Best of Both Sides: Dont bet on an AI trickle-down
(Illustration: C R Sasikumar)
There is an Adam Smith reading of the present moment: Even as the US tightens its grip on the most advanced chips and models, the invisible hand will, given time, diffuse capability outward, with prices falling and the frontier becoming the floor. The difficulty is the time. As John Maynard Keynes noted of this classical optimism, in the long run, we are all dead. In a domain where capability compounds by the quarter, the long run is the period in which the strategic gap becomes permanent. This is the harder reading that New Delhi should adopt. AI is being weaponised, and the country that holds the frontier has every incentive to keep it scarce. Modern statecraft has normalised the conversion of interdependence into leverage. We saw it with vaccine nationalism during the pandemic; we are seeing it with the Strait of Hormuz. We see it in critical mineral and rare earth export controls, where processing dominance can be switched on as a weapon. AI is the next and most consequential entry on that list because it is not just one commodity but the general-purpose input to all the others.The American architecture for this is already built. The 2025 AI Diffusion framework sorted the world into tiers of access and, for the first time, placed AI model weights under export control alongside the chips that train them. That specific rule has since been loosened into a more discretionary, country-by-country regime, but the principle survived and hardened. Its sharpest expression came only weeks ago, when Washington ordered Anthropic to cut off the latest versions of its most capable models, Mythos and Fable, to every user outside the US; the company complied by suspending access for everyone, curbs that have since been lifted from July 1. At the G7, the pitch to allies was that frontier models would be reserved for “trusted partners.” Read plainly, that means access to the most powerful general-purpose technology of the era is becoming a privilege that Washington can extend or revoke at will. That is not a market trickling outward. It’s an arms race with a gatekeeper. Unlike earlier technologies, AI feeds its own outputs back as inputs and reaches into cognition itself, so the leader’s edge compounds and a follower’s lag hardens into a structural gap. In my conversations across Washington this month, the same uneasy picture kept surfacing. Neither India nor Europe sits at the frontier of AI development. China, by contrast, has leaned into open-weight models that are cheaper and easier to adopt, which gives much of the world, India included, a powerful commercial pull towards Chinese systems exactly when the American frontier becomes harder to reach. That pull collides with real concerns about data sovereignty, security, and strategic dependence. India risks being squeezed from both ends: Rationed by Washington at the top, courted by Beijing at the bottom, and fully sovereign at neither. None of this is a case for fatalism, because India is not starting from nothing. It has the demographic ballast its rivals lack: The world’s largest population, a median age under 30, and one of the deepest pools of software talent anywhere. But scale is potential, not power, and soft endowments do not substitute for hard capacity. India still buys its chips rather than making them, and the binding constraints are physical: Advanced semiconductors, reliable energy, data-centre capacity, patient risk capital, and the university-to-industry pipelines that turn talent into frontier research. To its credit, the government has stopped treating this as a private-sector afterthought. The IndiaAI Mission has built a shared national compute pool that has crossed 34,000 GPUs and is scaling further, with subsidies for startups and universities. It has commissioned a dozen indigenous foundation models, and at the India AI Impact Summit in New
Discussion (0)