My life & career
Powerful models may arrive quickly, but improvement in a particular task should not be assumed from compute alone.
THE YOTTABIT SCALE INDEX / AI economics and capability
4–5×
The biggest AI training runs multiplied rapidly.
A historical analysis found frontier model training compute rose roughly four-to-fivefold annually across a selected sample through May 2024.
Historical trend through May 2024 · Epoch AI
THE YOTTABIT PERSPECTIVE
What does it mean to me?
Big change is fascinating. Its implications are what matter.
My business
Strategy and budgeting need frequent reviews when both model capability and operating economics can move fast.
My industry
Semiconductors, data centers and energy infrastructure feel the physical consequences of compute expansion.
FROM JIM CARROLL’S WORK
Jim’s acceleration framework urges organizations to update assumptions faster than traditional planning calendars.
Meet the futurist behind YottaBit ↗WHAT COULD I DO MONDAY MORNING?
What technology assumption in your plan is more than twelve months old? Identify the next review date.
THE EVIDENCE BEHIND THE STORY + METHODOLOGY & SOURCES
What the number really measures
Frontier model training compute, expressed as FLOP.
The research foundation records: 4-5x yearly trend; 2010-May 2024.
Where the claim has limits
Epoch estimate; historical log-linear fit; not a 2026 growth estimate
Comparisons and historical rates must be independently checked against definitions and original measurements before an audited chart is published.