Cloud AI bills make companies scrutinize their AWS, Azure and Google deals

If your company uses AWS, Azure or Google Cloud for AI, there's a good chance the old cloud agreement no longer fits reality.

The problem starts with access to computing power. Several IT leaders have found that they have an AI model ready for production but can't get the GPU capacity the agreement assumed. The market price has outrun the budget.

Then there's data transfer. When an AI workload has to cross regions or providers to find available computing power or a cheaper chip, the bill can rise sharply, without any more users or new systems. It's the architecture, not the usage, that drives up the price.

It also makes it more expensive to switch providers later. If you've consolidated everything with one cloud vendor, it can be costly in egress fees to move data out again if the vendor can't deliver the right capacity. That leaves you in a weaker negotiating position precisely when you need alternatives most.

It ends up as a question for the board: what will AI cost next year? Some IT leaders can't answer with certainty, because the contract is built for a different type of usage than the one the company actually has today.

The companies in the strongest position have done one thing before negotiating: separated AI economics from the rest of the cloud bill and mapped usage per workload, region and chip type. That provides a basis for negotiation, not just another high-level estimate.

Concretely, this means new types of agreements: reserved capacity for a specific GPU type instead of general compute credit, shorter commitment periods for the volatile AI needs, and longer agreements for ordinary cloud usage alongside.

Rule of thumb: if your company is on its way to using AI in production, the work of separating AI usage from the rest of the cloud budget should start before the next contract negotiation or renewal is on the table, not when the offer has already been written. For most private individuals and smaller companies without their own cloud infrastructure, this doesn't change anything in daily life right now.

What it means for you

If your company has AI projects in the cloud, you should review the agreement now, before the next negotiation or renewal. Separate AI usage from the rest of the cloud budget, and map what you use per workload, region and chip type. That puts you in a stronger position when you negotiate reserved GPU capacity or shorter commitment periods. If you or your company don't use cloud infrastructure yourselves, this doesn't change anything in your daily life right now.