Systematically dismantle the AI compute supply chain.
A radical proposal suggests targeting TSMC, ASML, and key silicon mines to stop AI progress dead.
In a provocative blog post, AI safety researcher David Scott Krueger outlines a radical intervention to address existential risks from artificial intelligence: systematically dismantling the global compute supply chain. Critiquing the common binary choice between oppressive government control ("Lock It Down") and unchecked development ("Let It Rip"), Krueger argues for a third path that targets the physical hardware enabling advanced AI. The core idea is that by removing the specialized chips, factories, and materials, AI progress can be brought to a "grinding halt" without the downsides of the other proposed solutions.
The proposal identifies highly concentrated, vulnerable chokepoints in the supply chain. The most advanced semiconductor fabrication plants (fabs) are operated almost exclusively by Taiwan Semiconductor Manufacturing Corporation (TSMC). These fabs depend on Extreme Ultraviolet (EUV) lithography machines made solely by Dutch company ASML, which uses specialized lenses from Germany's Zeiss. Furthermore, the high-purity silicon required originates primarily from the Spruce Pine Mining District in North Carolina. Krueger notes that the subsequent AI chips are predominantly housed in large, visible datacenters with massive energy and cooling demands, making them difficult to hide. Since these critical nodes are largely within the US and allied nations, he argues coordinated government action could feasibly shut them down.
- Targets physical hardware chokepoints: TSMC fabs, ASML's EUV machines, and Spruce Pine silicon mines.
- Proposes a "third way" beyond government surveillance or unchecked AI development to mitigate existential risk.
- Argues the supply chain's concentration in allied nations (US, Taiwan, Netherlands, Germany) makes intervention feasible.
Why It Matters
This radical idea reframes the AI safety debate around physically disrupting the multi-trillion dollar hardware ecosystem enabling the AI race.