It's the middle of 2026 and the best AI models on the planet are ones you can't use. Anthropic built Mythos, decided it was too dangerous to ship, watered it down into Fable, and cut off paying customers. OpenAI's newest model went to a handful of approved American companies. The chips arrive in India on a quota that moves with the mood in the White House. And in July, Beijing convened Alibaba, ByteDance and Z.ai to weigh restrictions on overseas access to its own best models, open weights included.
Every serious enterprise on Earth now has to build for a world where the model it depends on can be switched off overnight by an export order it had no say in.
Praveen's argument, first made in The Ken's AI column Zero Shot: that's the opening Indian IT services has waited half a decade for. When models can be turned off, when a Chinese model is brilliant but you can't be sure what's inside it, most enterprises need someone who can run all of them inside its own walls and keep the lights on when one goes dark. That's an integration and governance problem, and integration is the one muscle India has been building for thirty years. India has no frontier model of its own, which everyone treats as the reason it lost the race.
Praveen argues that this is specifically India's qualification for the job, because you can't be the neutral broker if you're also a competitor.
Pranay Kotasthane of the Takshashila Institution tackles the geopolitics side. He argues whether India’s non-alignment position is credible, and whether it can even hold this in the face of significant complicated interests globally around technology, chips and models.
Brady Ng, Deputy Editor of The Ken, and Zero Shot co-host, provides the US-China perspective. Chinese open weights grew out of a two-decade open-source culture formed behind the Great Firewall, and TikTok and Douyin already show what a dual-track export model looks like coming from one company. Brady shares the background and the factors that make this future viable for everyone.
Links
* Pranay Kotasthane, "Should the US Sell Advanced GPUs to China? An Indian Perspective" — Takshashila Issue Brief 2026-15, 29 April 2026 · https://takshashila.org.in
* Zero Shot episode with Kendra Schaefer on Chinese tech policy — https://the-ken.com/podcasts/zero-shot/china-let-its-ai-industry-run-wild-to-build-the-best-models-now-it-wants-control/
* Praveen's Zero Shot column on the non-aligned layer — https://the-ken.com/columns/zero-shot/the-non-aligned-layer/
* Reuters via Quartz — Beijing weighs curbs on overseas access to its top AI models, 7 July 2026 · https://qz.com/beijing-china-ai-model-export-restrictions-070726
* 2x2 EP50, "In an AI age, India does not have an open source strategy" — Pranay's first appearance, with Kailash Nadh, 10 July 2025
* 2x2 EP63, "India risks losing cultural relevance in the AI era" — 16 October 2025