The physical and computational resources, including data centers and GPU clusters, required to train large-scale artificial intelligence models.
Overview
- SpaceX possesses a training supercomputer called Colossus with a capacity of one million H100 equivalent GPUs. (Source: SpaceX announcement on X, via AI Daily Brief, 2026-08-24)
- [Emerging signal] OpenAI’s GPT Image 2.0 release signals that the company is leveraging significantly increased compute resources for model training, potentially foreshadowing similar improvements in other model families. (Source: AI Daily Brief Host, via AI Daily Brief, 2026-08-24)