Practical AI

AI policy and the battle for computing power

Mar 9, 2026
Listen Now

Summary

The episode examines how AI is reshaping global power by shifting the center of progress into private-sector firms and making computing power — not just data — the primary driver of modern AI capability. It highlights the geopolitical concentration of advanced chip manufacturing (notably TSMC and ASML) as a strategic vulnerability and lever between states, especially the U.S. and China. The conversation covers policy trade-offs for democracies: building compute advantages, coordinating international safety norms, and preserving democratic values while adopting and operating AI. The episode also explores immediate security implications, notably how AI accelerates cyber offense and defense by discovering vulnerabilities at scale, and debates around export controls and military uses of advanced AI.

Key Takeaways

  • 1Compute, more than data, is the dominant lever driving current AI progress.
  • 2Advanced chip manufacturing is highly concentrated and creates geopolitical vulnerability.
  • 3Democracies should combine competition with coalition-building to lead responsibly in AI.
  • 4AI is rapidly changing cyber operations by finding vulnerabilities at scale.
  • 5Policy choices about export controls and military deployment of AI are contentious and consequential.

Notable Quotes

""And this is a really important insight which is the more computing power you use to train an AI system, the more powerful the resulting AI system.""

""97% of the advanced computer chips in the world are made in Taiwan by a company called TSMC using incredibly advanced machines from a company in the Netherlands called ASML.""

""AI is not a partisan issue.""

""Anthropic... published that [their model] had found something like 500 high severity vulnerabilities in open source software.""

Episode questions

Why is compute considered more important than data for current AI progress?

OpenAI's scaling-law research and subsequent practice show that increasing compute during training reliably improves model capability; while data matters, compute tends to be the limiting scaling factor. That makes chips, power, and fabrication capacity central policy targets.

Why should non-specialists care about Taiwan's semiconductor industry?

Taiwan (TSMC) produces an overwhelming share of advanced chips; disruption would ripple across consumer goods, industry, and national security, costing potentially trillions in global GDP and delaying many technologies. Thus, chip supply concentration is both an economic and geopolitical risk.

How is AI changing cyber operations today?

AI models are now capable of discovering and exploiting software vulnerabilities at scale (example: industry model reportedly found ~500 high-severity flaws), increasing both offensive capabilities and defensive automation needs. This makes cyber an immediate domain for national-security AI strategy.

How can democracies 'win' in the age of AI while maintaining values?

Ben proposes three measures of success: invent the technology (lead in innovation), adopt it effectively across economy and security, and harvest/operate it consistent with democratic values (safety, labor impacts, anti-surveillance, integrity). Policy should strengthen each pillar.