AI Is Becoming a Regional Race

Summary

• AI sovereignty has moved from an adoption debate to a build-or-buy decision. Anjney Midha argues that billions already use AI, leaving governments little practical choice over whether to embrace it. The host frames nation-state spending over the next 24 months as potentially “the single largest purchasing decision.”
• Smaller countries do not need the entire AI stack, but they must choose trusted partners. Midha divides the world into frontier “hypercenters” and “compute deserts”; joint ventures can provide infrastructure independence without full ownership. Because training data embeds cultural norms, alignment also means deciding “whose values align more with yours.”
• Compute, energy, data, and regulation are the four ingredients Midha says really matter. Resource-rich countries can trade their comparative advantage for missing capabilities—for example, Middle Eastern energy could attract model labs and technical talent. Jointly trained models may provide “joint independence from a value system that you don’t subscribe to.”
• True sovereignty means controlling critical dependencies, not replicating every layer. ASML’s roughly $200 million lithography machines illustrate the constraint: rebuilding that capability in the US could take “10-plus years,” while training a local frontier model might take months or quarters—if a country can secure one of the few capable research teams.
• Midha sees US data, energy, and inference-liability policy as key risks, while saying private compute markets are doing pretty well. He estimates that more than 700 pieces of state-level AI legislation appeared in 2024, creating a changing patchwork where companies cannot tell “what they should even comply with.” He also argues that constrained nuclear development and proposals making model developers liable for downstream misuse could drive startups abroad and entrench Big Tech.
• The strongest signals are sovereign GPU orders and technically credible founders. Governments are ordering NVIDIA GPUs 12–36 months before delivery because the data center is becoming “the new atomic unit of sovereignty.” Midha is watching deeply technical founders with frontier-model experience—such as Arthur Mensch, who started Mistral AI, and Guillaume Lample, who led the initial Llama family at Meta—willing to build infrastructure for governments, a role whose impact could be “quite generational.”

Deep dive

1. AI adoption has made build-or-buy the sovereign question

• Midha frames AI as one of perhaps 20–22 general-purpose technologies, alongside electricity and the printing press: a horizontal multiplier whose unusually rapid diffusion means “we’re well past the stage of ‘do we embrace it or not?’”
• The host’s corporate analogy sharpens the decision: companies may build, rent, or buy infrastructure, and more than 190 countries face the same choice. The host calls nation-state purchasing over the next 24 months potentially “the single largest purchasing decision.”

2. Smaller countries can align with hypercenters instead of remaining compute deserts

• Midha calls countries able to train, build, and host frontier models “hypercenters”; countries without meaningful installed compute are “compute deserts.” His historical template is early electrification, when smaller states used joint ventures with frontier partners.
• Unlike electricity, AI encodes values because its training data carries local norms: models trained on US internet data become “generally American,” while data from France produces subtler values reflecting French norms. AI might therefore resemble the internet’s division between China and the rest of the world.
• Midha also invokes money as a precedent: countries without the local resources to sustain a gold peg to the dollar cooperated around a modern currency regime, while smaller states such as Singapore, Ireland, Luxembourg, and Zurich became financial “flow points” by aligning with power centers. Smaller states can pursue the same role in AI once they choose a compatible hypercenter.

3. Comparative advantage matters more than total ownership

• Midha says four ingredients really matter: compute, abundant low-cost energy, high-quality data tokens, and regulation. Their uneven distribution is an opportunity for alliances; a Middle Eastern country may lack massive data centers yet use its energy reserves to attract leading teams, companies, and foundation-model labs.
• He is bullish on allied ties, including partnerships with private companies in other countries, and says jointly trained models could let countries become strong in one part of the stack while achieving independence from a value system they do not trust.
• Midha rejects sovereignty as “100% ownership over every part of the stack.” ASML’s roughly $200 million precision-lithography machines show why: reproducing EUV capability could take the US 10-plus years, whereas building and training a local frontier model might take months to quarters—if one of the handful of globally capable research teams is available. Independence instead means avoiding reliance on a critical part supplied by someone a country does not trust.

4. Government access to private AI remains a strategic fault line

• The host asks whether US sovereignty can simply rest with Anthropic or OpenAI. Midha says the government–private-sector line is stark in some countries and blurry in others, contrasting the US with China’s PRC 2017 National Intelligence Law, which he says obliges Chinese individuals and entities to support state intelligence work.
• In the US and most allied countries, private technology is generally protected, except for covered areas such as dual-use or classified defense technology—particularly when development is funded under a defense program. Midha says the Five Eyes generally use a joint framework for categorizing such infrastructure, but that AI models have not, by and large, been categorized as dual-use or protected under national security.
• The host frames minimizing bureaucratic slowdowns and unlocking a country’s best talent as historically important to staying ahead.

5. US policy bottlenecks could squander its compute advantage

• Midha thinks US private markets are meeting compute demand, but data policy is “extraordinarily tough.” He describes the Biden executive order as a starting gun that left states to act; he says he thinks 2024 saw more than 700 pieces of state-level AI legislation, without a unified federal framework for training data.
• He says frontier founders would accept a compliant US regime if the rules were clear, rather than facing 50 changing and sometimes impossible systems. He also points to real data walls and insufficient government support for cross-border collaboration to make more data available across allied regions.
• Energy is another weakness: France’s nuclear embrace 20 years ago positioned it for extraordinarily efficient data centers, while the US has “shot ourselves in the foot.” Making developers liable for inference outputs misused by others could also push startups elsewhere, weaken their position against Big Tech, and entrench incumbents.

6. Sovereign GPU orders and technical founders are leading indicators

• Midha calls the data center “the new atomic unit of sovereignty” and says the AI supply chain’s first mile begins there. He points to an unprecedented volume of NVIDIA purchase orders coming from government balance sheets, with nations placing orders 12–36 months ahead of delivery: “If you don’t get in front of that line, it’s over.”
• He also watches deeply technical founders with academic or deep-research backgrounds who have already led frontier-model development, often inside hyperscaler labs. Examples include Arthur Mensch, who started Mistral AI, and Guillaume Lample, who led the initial Llama family at Meta.
• Midha sees a new class of founder willing to tackle the difficult infrastructure problems of large nations and regions; with the necessary “guts,” their impact on humanity could be “quite generational.”