AI as New Global Power?

February 27, 2026·13 min·1 clip

The U.S. promotes 'real AI sovereignty' at the India AI Impact Summit, but global south nations fear dependency on proprietary American systems.

The episode opens with Michael Zezas and Stephen Bird discussing the geopolitical implications of AI, specifically following Bird's participation in the India AI Impact Summit. Zezas frames the central tension: the U.S. is promoting global AI adoption built around what it calls real AI sovereignty, or strategic autonomy through integration with the American AI stack, while several nations — particularly from the global south and potentially parts of Europe — are skeptical of proprietary dependency. Bird identifies the Pax Silica agreement as a significant step in securing supply chains and AI technology access between the U.S. and India. He highlights Prime Minister Modi's focus on ensuring all Indian citizens have access to AI tools, citing a vivid rural health care scenario: a villager photographing a medical condition and receiving an AI-assisted diagnosis in the absence of any medical professional. Bird notes that American hyperscalers are attempting to serve these objectives through open-weights model versions made available to Indian health agencies and government. However, he frames the core tension as one between open access and explainability on one side, and the superior capability trajectory of proprietary models on the other. The scaling law argument is central: the big five U.S. AI firms have assembled approximately 10 times the compute used to train prior models, and if scaling laws hold, this will yield roughly twice the model capability. Bird argues this compute gap will be very difficult for open-source alternatives to bridge. Zezas frames the U.S. strategy as one of using technology allure to build geopolitical alignment, analogous to how military equipment dominance created a security umbrella that other countries wanted to be under for 80 years. Bird extends the analogy to U.S. dollar dominance, noting that as the predominant payment system, the dollar gives the U.S. the ability to impose sanctions — and a dominant AI stack could provide similar leverage over future use cases. Both analysts acknowledge the strategy's durability is uncertain: India has not yet signaled a clear preference between proprietary and open models. Bird identifies the April-June 2026 window as the likely moment for major U.S. LLM reveals, noting this timeline is based on tracking chip purchases and executive statements. He cites METR, a third-party organization that benchmarks model task complexity, noting that complexity has been doubling every seven months and that a recently introduced leading American model scored approximately 15 on METR's complexity scale, well above prior expectations. Bird also raises the prospect of recursive self-improvement as a concept gaining traction among LLM executives. The episode closes with the observation that the open versus proprietary model tradeoff will become increasingly critical through spring and summer 2026.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Thoughts on the Market