Compute Gap
The disparity in AI training compute between US frontier labs and Chinese labs.
The compute gap refers to the substantial disparity in access to AI training infrastructure between US-based frontier labs and Chinese AI organizations. American frontier labs have secured compute resources orders of magnitude larger than their Chinese counterparts, creating a structural advantage in the race to develop transformative AI systems.
The gap manifests in several ways. US companies like OpenAI, Anthropic, Google, Meta, and xAI have committed tens of billions of dollars to compute infrastructure. Individual data center commitments exceed 10 gigawatts of capacity in some cases. In contrast, Chinese labs operate under significant compute constraints due to both export controls and domestic infrastructure limitations.
The causes are multiple. US export controls restrict China's access to the most advanced chips. US companies have built relationships with chip manufacturers like Nvidia that give them preferential access to new hardware. And the scale of investment required for frontier training runs has created a winner-take-most dynamic that advantages already-leading labs.
Chinese labs have responded by focusing on efficiency. Research from Exponential View's May 2026 report found that Chinese labs have developed techniques to achieve comparable results with substantially less compute, creating what researchers call an efficiency moat. However, efficiency improvements cannot fully substitute for raw compute in all types of AI research, particularly training new foundation models from scratch.
Research this in Signals
Signals turns a topic into a sourced research record you can inspect and rerun. Your first scan is free, and this one starts with Compute Gap already loaded, so edit it or scan as is.