Using AI to Automatically Optimize Cloud Costs

The Birth of Cloud Computing

Back in the early days of cloud computing, if we would tell you that every business and consumer would eventually be running their entire compute and storage infrastructure digitally, we would sound crazy. But just 17 years later, it seems inevitable that the old world of data centers and on-premise servers will go the way of the physical newspaper and radio.

Indeed, even Andy Jassy, CEO of AWS, didn’t envision this future back in 2000. At the time, Amazon was simply struggling to scale despite hiring a lot of new developers. Jassy, who was Amazon’s Chief of Staff at the time, dove into the problem to discover that the hardware and infrastructural needs simply were not being met fast enough. Jassy set out to speed up this process internally, allowing the internal R&D team to access a pool of common compute and storage resources, making it easier for each to launch their own projects without needing to deal with the underlying infrastructure. Eventually, at a team summit in 2003, they realized this was a need for every other company on the planet, and so AWS was born.

The birth of Spotinst: Rising cloud costs

Today, there is a similar revolution beginning in the world of cloud computing. One still in its early stages, but seemingly destined to grow just as quickly as AWS. “Spotinst aims to be the largest provider of cloud infrastructure without running any servers,” says Amiram Shachar, CEO & Founder of Spotinst. The statement is bold on its own, but when you dive a bit deeper, not much bolder a claim than had you asked Andy Jassy the same back in 2003.

Just as Jassy found an internal struggle today, Spotinst began with Shachar’s identification of a problem at his company, then Lycos Global Group. As Director of DevOps, Shachar was pressed with the task of migrating all Lycos’ servers to the cloud, but costs quickly jumped. “Within a couple of months of moving to the cloud, our costs had actually risen dramatically, way beyond any calculator that we used to predict it” continued Shachar. “To the point where we were actually planning to move everything back to on-premise.”

Just around that time, Shachar encountered the relatively unused offering at AWS, called Spot Instances. Spot Instances were spare server capacity at AWS, cleverly packaged to enable monetization of these servers.

AWS offered these idle servers for massive 90% discounts with one major caveat – they’ll kick you off once a customer wants to pay full price for those servers. Shachar smelled opportunity. “By building algorithms that could predict the spare capacity terminations in advance, we could then migrate to another spare server beforehand, saving us 80% on our cloud costs.” After a few months, Shachar and his team were running nearly 75% of their compute power on AWS spare capacity. Just like Jassy back in 2003, the lightbulb eventually went off – there was a far larger opportunity at hand here.

The virtual cloud: run code without dealing with cloud providers

“All companies running on cloud struggle to keep costs down,” Shachar says. “But with a long list of priorities including security and maintaining site reliability, DevOps teams often don’t have the time to do all of the work required to optimize cost.” But while Spotinst’s first product was aimed at helping companies reliably leverage spare capacity to reduce costs, Shachar and his co-founders, Liran Polak and Aharon Twizer, had a bigger vision.

Today, managing cloud infrastructure has become just as complex as running on-premises. With a plethora of new AWS, Google, and Azure offerings launching every month, along with an increasingly vast list of companies that build off these offerings, figuring out the balance between cost and performance is a major challenge. But, Shachar envisions a world where you don’t have to.

The promise of the virtual cloud

The promise of Shachar’s vision is simple at heart – a virtual cloud. Spotinst will leverage machine learning to ensure that your applications are always running on the most cost-effective servers, given your performance requirements. Instead of exploring the long list of alternatives within and across the different cloud providers, Spotinst will take care of it all under the hood. Today, their platform recognizes any workload that could be run on a cheaper server. Just put those workloads under Spotinst’s control and their algorithms will take care of the rest – those workloads will forever run on the cheapest servers available. All without compromising performance or availability.

Using a new Serverless offering, where developers can simply write code without dealing with any of the underlying infrastructure, Spotinst aims to deliver on this promise in a major way. “All you should ever have to do is write code,” says Shachar. “Our algorithms will make sure it is always running on the most cost-effective infrastructure, across all cloud providers, for your performance and availability needs.” In other words, just write the code, and Spotinst will make sure it always runs at the lowest cost possible for your performance needs.

As Shachar says, “in 10 years, just as everyone is making the shift to cloud today, they will be making the shift to Serverless computing. Only with Spotinst, they’ll be able to also ensure they are paying the best possible price, all without doing anything.” The promise of companies all running a perfectly optimized cloud without ever dealing with the infrastructure seems bold. But as we learned from Jassy and AWS, in another 10-15 years, it may just be an inevitability

About the CEO

Amiram Shachar is the CEO & Founder of Spotinst, which leverages excess cloud capacity to help customers save ~ 80% on cloud costs. Before Spotinst, he was Director of DevOps at Ybrant Media and led the Israeli Defense Forces migration to the cloud.