Conference Presentation, Interview, Fireside Chat, Keynote
AI Is Becoming a Regional Race
- AI is projected to achieve one of the fastest diffusion rates of any general-purpose technology, potentially creating a fragmented global landscape similar to the historical division of the internet into a "Chinese internet" and the rest of the world.
- Over the next 24 months, governments are expected to prioritize "build or buy" decisions, with smaller nations seeking joint ventures or alignment with "hypercenter" frontiers to become modern financial "flow points" based on specific value systems.
- The global distribution of access to compute, energy, data, and regulation is predicted to remain uneven, with Middle Eastern countries leveraging energy reserves and other nations placing orders for GPUs 12 to 36 months in advance to secure their status as "hypercenters."
- Building advanced infrastructure such as EUV lithography capabilities is forecast to require a timeframe of over 10 years, leading to a near-term strategy where jointly trained models between countries emerge as a feasible alternative to total infrastructure independence.
- Sovereign AI is defined as avoiding reliance on a single trusted provider for critical stack parts rather than achieving 100% ownership of every component, though the line between government and private control is expected to remain stark in China versus blurry elsewhere.
- The United States is predicted to face competitive handicaps if a unified federal framework for data regulation and training is not established quickly, as the current patchwork of over 700 state-level laws and liability rules on inference outputs is viewed as infeasible and likely to drive developers to other jurisdictions.
- Legislation in the next year imposing liability on developers for inference outputs is expected to force startups to lose ground to incumbents, penalize US competitiveness, and potentially cause model developers to migrate to countries with less stringent copyright and IP rules.
- US energy production for data centers may be constrained by restrictions on nuclear power compared to nations like France, while the private sector in the US and allied countries is expected to remain protected from government technology sharing by default.
- A lack of government support for cross-border data collaboration is anticipated to hinder frontier research in the US and allied regions, whereas the speaker notes that unlocking talent with minimal bureaucratic slowdowns typically yields superior results.
- A new class of founders with deep technical research backgrounds and a focus on solving nation-state infrastructure problems is expected to emerge, delivering generational impacts while the speaker asserts that unlocking the best talent usually wins against fragmented regulatory approaches.