Anthropic's CEO promised AI could compress a century of medical progress into a decade — three weeks later AWS announced a $4B Trainium cluster commitment to support the training
"Powerful AI could compress 50 to 100 years of biological and medical progress into 5 to 10 years. AI-accelerated science could solve mental health, eliminate most cancers, prevent most infectious disease, and double the human lifespan. In climate, AI could accelerate the deployment of clean energy and the development of new materials."
AWS and Anthropic deepen strategic collaboration. Anthropic will use AWS Trainium and Trainium2 as its primary training chips, and AWS as its primary cloud and training partner, with $4 billion in additional investment from AWS to support model development including the build-out of large-scale Trainium clusters. — About Amazon — AWS and Anthropic deepen strategic collaboration, November 2024. source ↗
Amodei's October 2024 essay framed the promise of powerful AI as a moral imperative justifying enormous resource commitment: doubled lifespans, eliminated cancers, accelerated clean energy. The essay does not quantify the energy and water cost of producing the AI that would deliver those benefits. Three weeks after publication, AWS and Anthropic jointly announced a $4 billion deepening of their compute partnership centered on large-scale Trainium and Trainium2 clusters — the physical hardware required to train the next generation of Claude. A single large training run for a frontier model is estimated at 50–100 GWh; at the platform's declared baseline of 10,500 kWh per US suburban household per year, even the lower bound (50 GWh) equals the annual electricity of ~4,762 suburban households. The essay's normative case for AI's benefits is sourceable; the energy and water cost of producing it is structurally outside the essay's frame.
Baseline: 2024 US suburban household consumption: 10,500 kWh electricity / 109,500 gal water / 16 tonnes CO2e per year. · Methodology · How this is verified