3 min readJun 22, 2026 06:20 AM IST
First published on: Jun 22, 2026 at 06:20 AM IST
Over the past few weeks, two governments showed us how they are thinking about AI. Beijing restricted overseas travel for top AI engineers from private firms like DeepSeek and Alibaba. Washington issued a directive to Anthropic to suspend its frontier models for foreign nationals, including its own employees. A commercial product used by millions of people was pulled off the shelf. One country treated software engineers like nuclear physicists, and the other treated a software product like a munition. Both policy actions flow from the same story about what AI is.
National security thinking in the US and China treats AI as a zero-sum arms race or a weapon of mass destruction to be contained through non-proliferation. This narrative has been promoted by frontier labs to gain market share and shape the regulatory landscape. Other policy instruments seen recently, such as export controls and hardware tracking of chips, also stem from this narrative. The framing makes coercion feel like prudence.
However, this analogy misrepresents the technology. Unlike rare, controllable fissile material, AI models are mathematical artefacts that are easily reproducible and constantly diffusing. Chinese firms built competitive models despite chip export controls, and powerful open-source models remain freely downloadable. The bottlenecks to unlocking productivity lie in widespread adoption: Bridging the capability-reliability gap, learning curves, organisational reforms and regulatory compliance.
A more honest framing is the geopolitical innovation race. Here, a dense web of public and private actors across the world compete and collaborate, driven by a mix of national security, economic advantage, and market incentives. The policy instruments that follow are constructive: Compute investment, skilling, industrial policy, strategic partnerships, organisational reforms.
The response for a middle power like India is to lean into AI as a general-purpose technology while acknowledging the geopolitical innovation race. The value lies not just in inventing the frontier but also in diffusing it widely and well. The Supreme Court’s draft guidelines on the use of AI tools in courts are a telling example. The returns of the AI era will accrue to the countries that put it to work in health, agriculture, law, and public services.
That points to three priorities. Diffusion over denial: Spend political energy on adoption infrastructure, regulatory clarity, and sector deployment. Plurilateral resilience over self-sufficiency: Coalitions of the like-minded — India and the EU have complementary strengths — can share open-source development costs and build trusted supply chains. Finally, open over closed: The Fable 5 and Mythos 5 suspension shows that dependence on a nationally controlled, closed model risks access being withdrawn overnight. Open-weight models, open standards like RISC-V, and open alternatives to proprietary stacks are resilient because no single government can switch them off.
The writer is associate fellow, Takshashila Institution
The response for a middle power like India is to lean into AI as a general-purpose technology while acknowledging the geopolitical innovation race. 
Share
