Per-token AI pricing has dropped 98% since 2022. Enterprise AI bills have tripled in the same period. Managing consumption has become an operations problem.
Meta sent an internal memo to 6,000 employees in late June warning that AI usage costs were approaching billions. The company tracked consumption on an internal leaderboard called “Claudeonomics.” Employees consumed 73.7 trillion tokens in roughly 30 days. Meta is now dismantling the leaderboard and replacing it with a centralized monitoring platform called AI Gateway to track usage and spending across teams in real time.
Uber reportedly exhausted its 2026 AI coding budget by April across roughly 5,000 engineers. Monthly costs per engineer ranged from $500 to $2,000 for power users. Uber responded with a $1,500 per month cap per tool. Walmart, Amazon, and Cisco followed with similar controls.
The FinOps Foundation’s 2026 State of FinOps report found that 73% of enterprises reported AI costs exceeding original projections.
The 4,500x problem
The model-price comparison cited here puts the range at roughly 4,500x. That comparison needs a date, token basis, and named models before it can support a purchasing decision. For a local estimate, compare the cost of completing the same task at the required quality.
The term “token maxing” describes heavy consumption, but a usage count needs task context. Autocomplete and parallel agents working across a codebase can have very different consumption patterns. Track the model, task type, retries, and completed work alongside the token bill.
The cited estimate puts agentic workflow consumption at 5 to 30 times that of a standard chatbot interaction. The task definitions behind that range need checking. A budget based on occasional chat use should be retested before supporting long-running agents.
Average annual enterprise AI budgets grew from roughly $1.2 million in 2024 to $7 million in 2026. Per-developer usage is up 18.6x in under a year. Those figures make consumption assumptions worth revisiting alongside unit prices.
Apply consumption controls to the workload
Cloud cost management offers a useful comparison. A development environment running 24/7 for a workload needed only 8 hours a day illustrates why usage duration matters alongside unit price. This hypothetical example illustrates a cost-control problem. Claims about a particular company’s cloud bill between 2018 and 2022 would need separate evidence.
FinOps practices such as cost allocation, usage policies, spend alerts, and tagging offer ways to make that consumption visible.
For AI, model routing policies, usage tiers, and spending dashboards provide ways to inspect consumption and intervene. They do not guarantee that a team will remain within budget; check alerts and enforcement against actual use.
What the spending caps actually reveal
The reported Meta and Uber responses are reasons to review routing and budget assumptions. They do not establish that a single missing control caused either company’s spending. Separate what the reports say about consumption from an explanation of why it happened.
Assign someone to review team-level costs and decide when to change routing or access. Include a process for approving expensive work that is justified by its contribution.
Give cost governance an owner
Before increasing the budget, name the person who can inspect consumption, enforce limits, and approve exceptions. Agree on what will be measured: completed work, review effort, and total cost as well as token use.
Test an alert and a spending limit during the trial. Finance and the workflow owner should both be able to see what happened and what action remains available.