The Chinese are confident about winning the AI race in terms of the parameters they have set for themselves. The confidence bordered on arrogance. They recognise only the United States as their peer competitor. Others don’t even come close. India was dismissed as being “nowhere in the picture”. In the race with the US, China is following a strategy different from that of the US. Many more entities are involved in this domain in China, unlike the concentration in a handful of high-tech multinationals in the US, such as Google, Microsoft, Meta, Apple, Amazon, and chipmakers like Nvidia.
The US was described as “chasing the magic bullet” of a singular proprietary artificial general intelligence (AGI), which would solve all human woes. China, on the other hand, was building a ubiquitous, open-source and deeply embedded network of AI that permeates every facet of the physical and economic world but is overseen by the strategic hand of the party-state.
The academic core resides in universities like Tsinghua, billed as the MIT of the East, which houses the Institute of AI Industry Research. Peking University leads in basic research and ethics of AI while Fudan University in Shanghai has become the hub for the integration of AI into finance and economics. Among the corporate leaders are Alibaba, through its Qwen models, and ByteDance, which leverages its massive visual and textual data flows from TikTok and Douyin to create powerful models. There are specialised players like iFlytek, focusing on voice and translation and SenseTime, which focuses on vision. There is a continuous feedback loop between the laboratory and the market.
The conference displayed examples of this “application” model in manufacturing and robotics, in pharmaceuticals (for drug discovery) and in defence (developing swarm technologies and autonomous logistics). This approach is underpinned by a policy commitment to open-source, which allows for rapid assimilation. The aim is to become the “Linux of AI”, a foundational layer, which is open to adoption by everyone precisely because it is free and efficient. It can set global technical standards while the US keeps its best models behind proprietary walls.
More than one speaker warned of a massive AI bubble building up in the US, with billions of dollars being poured into the quest for ever more powerful models even while productivity gains and potential profits are mostly expectations for the future. They said that the scale of such a bust would be far greater than the global financial crisis of 2007-2008, whose consequences are still playing out.
One of the more interesting interactions was with a senior functionary of the Chinese Communist Party School, which is the ideological fount for the party. He referred to an active debate among younger party cadres and intellectuals concerning the relevance and role of the party in an AI-dominated future. Some had argued that the party may lose its rationale in an AI future and that there may be no more room for any “isms” like socialism or capitalism. Such “erroneous” views were strongly rejected. It was asserted that the party’s wisdom and guidance would be required even more to guide AI into the right channels, and that AI would become a powerful tool in the hands of the party to promote a socialist future for China. The party would ensure that AI created the collective good rather than only shareholder value. It was interesting to note that such debates were taking place in party forums.
Where is India in the AI domain?
Clearly, India must follow a frugal path towards AI development. It simply does not have the resources on the scale deployed by the US and China. The India AI Mission has allocated about ₹10,300 crore, which is a modest figure by comparison. Relying on American or Chinese models, even if they are open-source, does create the risk of a dependency trap. But there are opportunities for leapfrogging. Like China, India, too, is following an applications-first approach. Projects like Bhasini are using AI to break the language barriers across India’s 22 official languages. The integration of AI into the “India Stack”, comprising digital identity and payments, show a clear focus on social utility. Indian startups are using small language models (models with 8 million to 1 billion parameters) for cost-effective solutions that can run on local devices without requiring data flows to foreign Clouds.
Graphic processing units (GPU) are the “oil” on which AI runs. Unlike central processing units in a standard computer, which work sequentially, one step at a time, GPUs work in parallel. When AIs like Gemini or ChatGPT “think”, they process huge volumes of data in multiple, virtually simultaneous steps. India aims to increase its current stock of 38,000 GPUs to 100,000 by the end of the year, with a target of 1 to 2 million by 2030. The state is providing GPU “time” as a service at a subsidised rate of only ₹65 per hour to foster application development by startups.
There is still a long way to go. The US has a current inventory of 5 to 7 million GPUs, and China has 1.5 to 2.5 million. The target for the US is 25 to 30 million by 2035 and China plans to achieve 10 to 15 million by the same date.
AI is one domain where late-mover advantage appears to hold greater promise in India. Like China, it may be prudent to focus on applications that advance economic and social aims rather than chase the magic bullet of AGI.
The author is a former foreign secretary
 "artificial intelligence, China, Technology")