The global AI landscape seemed defined by a sharp divide: China championed open-source accessibility and rock-bottom prices, while the US dominated the market with powerful - and far pricier - proprietary models. Recently, however, those boundaries have begun to blur.
While US tech giants including Nvidia and OpenAI have rallied behind open-weight models and warned against restrictions, there have been few frontier US open-model providers.
But earlier this week, Meta took action by launching Muse Glimmer, a 30-billion-parameter open-weight model, while also announcing plans to release the weights of the flagship Muse Spark 1.2 model. Meta CEO Mark Zuckerberg said it would resume releasing open-source models.
The move was prompted by cost-efficient Chinese open-weight models becoming an appealing choice for global enterprise customers eager to cut AI expenditure, arousing anxiety among US rivals. #🇨🇳 ChinA.I. 🤖🧠🦾🤖
However, Chinese AI firms are no longer bearing the low-price strategy amid surging compute demand.
DeepSeek, which officially launched its DeepSeek-V4-Pro and Flash variants recently, announced a “significant” price hike of up to 11-fold at peak hours. The new policy will take effect on Monday, which is more expensive than certain American systems, such as OpenAI’s GPT-5.6 Luna.
Meanwhile, the developers of open-weight models have imposed new revenue sharing licenses, targeting big commercial users. Both Alibaba Group Holding, which released the model weights of Qwen3.8-Max, and Moonshot, behind the popular Kimi K3, asked users that leverage the systems to generate certain income to strike special deals. Alibaba owns the SCMP. Besides models, Chinese firms start to monetise their AI applications. Alibaba started to offer three membership tiers for an office assistant on its Qwen app, with annual plans ranging from 200 yuan (US$29.60) to 1,499 yuan.
The move follows ByteDance’s earlier adoption of subscriptions for professional versions of its Doubao chatbot, which range from 68 yuan to 500 yuan per month.
Alibaba and ByteDance’s subscription strategies mirror American peers such as OpenAI and Anthropic, which have pioneered paid tiers for premium features. Domestic rivals, including Moonshot AI and Z.ai, have also introduced paid plans for advanced tools.
Running large language models and multi-modal agents at scale requires vast computing resources, making monetisation critical to offsetting rising infrastructure costs as the technology matures.
One example is Tencent Holdings, which nearly tripled its capital expenditures in the second quarter, to 52.8 billion yuan, to supercharge its AI push. Alibaba will report its earnings next week.
https://www.scmp.com/news..
