct contracts and service agreements. But, US AI labs will maintain their position despite Chinese competition, according to Dong Chen. Geopolitical fragmentation and Western regulations will likely establish a barrier against Chinese models in sensitive sectors, though Chinese models are prepared to succeed in more cost-conscious market segments.

• Global businesses are increasingly switching to cheaper Chinese open-weight AI models due to rising costs of US proprietary software.

• Chinese models like Zhipu's GLM-5.2 and DeepSeek's V4 Flash have seen substantial increases in token volume and platform traffic.

• Analysts predict a significant rise in Chinese open-weight model adoption and token consumption globally by 2030.

• Major financial institutions and cryptocurrency exchanges are exploring or adopting open-weight models to manage AI expenditure.

• Despite arguments for continued US market dominance due to brand strength and regulations, cost-conscious sectors may increasingly favour Chinese alternatives.

Source: SCMP