Conference Presentation, Interview, Fireside Chat, Keynote
AI Is Becoming a Regional Race
- General Purpose Technology (GPT) Context: Modern AI (deep learning, generative models) is classified as a GPT, joining a historical list of roughly 20-22 technologies like electricity and the printing press that act as horizontal economic multipliers across society.
- Adoption Trajectory: AI is diffusing through society at one of the fastest rates for any GPT, rendering the binary question of "embrace vs. hostile" largely obsolete as billions have already adopted the technology.
- Current Strategic Dilemma: The primary decision facing nation-states is now "build or buy," representing the single largest purchasing decision likely to occur in the next 24 months.
- Infrastructure Independence: The core strategy is not total sovereignty over the entire stack, but rather independence from critical components where trust deficits exist or where reliance on untrusted partners creates vulnerability.
- Hypercenters vs. Compute Deserts: The global landscape is bifurcating into "hypercenters" (nations capable of developing and hosting frontier models) and "compute deserts" (regions with no relevant install base of compute capacity).
- Joint Ventures for Small Nations: Small countries without domestic compute or talent can achieve value-aligned independence by forming joint ventures with frontier nations that match their specific cultural and value systems.
- Value Encoding in AI: Unlike physical infrastructure, AI models encode the values and norms of the data they are trained on; a model trained on US data reflects American norms, while French data yields different cultural encodings.
- Value Alignment Strategy: Small nations must first identify which "hypercenter" value system aligns with theirs, analogous to historical currency regimes where smaller nations allied with a single global reserve currency (e.g., the US dollar) rather than attempting to maintain independent gold pegs.
- Four Critical Ingredients: The AI stack relies on four unevenly distributed resources: compute, abundant low-cost energy, high-quality data tokens, and regulation.
- Comparative Advantage: Nations should leverage their specific natural or human assets (e.g., Middle Eastern oil for energy) to partner with other nations for missing resources, rather than attempting total vertical integration.
- Feasibility of Total Independence: Total ownership of the entire stack (lithography to model layer) is infeasible in the short term; for example, replicating the capabilities of ASML (the sole producer of EUV lithography) would take over 10 years and requires $200 million per machine.
- Sovereign AI Definition: True sovereignty means avoiding reliance on an untrusted partner for a critical part of the stack, rather than owning every layer of production.
- Legal Divergence on Private Sector: There is a stark contrast between the US/allied model (where private companies are generally protected from mandatory government technology access) and the PRC model (where the 2017 National Intelligence Law mandates private entities support state intelligence work).
- Regulatory Fragmentation in the US: The US faces significant risk from a "patchwork" of state-level data regulations, with over 700 AI-specific laws passed in 2024 alone, creating an impossible compliance environment for companies compared to nations with unified frameworks.
- Data Wall Risks: Frontier research in allied countries is hindered by strict copyright/IP enforcement and a lack of cross-border government collaboration to share data, allowing labs in regions with less stringent rules to race ahead.
- Energy Deficits: The US has hamstrung its AI potential by failing to embrace nuclear energy, whereas countries like France utilized nuclear infrastructure 20 years ago to secure efficient data center power.
- Liability and Inference Regulation: Proposals to hold model developers liable for user misuse of inference outputs risk driving developers abroad and entrenching big tech incumbents by forcing startups to absorb unsustainable compliance costs.
- Leading Indicator for Sovereignty: The purchase of GPUs (specifically orders placed 12-36 months in advance) is the primary metric for identifying emerging "hypercenters" and new national atomic units of sovereignty.
- Emerging Founder Archetype: A new class of founders is emerging—deeply technical researchers from major hyperscaler labs (e.g., former DeepMind or Meta teams like Arthur Mensch or Guillaume Lample) who are mission-led to solve national infrastructure problems.
- Time Horizon: Building domestic AI infrastructure is a decade-scale endeavor, necessitating immediate strategic alignment and forward-looking procurement decisions.