At FinOps X 2026 in San Diego, the Tokenomics Foundation was announced — a vendor-neutral body inside the Linux Foundation focused on governing AI token economics. The core insight: the real cost of AI infrastructure isn't the model provider invoice, but what happens upstream in GPU clusters, autoscaling decisions, and agentic retry chains. SAP data showed 12 months of exponential token growth with total spend still doubling despite falling per-token costs. The author argues that only ~15% of developer tasks genuinely require frontier models, and that infrastructure-layer automation — GPU rightsizing, spot instances with live migration, burst autoscaling, and multi-cloud routing — is the path to sustainable AI spend governance. Cast AI is also building a new product called Kimchi targeting token consumption optimization.
