Businesses that once used AI tools under flat-rate subscription models are now facing rising costs as providers switch to usage-based pricing. On 14 May, Anthropic announced it would separate agentic usage from its Claude subscription, introducing a metered credit pool starting 15 June. GitHub Copilot will also transition to a token-based AI Credits model on 1 June. These changes are part of a broader trend where companies are moving from fixed monthly fees to models that charge based on actual AI usage, measured in tokens—units of text processed by AI systems. Some businesses are seeing their bills increase by up to 27% due to changes in how AI models process data, even without updates to public pricing pages. The shift in how AI usage is measured has created unexpected challenges for enterprise procurement teams. Most CFOs were not informed about the changes, and some companies have already exceeded their AI budgets. For example, Uber reportedly used up its 2026 AI budget by April, forcing its CTO to reconsider AI strategies. KPMG’s recent survey found that U.S. enterprises expect to spend an average of $207 million on AI over the next year, nearly double the amount from last year. Goldman Sachs data shows that many large companies are significantly overspending on AI, with Salesforce CEO Marc Benioff estimating his company’s Anthropic bill could reach $300 million this year. He has called for a "smart router" to better manage AI usage. The shift in pricing models is creating challenges for finance teams, who often lack the infrastructure to manage fluctuating costs. Unlike electricity, which has well-established metering and regulation systems, AI consumption currently lacks similar frameworks. Many companies are struggling to track AI spending, which is often scattered across departments and not centrally monitored. Finance teams often only see the full picture of AI spending when the data is compiled, leading to last-minute budget adjustments. Some businesses are restricting access to advanced AI tools due to cost concerns, not because the technology is ineffective. To better manage these challenges, buyers are advised to include clear provisions in contracts, such as notice of price changes, stability clauses for token measurement, audit rights over consumption data, and exit and portability terms. Many existing contracts, written in 2023 or 2024, did not account for the rapid changes in AI pricing models. As well, some temporary high-capacity AI usage options, like OpenAI’s Codex trial and Anthropic’s 50% capacity boost, are set to expire in mid-July. This leaves companies with a narrow window to take advantage of subsidized usage before bills for these costs arrive in August. Those that prepare well—by tracking AI spending, building strong supplier relationships, and structuring contracts like utility services—will be better positioned to navigate this new landscape.