Introducing Spend Management: Spending limits and alerts for Cloudflare
Source
blog.cloudflare.com
Date
Why it matters
You can now cap Workers and Durable Objects spend so Cloudflare pauses activity at the limit instead of only alerting. Combined with AI Gateway budgets, this bounds runaway AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → or inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → costs.
Key takeaways · AI-distilled
A cap does not stop everything. Cloudflare says it pauses only customer-code invocations that cost money: in Workers, Cron Triggers and Queue consumers stop, but static assets keep being served. Raising the limit, removing the policy or a new billing cycle lifts the block.
Cloudflare reports per-product spend in near real time, meaning under 10 minutes, so caps act on slightly delayed data. It says it plans to shorten the gap between measurement and enforcement, hopefully to under a minute.
Limits run per billing cycle, which starts when the paid subscription began, not on the calendar month or when you set the cap. A subscription started October 10 generally resets on the 10th.
To make metering scale, Cloudflare aggregates usage in two stages, into roughly one-minute and then five-minute windows, and partitions pricing by account. Usage that arrives before its subscription data is held as pending rather than treated as free.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
