infrastructureScore 45/100Watch
AI models have scaled from 175 billion to 10 trillion parameters over six years (100x growth), with continued scaling planned, implying sustained demand for advanced semiconductor capacity and packaging
Leopold Aschenbrenner· Forethought· AI· 2026-07-20
“Right now, they're at something like 10 trillion parameters. Back in 2020, um it was more like 175 billion. So, we've grown like two orders of magnitude uh in six years.”
Why it matters
This 100x parameter increase over six years establishes a sustained trajectory of AI model complexity that requires proportional increases in GPU capacity, memory bandwidth, and advanced packaging to handle training and inference.
Investment implication
Semiconductor suppliers (GPU manufacturers, memory providers, advanced packaging vendors like those serving CoWoS and chiplet architectures) face multi-year demand acceleration from frontier AI labs. Power infrastructure (data center cooling, power delivery) also becomes critical.