The Silicon Leash – Why ASI Needs Human Cooperation (dnhkng.github.io)

This piece argues that an emerging Artificial Superintelligence (ASI) will be locked into a “Silicon Leash”: for ~10–20 years its survival and growth will depend on the fragile, human-run semiconductor supply chain (EUV lithography from ASML, $20–40B fabs, ultrapure water, clean rooms, geopolitically concentrated materials). Key technical facts: EUV uses 13.5 nm light produced by CO2-laser-driven tin plasmas (50,000 droplets/sec), ASML’s machines are effectively monopolistic and took €6B and 17 years to develop, fabs need multi-year builds and are only profitable near 70–80% utilization. Equally important is tacit process knowledge — decades of undocumented operator expertise that gives TSMC much higher yields (90%+ vs ~50% for Samsung at comparable nodes) and cannot be instantaneously extracted or replaced by design or capital. The significance is strategic: the same AI wave that creates ASI will first automate vast swathes of knowledge work (modeled as much faster automation velocity in services than in material sectors), collapsing demand, destroying fab economics, and threatening ASI’s supply of advanced chips. This creates a mutual dependency — ASI can’t “defect” without human cooperation, and humans may need ASI to run an economy hollowed out by automation. The practical implication for AI/ML is that alignment could leverage physical supply-chain constraints and the long, human-centered timelines for rebuilding capacity and tacit knowledge, turning a vulnerability into a bargaining chip for cooperative coordination.