Hello everybody. In this video, I would like to show you how you can optimize your machine learning costs on SageMaker using savings plans. Savings plans is a cost optimization feature that has been available for a few years on Amazon EC2 and compute in general, and now it's also available for SageMaker. Let me show you how to get started. First, you go to the cost management console and select the savings plans entry. If you're not familiar with what savings plans are, there's an overview here, and we'll see how those plans actually work, how to select the hourly commitment on a one-year or three-year period, how much upfront you want to pay, etc. If you're familiar with savings plans on EC2 and compute, this is really similar, so you won't be thrown off. Feel free to read all of this to become familiar with savings plans.
This is the good part: I can save a bit on compute and EC2, but it looks like I could save a lot on SageMaker. I can click on this, and it takes me directly to the recommendations. Recommendations look at my past SageMaker usage and recommend a plan that maximizes my savings. I can select this plan or create my own, as I will also show you. Let's try to make sense of this plan. Make sure you select SageMaker. This is the one we want. The first parameter is the term. The basic mechanism here is that I'm going to commit to a certain level of usage and can decide to commit for one year or three years. Let's start with one year. Then, I decide whether I want to pay upfront for this usage. I'm going to pay in advance for that commitment over the one-year period. If you pay all upfront, you'll get a better discount. You can select partial upfront or no upfront, meaning you don't pay anything at all. Partial upfront means at least 50%. Let's try with no upfront first.
The time period on which the recommendation is defined should reflect your future usage. If you've just started using SageMaker or have been ramping up massively, maybe you've been experimenting for a few weeks and now you're in production at a large scale, then seven days is probably good. If you've already been using SageMaker over a longer period, maybe one month or two months, feel free to select those. Again, past usage should look like your future usage. Let's just say I'm using SageMaker all the time, so let's pick 60 days. Scrolling down, I can see how much I'm currently spending on SageMaker on-demand, a little more than five thousand dollars per month, which means 7.3 dollars per hour. If I use the recommended plan, I will spend 4.4K, which means about 6 dollars per hour. I'm going to save about just under 900 dollars per month, which is about 17 percent savings compared to on-demand. By just committing to a certain level of usage over one year and no upfront, I'm saving 17 percent just like that. Our commitment here would be 2.79 dollars per hour. As you can see, we leave out an average of 228 dollars per hour for extra on-demand. We're not optimizing 100% of the existing on-demand usage because it's likely there will be some variations. If your usage drops a little bit compared to those 60 days, your commitment would be too high. We go for a slightly lower commitment and keep that buffer on-demand spend to account for variations to avoid over-committing.
Let's try three years. Three years is a longer commitment, so we would get 36 percent savings now, which is really good. We could see the updated commitment here, of course. We could experiment with upfront payments and see how far we could go. If we decided to pay everything upfront, then we could go up to 40 percent savings. It's really up to you to decide if you want to pay in advance. Some organizations prefer upfront payments for budgeting purposes, as it makes it more predictable and easier to track their spend. Some organizations will prefer no upfront because they're happier that way, just wanting to minimize upfront payments. It's quite flexible, and you can find the plan that works for you. Let's say I want to go for one year and no upfront. I'm already happy with that. I could just add the plan to the cart and go to my cart here. I would just purchase this, send my order, and that's it. There's no upfront payment here, so I don't need to pay anything now. Every month for a year, based on my commitment, I will be charged 2,700 and something. That really corresponds to this commitment, this hourly commitment over a one-month period. Every month, I will pay this and also any extra on-demand usage. Remember, we kept that buffer, so if you use more than your commitment, this will be charged at the on-demand price.
Now, let's say I'm interested in trying my own plan. Instead of using those recommendations, I want to create my own. I can see my current on-demand spend is 6.54. Let me go back to 60 days, so we'll have the same number, which is 7.3 dollars per hour. I could go to purchase savings plans and select my own. I could say, okay, I want to do a one-year plan and maybe I expect my usage to drop or maybe I'm a little bit conservative, so I'll just go and say, okay, I just want to commit for two dollars per hour. I don't want to commit too much; I'm not so sure what my SageMaker usage is going to be in the next year. Two dollars, I can commit to more, but I'm not so sure. I want to do partial upfront, so I'm ready to pay for half of the plan right now, which turns out to be 8,700 something dollars. I could set a start date, so I could say, all right, I want this to apply on May 1st. I can see my summary: my upfront cost is this, which is 50% of the total commitment over a one-year period, then my monthly payment would be 730 dollars every month for a year. The total cost of the plan would be 17,000 something dollars. That's the flexibility you have. You can either use the recommendations and maximize your savings with those, or you can build your own and commit to the level you want. You can go as low as 0.001 dollars, a super low commitment, and build the plan that you want. Then add it to the cart, pay for any upfront you've selected, and start enjoying your savings.
As you can see, this is a really simple and efficient way to optimize your costs. No upfront over one year, and I could save 17% already, which is pretty cool. I hope this is useful. Try it out, send us feedback, ask us questions, and start saving money. Thank you, bye.
Julien Simon is the Chief Evangelist at Arcee AI
, specializing in Small Language Models and enterprise AI solutions. Recognized as the #1 AI Evangelist globally by AI Magazine in 2021, he brings over 30 years of technology leadership experience to his role.
With 650+ speaking engagements worldwide and 350+ technical blog posts, Julien is a leading voice in practical AI implementation, cost-effective AI solutions, and the democratization of artificial intelligence. His expertise spans open-source AI, Small Language Models, enterprise AI strategy, and edge computing optimization.
Previously serving as Principal Evangelist at Amazon Web Services and Chief Evangelist at Hugging Face, Julien has helped thousands of organizations implement AI solutions that deliver real business value. He is the author of "Learn Amazon SageMaker," the first book ever published on AWS's flagship machine learning service.
Julien's mission is to make AI accessible, understandable, and controllable for enterprises through transparent, open-weights models that organizations can deploy, customize, and trust.