Compute Forecast — AI 2027
AI 2027 predicts AIs trained with 1000x more compute than GPT-4 and the internal deployment of hundreds of thousands of AI research assistants by 2027. This supplement introduces the compute production model and the inference compute model behind these predictions.
Compute Forecast — AI 2027 Compute Forecast Romeo Dean | April 2025 ai-2027.com Figure 1: The training runs in AI 2027 , informed and justified by our compute production , compute distribution and compute usage sections. Summary Compute is one of the key inputs to AI progress. In this supplement we only consider AI-relevant compute . We frequently refer to H100-equivalents (H100e) throughout the supplement, which we define as AI-relevant compute with processing performance converted into units of the Nvidia H100 GPU. Processing performance is most directly relevant to training compute, but is
related reading
- AI 2027ai-2027.com
- AI 2027ai-2027.com
- I. From GPT-4 to AGI: Counting the OOMs - SITUATIONAL AWARENESSsituational-awareness.ai
- AI 2027ai-2027.com
- Can AI scaling continue through 2030? | Epoch AIepoch.ai
- Frontier labs don’t use most AI compute (yet) - by Josh Youepochai.substack.com
- AI progress is about to speed up | Epoch AIepoch.ai
- Supplements — AI 2027ai-2027.com
- Navigating the High Cost of AI Compute | Andreessen Horowitza16z.com
- IIIa. Racing to the Trillion-Dollar Cluster - SITUATIONAL AWARENESSsituational-awareness.ai
- Why compute might get 10x more expensive in coming yearsdwarkesh.com
- My picture of the present in AI — LessWronglesswrong.com