# Training power doubles every two years — 900MW models by 2028

- **Speaker:** Greg Osuri
- **Attribution:** attributed
- **Said:** 2025-04-11
- **Theme:** Energy & AI — https://www.gregosuri.com/predictions/energy-ai/
- **Source:** Greg Osuri, CEO of Akash Network, on why compute should be shared with a decentralized marketplace! (LegendsNLeaders)
- **Watch at:** 00:07:39 — https://www.youtube.com/watch?v=vMx6G0ZOxeo&t=459s
- **Canonical:** https://www.gregosuri.com/predictions/energy-ai/2025-04-11-training-power-doubles-every-two-years-900mw-model-badr/
- **Cite as:** Greg Osuri, "Training power doubles every two years — 900MW models by 2028," Greg Osuri, CEO of Akash Network, on why compute should be shared with a decentralized marketplace! (LegendsNLeaders), 11 Apr 2025, 00:07:39. https://www.youtube.com/watch?v=vMx6G0ZOxeo&t=459s

## Quote — verbatim

> If you look at the amount of installed capacity needed to train state-of-the-art models... it's doubling every 2 years. So Grok-3 roughly took about 25, 35 megawatt capacity to train... right now the state of the art is 35 MW; in \[2026\] it will be 70 MW, and \[in\] '28 we're looking at 900 MW... It's nearly impossible to get over 10 MW capacity in America.

## Context — site annotation, not spoken

Quantified forecast of the training-energy wall; grounded in xAI's Memphis data center demanding up to 150MW. (Caption garbles "2026" as "2006".)