It wasn’t so long ago that artificial intelligence was a niche concern, a subject for research labs and sci-fi thrillers, not the front page of global diplomacy. But today, AI isn’t just about technological breakthroughs—it’s about power, governance, and who gets to write the rules of the future. The recent Paris AI Summit saw the U.S. seek to shape the trajectory of the debates around AI governance, with Vice-President J.D. Vance’s focus on building AI, infrastructure, as opposed to “hand-wringing about safety,” underscoring Washington’s determination to shape the global regulatory framework in its own image.
Meanwhile, China’s DeepSeek, a frontier AI model with performance benchmarks rivalling ChatGPT-4, signals that Beijing isn’t content playing second fiddle in this race. Furthermore, the AI Diffusion rule, a U.S. policy announced by the Biden White House in its twilight-hour decision – which controls the spread of cutting-edge AI capabilities through a three-tier classification of countries – adds yet another layer of strategic competition. And in the midst of all this, India, a rising digital powerhouse, finds itself at a crossroads—striving to carve out its own leadership role in AI while navigating the competing pressures of regulation, innovation, and geopolitics. As New Delhi prepares to host its own AI Summit in late 2025, the real question is not just where India stands today but how might it appraise the recent developments to help it shape the AI order of tomorrow. This piece raises some open-ended questions to consider how India may approach this.
Recent Developments
We are crossing milestones in the field of AI faster than ever before. The arrival of ChatGPT was a seminal moment in the field of generative AI, attendant with concomitant advancements in computing power and machine learning. However, recent events ranging from developments in Washington to Paris to Beijing herald the arrival of an era in which technological breakthroughs and geopolitical events around them are outpacing our ability to track them.
For instance, last month, U.S. President Trump announced a joint venture between OpenAI, SoftBank, and Oracle to create the ‘Stargate’ initiative – which could see investments being made up to $500 billion over the next four years to create computing infrastructure such as data centres to power the next generation of AI models. Stargate was also positioned as a project to maintain American supremacy in AI technology, a leadership aspiration which has been increasingly challenged by China. Recent developments around a novel open-source artificial intelligence model made by the Chinese firm DeepSeek, with performance benchmarks at par with those of U.S. tech giants, have demonstrated that the perceived gap between the two may be narrowing. The fact that this was done in the wake of extensive U.S. export controls rules enacted to curb China’s access to cutting-edge AI chips, makes this accomplishment even more compelling.
When these developments are seen in the context of recent remarks at the Paris AI Summit by Vice President Vance, there is a clear sense that the U.S. is keen to maintain this lead which it has in AI Indeed, Vance asserted that “his administration will ensure that American AI technology continues to be the gold standard worldwide and we are the partner of choice for others.”
How does India figure with respect to these developments in this global race in AI?
The developments with DeepSeek may prompt a relook at how India has so far invested in the AI value chain. While most investment in India is reportedly concentrated in the applications layer – where developers and coders build on top of the large language models (LLMs), little attention has been paid to actually investing in building the LLMs themselves or the requisite computing power that undergirds it. India has traditionally been known for its scrappy approach to innovation, delivering cutting-edge R&D at lower costs, as evidenced by its civilian space programme since the late 1960s. The argument goes that if China could figure out a way to train LLMs at a fraction of the cost incurred by their counterpart U.S. firms (regardless of the debate around the actual costs incurred), despite export control restrictions, then why can’t Indian firms strive for the same? Could this be pioneered in India through a novel way to look at assembly and packaging of chips, or through innovations in a different approach to training LLMs altogether? Towards this, India’s IndiaAI Mission has already released a call for proposal to look at building all kinds of foundational models, including SLMs as well. It remains to be seen how this plays out.
India’s Considerations
Regarding Stargate, and the headlines it has generated, India has been keen to build and provide computing power as well. It has recently launched an IndiaAI Compute Portal, dubbed AI Kosha, that will seek to give access to computing power for more than 18,000 GPUs to start-ups, and researchers. However, while this may appear to be a far cry from the massive GPU build-up seen in the U.S. and China, perhaps it is worth looking at arriving at reliable estimates for the downstream use-cases for such computing power before coming to a definitive conclusion on the degree of computing power required for India. It is here that India should seek to leverage its partnership with the United States to seek seamless access to GPUs and AI infrastructure. The recent U.S.-India February 2025 Joint Statement also speaks to this need by considering deploying such infrastructure in India, although the granular details on how this may be done are perhaps still being worked out.
To keep up in the AI race, India needs to focus on creating LLMs and not just applications. It must leverage U.S. tech partnership to build sufficient computing power and adopt light-touch AI regulation
However, this would require both the Trump administration and the Indian government to discuss at whether it is possible to rewire and relook at the AI Diffusion Rule – an eleventh-hour proclamation by the Biden White House to regulate access to U.S.’s cutting edge capabilities in advanced AI chips. Here, India may be keen to address concerns regarding a roadmap to become a Tier 1 country, or at least be in a grouping which has less constraints than those imposed on Tier 2 countries, its current grouping in the AI Diffusion Rule. This may already be in the works. Other constraints imposed on entities in Tier 2 countries like India may be clarified — such as the requirement for all non-U.S. companies that seek a National VEU (Verified End-User) authorisation to accept the US rules on outbound investment, as they currently apply to U.S. firms. A National VEU authorisation is a classification provided under the AI Diffusion Rule to enable entities to purchase computational power equal to around 320,000 advanced GPUs over two years. This is so since the AI Diffusion Rule will also have a downstream impact on Indian firms that seek a National VEU status – as they would have to comply with these U.S. outbound investment rules.
Conclusion
When the worlds of technology and diplomacy first convened around AI last year—starting at the U.K.’s Bletchley Park and later continuing in Seoul—the dialogue was firmly centred on safety and fairness. However, the narrative has since shifted, with the focus now expanding to prioritise AI-driven innovation rather than being singularly dominated by concerns over risk and regulation. As India looks set to co-host the AI Summit in 2025, it may consider examining the shifting trajectory of this trend, given its desire to cultivate start-ups in the field of AI. Here, one of Vance’s remarks at Paris were particularly telling – where he sought to assure listeners that the Trump administration will “not be the one to snuff out start-ups” and will instead, “keep Big Tech, Little Tech, and all other developers on a level playing field.” This ought to be encouraging for India, as it considers its own regulatory framework, which currently, as per recent research put out by Carnegie India scholars “broadly supports a ‘pro-innovation’ approach to AI regulation.” However, India needs to do more on computing power – yes, the innovation in training LLMs with the advent of DeepSeek deserves to be studied, however, access to large amounts of computing power is likely to remain the critical factor in achieving even that outcome. The argument here being that that A.I. development is not just about a series of novel training runs, but that it needs be augmented by, and necessarily involves constant iteration that can only be achieved through massive computing power that supports such experimentation. As India navigates its own way in this ever-changing world, its position on AI innovation as the co-host of the AI Summit with France (that has also argued for a simplified approach to AI regulation) could also be a platform to articulate the Indian way on AI.
