Google Cloud spend caps stop AI use at the budget limit

Google Cloud made spend caps on Google Cloud Budgets available to all customers. A spend cap stops use of a system when the cost is at the limit. Google tells that caps for AI systems work in minutes, with estimated live costs. The release adds an API (a method for software to send commands), live cost reports and caps that start use again each month. Google gave no test results for the speed.

What is new

Three changes come with the release:

• The spend caps API lets software set, lift, update or remove caps.
• Cost reports show the live estimated costs that caused a stop.
• Caps start use again at the start of each month. A manual override stays.

What a cap includes

Caps stop requests that have a cost for each use. They also stop requests that a committed use discount includes. Provisioned Throughput, which is reserved capacity, does not stop. Alerts go to billing administrators and project owners at 50%, 80% and 100% of the cap.

What is not known

Caps use estimated costs, not the costs on the last bill. Google gave no data on how fast a cap works. Google tells you to start with developer sandboxes and test clusters.

Insight: Caps protect test spend, not reserved capacity

Google Cloud will open A5X instances with NVIDIA Rubin chips, and it has a Claude deployment that Anthropic operates. Its new caps stop spend for each use, but do not stop reserved capacity. This reads as a protection for test work, not for committed spend. The contract is where you set the correct limit. Leaders get a control for sandboxes only.