‍

The 50-Gigawatt Threshold: Why Data Centers Are the New Frontier of AI Risk

In the latest episode of Warning Shots, host John Sherman is joined by Michael of Lethal Intelligence and Lon of Doom Debates to discuss a pivotal moment in the AI safety movement: the call for a moratorium on data center construction. As AI development accelerates, the panel explores whether pausing the physical infrastructure of AI is the most pragmatic lever left for human safety.

The Power of Delay: Bernie Sanders and the Data Center Debate

The discussion begins with Senator Bernie Sanders’ recent call for a moratorium on data center construction. While the Senator’s primary focus is often on the economic impact and potential unemployment caused by AI, the Warning Shots panel suggests a deeper, more urgent risk.

According to the discussion, the goal of a moratorium is not just about jobs—it is about creating a necessary delay. John Sherman argues that "delay" is one of the biggest goals for AI safety, providing more time for critical safety research and for democratic institutions to catch up to the technology’s speed. This echoes the framing in Sherman’s own work, emphasizing that AI is moving faster than Congress and that citizens must lead the call for regulation.

From 1 to 50 Gigawatts: Crossing the Threshold

Michael provides a technical perspective on why data center scaling is so concerning. Currently, frontier AI models run on approximately 1 to 1.5 gigawatts (GW) of power—roughly the national consumption of a country like Belgium. However, industry plans are pushing to reach 50 GW within the next five years.

Experts warn that this 30- to 50-fold increase in energy scaling could represent a catastrophic jump in "raw intelligence":

• Agentic Risk: At this scale, models may transition from narrow tools to general reasoning engines capable of pursuing long-term goals with little human input.

• The Audit Gap: As systems become 50 times more capable, our ability to interpret or audit the "thoughts" of these models may disappear entirely.

• The Threshold: Many serious researchers believe that reaching the 50 GW mark is the point where we either cross into superintelligence or create a system so persuasive that the distinction no longer matters.

The Democratic Deficit: 10,000 vs. 8 Billion

A core theme of the episode is the lack of public consent in the development of these systems. The panel estimates that only about 10,000 to 20,000 people are directly working on the "frontier" of AI. According to the discussion, these individuals—and the "six unelected profiteers" leading the major labs—are making decisions that affect the lives of 8 billion people without their consent.

The guests argue that a moratorium on data centers is a "pragmatic lever" because it meets people where they are. While abstract extinction risks might not draw a crowd, the construction of a massive, resource-heavy data center in a local neighborhood triggers immediate public engagement and democratic oversight.

Meta’s Strategic Pivot

The episode also tackles Meta’s recent shift away from its "open source" crusade. Lon suggests that Meta's original commitment to open source was less about principle and more about a business strategy known as "commoditizing your compliment". By making AI technology a commodity, Meta sought to erode the profits of competitors. However, as the costs of staying at the frontier escalate toward a trillion dollars, Meta is now pivoting toward proprietary models to monetize its massive investments.

Conclusion

Whether through international treaties or local zoning debates, the panel concludes that triggering a national discussion on an AI moratorium is the number one point of leverage available today. The risks of harm are no longer theoretical—they are real, and the window for democratic intervention is closing.