What's going on

Every major cloud rewards commitment. AWS Savings Plans, Azure Reservations and Google Cloud committed use discounts all trade a lower rate for a promise: you'll spend a set amount for 1 or 3 years. That works when usage is steady. AI usage rarely is.

Why AI makes it harder

• Token volume moves fast. Every launch, new customer or feature can change it.
• GPU needs shift. Training, fine-tuning and inference each need different capacity, and a model swap can change all of it. We dug into this in The GPU Blind Spot in Cloud Commitment Planning.
• Projects move at startup speed. A pilot can hit production overnight. It can also get shelved.

So teams pick a poison

Over-commit. Lock in for years, then keep paying for capacity you stopped using when plans changed.

Under-commit. Stay on-demand and pay full price for workloads that could cost far less.

Most teams pick the second one. It feels safe. It's also an expensive choice, and it adds up fast when GPU-heavy AI workloads are on the bill.

The fix: Guaranteed Commitments

A Guaranteed Commitment is a standard cloud commitment with a money-back guarantee on top. Under the hood, it's a regular Reserved Instance, Savings Plan or CUD, bought and managed in your own cloud account. On AWS we call them Guaranteed Reserved Instances (GRIs) and Guaranteed Savings Plans (GSPs). On Google Cloud they're Guaranteed CUDs (GCUDs). 

Here's what you get:

• Short terms. Pick a 30-day or 1-year guarantee window.
• Real savings. Up to 46% off on-demand pricing, with terms as short as 30 days.
• A safety net. After the guarantee window, if your usage drops, you get money back for the unused part. The guarantee is insured by an AM Best-rated carrier.

Think of it as insurance for individual cloud commitments. You get the discount, and Archera takes on the risk of guessing wrong.

What it looks like for AI teams

• Cover GPU and AI compute. On AWS, Guaranteed Commitments cover GPU instances and SageMaker Savings Plans. On Azure and Google Cloud, they cover the virtual machines your GPU workloads run on. Check what's eligible in the price catalog.
• Commit for 30 days while you test. Get committed rates during a pilot. If the project slows down after the first 30 days, the unused part is covered.
• Hold capacity longer. AI capacity is limited and costly to hold onto. Archera lets you keep it longer at a better rate.

Floyo, which runs GPU-heavy AI workloads, put it this way:

"Teams no longer have to worry about over-committing or being stuck on the wrong instance types. It's simplified our FinOps workflows tremendously."

-Ashish Gupta, Head of AI Infrastructure, Floyo.

‍Read the Floyo story.

Another team saved over $100K on AWS GenAI workloads with Archera.

How it works

1. Connect. Link your AWS, Azure or Google Cloud billing. Setup takes about 5 minutes, with no code changes.
2. See your options. Archera analyzes your usage and shows coverage, cost and savings for each plan before you buy.
3. Apply the plan. Commitments are bought and managed in your own account. No reselling and no bulk buying.
4. Stay covered. If usage drops after the guarantee window, you get money back for the unused part.

Want answers without digging through dashboards? Archera AI answers questions about your spend, coverage and savings in the Archera app, in Slack, or in your own AI tools.

What it costs

The Archera platform is free. That includes cost visibility, forecasting and managing your native commitments. You only pay a premium on Guaranteed Commitments, as a share of the savings they create, and you'll see your projected net savings before you buy. You can buy through the AWS, Azure, and Google Cloud marketplaces. See pricing.

Bottom line

AI is unpredictable. Your discount doesn't have to be a bet. Get long-term pricing with terms as short as 30 days, and let Archera carry the risk.

‍

‍See how other teams do it.

‍