People often ask us how Usage (usage.ai) works. We delayed answering that for a while because AWS has changed so much over the years, and we had to evolve our business with it. Now that things have stabilized, here’s how Usage actually works.
Let’s walk through the Usage story from the beginning. How we started, where we are today, and where we are going. With this information, you could build your own AWS, GCP, or Azure optimization strategy. But you do not have to, since our automated dashboard is free forever.
A Reserved Instance is a discount mechanism, not a physical instance. It can save you up to 57 percent with zero upfront capital if you are willing to make a long-term commitment. A Savings Plan works in a similar way but is based on spend rather than a specific configuration. Both RIs and SPs exist on AWS, GCP, and Azure. For RIs, you commit to a number of normalized units for a specific instance type and region for one or three years. In return, you receive a discount for as long as the normalized units are being used. If utilization drops, your savings shrink back to on-demand levels. If there is no utilization, you end up paying more than you would have on-demand. There is real risk in buying these commitments.
For example, if you purchase a three-year no-upfront m5.xlarge RI and you have two m5.large instances running in the same region, that RI will fully cover both instances and give you the full 57 percent savings. If you shut down one of the m5.larges, only half the discount is applied. If both are off, the RI is zero percent utilized and you are still paying for something you are not using.
Savings Plans operate the same way. To get full savings on an m5.xlarge SP, you might commit to spending 5 cents per hour for three years. The difference is that SPs are flexible. As long as the value matches, that m5.xlarge-equivalent SP can apply to any instance family or region. You could be running two r5.large instances, and if the spend lines up, you still get the full benefit.
We started Usage four years ago with AWS EC2 RIs. We chose EC2 RIs because they offered a unique advantage. You can purchase one for one or three years and sell it on the RI Marketplace. Even better, if you choose no-upfront, you avoid paying sales tax, which only applies to upfront payments. That means you can buy a three-year no-upfront RI, get the full 57 percent discount, and sell it later if your needs change. As long as your instance type and region are in demand, there is usually a buyer.
But the RI Marketplace is not as strong as it used to be. In January 2024, AWS made changes that hurt liquidity. They stopped customers with Enterprise Discount Programs from selling RIs and limited how many RIs a customer can sell during the life of their account. Marketplace activity dropped significantly. Transactions still happen, but far fewer than before.
Now that you understand how RIs and SPs work across AWS, GCP, and Azure, how do you actually start saving?
Choosing what to buy is not simple. Each RI needs a configuration. Each SP needs a dollar amount. Once you start combining them across providers, it gets complex quickly.
There are two places to go for help. First, all major cloud providers offer RI and SP recommendations in their own dashboards. Second, the free Usage dashboard refreshes daily and tells you exactly what to buy. If you enable autopilot, Usage will even purchase the commitments for you.
When you are ready to commit, you need to decide on coverage. Do you want to cover 80 percent of your on-demand usage? Or 100 percent? And for how long? One year or three?
The right answer depends on your risk appetite. If you are confident your infrastructure will stay the same or grow over the next three years, and that there is no chance of major changes like rightsizing or moving to serverless, go with 100 percent coverage for three years. You will get maximum savings. If you think changes might happen, you could choose 80 percent coverage. You will save a bit less but keep some flexibility. If you only feel sure about the next year, commit for one year. At 100 percent coverage for one year, you will save around 20 to 25 percent. That is lower than the 50 to 57 percent you get from a three-year commit, but the risk is lower too.
You get the idea. The more risk you can handle, the more you can save. Large companies have entire teams of data scientists just to forecast three years of cloud usage.
Be careful not to overcommit. We have seen many customers lose millions this way. Some have lost tens of millions. Always coordinate with Engineering, DevOps, and leadership to make sure your three-year outlook is clear before locking anything in.
Once you buy your first commitment, your work is not over. If your environment is large, you probably bought several types. For compute, you may have a mix of RIs and SPs. For things like RDS, Redshift, ElastiCache, and OpenSearch, you only have RIs available.
You need to track expiration dates. Once an RI or SP expires, your pricing defaults back to on-demand and your bill goes up. The best approach is to check your expiring commitments every week. Each cloud provider shows this in their console. Usage’s free dashboard shows your entire inventory across every provider in one place.
As expiration dates approach, talk to your team about what you will need in the next one to three years and purchase accordingly.
Buying commitments is easy. So is renewing them. The real challenge is managing risk. Staying on-demand gives you total flexibility but the worst pricing. Committing all your spend for three years gives you the best pricing but the most risk.
Here are common patterns we see:
Eighty percent coverage with one-year terms. This gives you room to adjust and keeps savings around 20 percent.
Sixty to eighty percent coverage with three-year terms. This is more aggressive. It gives you some flexibility and brings savings to around 35 to 40 percent.
As your confidence grows, you can increase coverage and get closer to maximum savings.
With Usage, you get full automation. We purchase three-year no-upfront commitments on your behalf. You get 50 percent savings. If anything goes underutilized, we send you the money back through our cash-back program. Your savings are maximized, your risk is zero, and your effort is minimal. In a separate blog post, we’ll go into more detail on how Usage manages this risk across its book of customers, backed up by numbers, including a key metric called Loss Ratio and how that’s evolved over time.
Share this post
