A daily benchmark for LLM spending has fallen from its spring peak, yet the reason for the decline is still unclear.
The fight over cheaper model access is less about a single token bill than about who owns routing, policy, and dependency control across AI deployments.
A University of Toronto proof-of-concept shows how a self-replicating worm can use a locally hosted open-weight model to choose its next move without human intervention.
The 2025 AI race is being shaped by three different forces: private money, open-weight model distribution, and a more demanding regulatory perimeter.
Beijing’s open-source LLMs are surging past Western rivals, but new power brings new risks-and a new era of AI geopolitics.