technologyScore 55/100Research
Google achieving 10-100x model efficiency improvements over 2 years via distillation; frontier model training still at large scale, creating continued capex demand despite efficiency gains
Demis Hassabis· Google DeepMind· AI· 2026-07-19· about Google DeepMind (GOOGL)
“uh the model efficiencies are like 10x you know even 100x better for the same performance... Now the the reason that that isn't reducing demand is because we're still not got to AGI yet. So also the frontier models you keep wanting to train and experiment with uh new ideas at larger and larger scale...”
Why it matters
Google has achieved massive efficiency gains but continues scaling frontier models. This creates bifurcated capex profile: massive spending on frontier training clusters + efficient deployment infrastructure. Energy demand may not decline despite efficiency breakthroughs due to continued frontier model exploration.
Investment implication
Semiconductor and datacenter infrastructure suppliers (GPU vendors, ASIC designers, cooling/power providers) will see sustained high capex demand despite model efficiency improvements. Energy providers should plan for continued AI power demand growth. Distillation tools and model compression vendors may see growing adoption.