AI policy and the battle for computing power

Summary of AI policy and the battle for computing power

by Practical AI LLC

48m•March 9, 2026
Listen to Episode

Overview of AI policy and the battle for computing power (Practical AI Podcast)

This episode features Ben Buchanan (assistant professor at Johns Hopkins SAIS; former White House Special Advisor on AI) discussing how compute — not just data or algorithms — is central to modern AI, why control of the semiconductor supply chain (especially Taiwan/TSMC) is a geopolitical lever, how governments should work with the private sector, and what democracies must deliver to “win” the AI era while preserving safety and values.

Guest background

  • Ben Buchanan — former White House special advisor on AI (Biden administration), PhD in cyber operations, author (The New Fire; upcoming The Bitter Struggle), professor at Johns Hopkins SAIS.
  • Current work: advising American AI and cybersecurity companies and teaching graduate students.

Main themes and takeaways

  • The AI revolution differs from past tech revolutions because it is predominantly driven by the private sector rather than government R&D.
  • Compute (chips + power) is the primary limiting factor for state-of-the-art AI; controlling compute creates strategic leverage.
  • Democracies currently hold advantages in the compute supply chain (TSMC, ASML, U.S./Japanese firms), and policy should preserve that lead.
  • Effective AI policy requires educating policymakers, building guardrails, and crafting international norms — not just technical or economic playbooks.
  • Safe adoption of AI yields opportunity: “we get AI opportunity through AI safety.”

Key technical insights

  • Scaling laws (OpenAI paper, ~2020): performance of neural networks improves predictably with more compute (and data), making compute a core driver of AI progress.
  • Compute is physical: advanced chips, chip-making equipment, and electricity are tangible strategic assets — not just abstract “data.”
  • Semiconductor manufacturing concentration: a very large percentage of advanced chips are produced in Taiwan (TSMC) using specialized lithography tools (ASML, Netherlands), creating central points of vulnerability and leverage.

Geopolitical and policy actions discussed

  • U.S. strategies pursued (Biden administration):
    • CHIPS and Science Act — incentives to onshore advanced chip manufacturing (e.g., fabs in Arizona).
    • Export controls on advanced AI chips and chip-making equipment to limit China’s access.
    • National Security Memorandum (executive order–like guidance) directing defense/intel to work with private sector, with classified and unclassified components.
    • Hiroshima/G7 process and diplomatic efforts (UN resolution co-sponsored, multilateral outreach).
    • Political Declaration on the Use of Autonomy in Military Systems — multilateral principles for autonomy in military contexts.
  • Rationale: restrict scarce, dual-use compute resources from adversaries; preserve democratic advantage and buy time to invest in safety and standards.

Government–private sector relationship & guardrails

  • Two distinct relationships:
    1. Government ↔ chip-makers (e.g., NVIDIA, TSMC): export controls and supply-chain policy to manage distribution of scarce hardware.
    2. Government ↔ AI model developers (e.g., OpenAI, Anthropic): collaboration for procurement/adoption, standards, safety/testing, and responsible military use.
  • Approach advocated:
    • Not purely laissez-faire: “let the private sector cook” is inadequate for strategic/defense risks.
    • Collaborative regimes that protect national security while enabling innovation and adoption.
    • Case-specific guidance: some military/autonomy uses will require more human judgment or rules (DOD policy 3000.09 referenced).

International coordination and diplomatic reality

  • Multilateral engagement (G7, UN, bilateral talks) is essential to build interoperable standards, norms, and export regimes.
  • Tensions and frayed alliances complicate coordination; diplomatic ability matters for aligning regulation and safety standards.
  • Buchanan argues democracies should lead, but also engage (including quiet talks) with autocracies where necessary.

Cybersecurity implications

  • AI dramatically affects cyber offense/defense:
    • Vulnerability discovery: AI tools are proving capable of finding high-severity software vulnerabilities at scale (Anthropic example cited).
    • Faster vulnerability finding can be used for both better defense (patching) and more powerful offense (exploitation).
  • Cyber operations are a persistent component of national security; AI changes the pace and scale of capability.
  • Nations that lead in AI-enabled cyber capabilities will gain significant strategic advantages.

Measuring democratic “success” in the AI era

Buchanan proposes three metrics to assess whether democracies are “winning” with AI:

  1. Invention — Are democracies inventing and leading core AI technologies?
  2. Adoption — Are democracies effectively adopting AI across defense, economy, and services to boost prosperity and security?
  3. Values — Are democracies using AI consistent with their principles (privacy, rule of law, civil liberties, avoiding surveillance authoritarianism, protecting kids online, managing disinformation and job impacts)?

Notable quotes & metaphors

  • “Compute, not data, was driving the bus.” — Compute as the central driver of progress.
  • “To resist metaphor is to endure the thing itself.” — Prefer understanding AI on its own terms rather than overreliance on analogies.
  • Railroad analogy: early railroads were transformative but unsafe; over decades, private and public solutions combined to improve safety and speed. Lesson: safety and speed can be complementary, not oppositional.
  • “We get AI opportunity through AI safety.” — Safety enables sustainable innovation and public trust.

Recommendations / Actionable items (for policymakers, industry, and researchers)

  • Policymakers:
    • Protect and diversify the semiconductor supply chain (continue CHIPS investments).
    • Maintain export controls for scarce dual-use compute that would materially empower adversary militaries.
    • Fund high-skilled immigration and talent policies to attract AI researchers.
    • Create flexible procurement pathways for defense/intel to responsibly adopt industry advancements (while enforcing safety guardrails).
    • Lead multilateral efforts to set norms, interoperability, and export regimes among democracies.
  • Industry:
    • Engage constructively with governments on safety, testing, and standards.
    • Invest in robust red-teaming, vulnerability disclosure, and patching processes.
  • Researchers/Practitioners:
    • Prioritize reproducible safety research and cybersecurity use-cases.
    • Contribute to international standards and governance discussions.
  • General public:
    • Expect the interplay of safety and economic opportunity; advocate for policies that protect values while enabling innovation.

Episode value and who should listen

  • Strong value for policymakers, security professionals, AI researchers, industry leaders, and anyone interested in how geopolitics and supply chains shape AI power.
  • Useful primer on why compute matters geopolitically and how policy choices (export controls, CHIPS Act, alliances) can shape the balance of power.

Closing note

Ben Buchanan emphasizes that democracies have meaningful advantages in AI invention today — especially in compute — but success depends on adopting the technology wisely and ensuring it aligns with democratic values. The policy balance is both strategic (supply chains and export controls) and ethical (safety, norms, and multilateral governance).