The Beginning of Scarcity in AI

Article

• Tomasz Tunguz argues AI compute is entering a genuine scarcity era
• Demand from frontier labs outpacing GPU/ASIC supply; ASML a bottleneck
• Predicts 5–10 years of constrained compute; raises questions about who wins

Discussion

• mattas pushes back: labs burning cash faster than revenue — scarcity framing masks unit economics problem
• dmazin sees upside: scarcity will force better harness design and small model innovation
• paulddraper rejects thesis: supply can scale, models are getting more efficient (DeepSeek)
• isawczuk calls it artificial scarcity; com2kid notes open-weight models close the gap for most use cases

Discuss on HN