New AI Cuts Cloud Bills 45% by Shopping Amazon, Google, Microsoft

Every time you open an app, somewhere a company is renting computers from Amazon, Google, or Microsoft to run it. Those three charge different prices, offer different discounts on unused capacity, and go down at different times. Most companies pick one and stay there, which means they overpay. HELIOS is a new AI system from researchers Guanyu Ding and Ying Wang that juggles all three at once, always hunting for the cheapest safe spot for each job.

The clever part is that the AI doesn't make split-second calls itself. Asking a large language model — the kind of AI behind chatbots — to approve every single decision would be far too slow and expensive. Instead, HELIOS lets the AI write small rulebooks in advance, tests them in a simulated cloud, and keeps only the winner. That winning rulebook then runs live, responding in milliseconds.

The results are striking. Across real workload data and 2026 prices from all three providers, the AI cut penalized operating costs by 45% compared to sticking with a single cloud, and beat hand-tuned multi-cloud strategies by 9–19%. Safety guardrails turned out to be essential: under extreme stress testing, an unguarded version of the same AI let premium services go down 97–98% of the time, versus just 0.17% with the guardrails on. The team released their simulator and prompts so others can check the work.

So what does that mean for you? Cloud bills are a huge hidden cost inside nearly every digital service you pay for. If providers can run things 45% cheaper, some of that savings can reach your subscription price. Just as importantly, fewer outages means your payments, maps, and messages keep working when one provider stumbles.