Compute

In an AI context, compute refers to the total computational power spent training or running a model, usually measured in FLOPs. It sits at the center of Scaling Laws discussions: more compute generally helps but with diminishing returns. Frontier labs are now pouring billions into GPU and TPU fleets, while alternative silicon vendors like Cerebras and Groq try to differentiate on efficiency. Compute has crossed from a technical term into mainstream geopolitics and energy policy.