Since I published the three-part essays on key trends of China’s AI development, I have met with a few VCs heavily involved in AI investment in the country. From the conversations, an outline of Chinese AI companies’ strategy became clear and it dovetailed into my predictions.
In short, AI players in China intend to succeed in the same way as China’s EV, photovoltaics, and biopharma companies – winning by changing the economics of these businesses. In plain language, they intend to outcompete by making AI adoption large scale and low cost, neutralizing rivals with high cost, high margin business model.
Context
Since 2018, the US government has attempted to strangle China’s AI development through chip export control and denial of access to the most advanced AI models developed in the US. The launch of DeepSeek broke the chokehold and demonstrated China’s resilience and capacity for innovation.
Subsequently, the US government has started to ban DeepSeek in government agencies, and OpenAI is lobbying to ban DeepSeek more broadly in the US. It’s conceivable that the US regime will pressure its client states to ban DeepSeek as well, as it did with Huawei. Similar restrictions are likely to put on other Chinese AI companies.
The response from Chinese companies is interesting. Since DeepSeek’s launch, there is a flood of high performance AI models coming out China – Alibaba’s Qwen, ByteDance’s Doubao, Tencent’s Hunyuan, Baidu’s Ernie.
Unlike their US competitors, these AI models are open source and free – essentially given away for anyone in the world to download, modify, and integrate.
Why are they doing this and what is the strategy behind it?
Comparing strategies
Since ChatGPT’s release in Nov 2022, US tech giants from OpenAI, Microsoft, Google, and Meta have pursued the same strategy – they have stockpiled the most advanced AI chips from Nvidia, invested tens of billions in data centers, developed closed proprietary LLMs, and are charging high subscription or licensing fees to monetize their products.
These tech firms treat AI as an exclusive resource, restricting access to their most powerful models behind paywalls. OpenAI, Google DeepMind and Anthropic limit full access to their most advanced AI models, offering them through paid subscriptions and enterprise deals. These AI programs are valued at many billions of dollars as investors expect outsized returns.
Basically, the Silicon Valley firms’ AI investment is premised on a high cost high margin business model, protected by a moat of proprietary IP. The business model is further reinforced by the prohibitively high investment requirements for raw compute – affordable to only the most deep-pocketed tech giants in the world and effectively pre-empting any competition.
China’s strategy is the exact opposite. Denied from accessing leading edge compute resources, even the biggest Chinese companies are forced to find innovative solutions to develop high performance models without the best chips. Rather than raw compute, Chinese companies have focused on smart engineering and algorithmic optimizations to develop their AI models.
As their models start to reach parity with US models, they have decided to open source their product to pool resources from all developers and accelerate improvements.
This approach has several distinct advantages –
– Low dependence on the most advanced AI chips
– Low capex need
– Decentralize development to leverage global AI talent pool
– Allow developers who have access to the best chips to contribute to model refinement
– Faster iterations. AI advances through iteration. Every new release builds upon the last, refining weaknesses, expanding capabilities and improving efficiency.
Through open sourcing, Chinese AI companies are creating an ecosystem where global developers continuously improve their models – without shouldering all the development costs.
Such an approach will fundamentally reshape AI’s economics. If open-source AI becomes as powerful as proprietary US models, the ability to monetise AI foundational models is neutralized. Why pay for closed models if a free, open, and equally capable alternative exists?
By making foundational AI models free and abundant, Chinese companies will destroy the paid business model – closed, proprietary systems built on huge capex investment.
As a side bonus, such a move will reduce the relevance of chip control and nullify the financial advantages of US AI firms.
Of course, free open-source AI model is not an end in itself. The ultimate goal of Chinese AI players is to move AI from foundational models to applications where China has inherent advantages – data and market. Monetization will happen at the application level as AI is integrated into industries and consumer use cases.
Rather than making money on AI models, Chinese AI companies will generate revenue and profit by selling AI solutions, building embodied AI, and incorporating AI in consumer goods and services. Vast profit pools exist in humanoids, autonomous driving, smart infrastructure, manufacturing and healthcare applications, and more.
Chinese government is already accelerating the application of AI technology in state owned businesses from telecom, banking, ports, and energy to public services such as hospitals, schools, and government offices. Private Chinese companies in the auto, electronics, pharmaceutical, and consumer goods industries are embracing AI as well. Once mass market adoption is reached, AI will be ubiquitous and cheap.
Of course, the open source nature of Chinese AI models will spur competition from other countries since everyone competes at a level playground. China’s bet is as the models get refined and upgraded continuously, everyone in ecosystem will benefit. China is likely to benefit the most as it has the biggest market and the most data which are essential to develop best-in-class applications.
If China pulls this off, its success in AI would be another victory most recently exemplified by its EV industry – winning the race by “changing lanes” and outsmarting competition.
