News
Public · Published
By 2028, training the latest AI model could require a gigawatt-scale data center roughly the output of a nuclear reactor. @gregosuri explains why the answer may be distributed training: moving compute to where energy already exists. #akash #ai #crypto #depin
By 2028, training the newest AI models could require a gigawatt‑scale data center, roughly the power output of a nuclear reactor, and the proposed solution is distributed training that places compute where energy is already available.
Published:
Updated:
What happened
By 2028, training the newest AI models could require a gigawatt‑scale data center, roughly the power output of a nuclear reactor, and the proposed solution is distributed training that places compute where energy is already available.
Confirmed
Global impact / market context
If AI training shifts to distributed sites with existing energy, companies can avoid huge capital outlays for new power plants, lower operating costs, and speed up model development, benefiting investors in both AI and energy‑focused assets.
Analyst inference
AI model training is becoming so compute‑intensive that by 2028 it may need a data center the size of a gigawatt power plant, comparable to a small nuclear reactor’s output.
Confirmed
What to watch
- Adoption of distributed training platforms like Akash that locate compute near existing power sources, potentially reducing the need for dedicated gigawatt data centers. Analyst inference
- Regulatory developments around large‑scale energy use for AI, especially policies affecting nuclear‑level power consumption and renewable‑energy integration. Analyst inference
- Capital spending trends of cloud providers and AI firms as they decide between building massive centralized facilities versus leveraging decentralized, energy‑rich locations. Analyst inference