Anson Ho is a senior researcher at Epoch AI studying the drivers and effects of AI progress. His work connects hardware-efficiency limits, model evaluation and analysis of technological change.[1][2][4]
Start with Limits to the energy efficiency of CMOS microprocessors →
Links
Personal site
X · ansonwhho Senior Researcher · Epoch AI
Background
His research portfolio includes work on algorithmic improvement, a statistical framework for aggregating benchmark scores and contributions to the International AI Safety Reports. The common task is to turn scattered measurements into a more explicit account of AI development and its possible effects. His background includes a physics degree in the UK.[1]
Contributions
Estimate a physical limit on computing efficiency
Ho’s paper with Ege Erdil and Tamay Besiroglu models the energy dissipated in transistor switching and wire interconnects. It estimates a bound on floating-point operations per joule while accounting for uncertain inputs. The result is a framework for forecasting the remaining room for CMOS efficiency improvements, rather than a forecast that a particular chip will achieve the bound.[2]
Separate revenue from model availability
In his report with Cheryl Wu, Ho examines consumer apps, APIs, enterprise work, licensing and indirect benefits to existing products. Releasing weights does not eliminate every revenue source, but it can change competition and pricing power. The analysis distinguishes model-focused companies from larger firms whose AI products support other business lines.[3]
Tools, writing & talks
Writing, research & talks
Start here
A model of transistor and interconnect dissipation for forecasting efficiency limits.
Read the two-source energy model before its numerical range. Basic physics and computing knowledge helps; the summary explains how the bound differs from a chip roadmap.
Coauthor[2]Epoch AI lays out five revenue routes for Chinese AI labs: consumer apps, API access, enterprise deals, licensing, and cloud complements. A GLM 5.3 Flash case study shows open weights quickly pushing first-party token share down.
Catalog workOriginal sourceA report separating several AI revenue streams and their structural constraints.
Coauthor with Cheryl Wu[3]