SkyPilot, a startup founded by Databricks co-founder Ion Stoica, has secured $20 million in funding to tackle a growing pain point in the AI industry: the complexity of running compute-heavy workloads across multiple cloud providers. The company says its platform is designed to cut through the vendor lock-in that has become a costly headache for many organizations, especially as GPU prices climb.
Why the funding matters
The funding round, announced this week, comes as AI companies increasingly find themselves tied to a single cloud provider, often facing steep price hikes for GPU capacity. SkyPilot's software aims to let teams deploy jobs on Amazon Web Services, Google Cloud, or Microsoft Azure — or any combination — without rewriting code or manually reconfiguring infrastructure. The idea is to give users flexibility to chase cheaper compute or burst into spare capacity as needed.
Stoica, who also co-founded the Apache Spark project and the Ray framework, said in a statement that the industry is moving toward a multi-cloud reality. “AI workloads are becoming too diverse and too expensive to be locked into one provider,” he noted. “SkyPilot gives teams the freedom to choose where their models run, based on cost, availability, and performance.”
How the platform works
SkyPilot acts as a kind of orchestration layer. Instead of a user manually configuring a cluster on one cloud, the platform picks the best spot — across clouds — to run a job. It handles provisioning, auto-scaling, and cleanup. The tool is open source, though the company offers a managed version for enterprises that want a hands-off experience.
Early adopters include AI labs and startups that need to train large models on expensive GPUs like Nvidia’s H100s. By spreading jobs across clouds, they can avoid the markups that come with a single vendor’s special contracts. SkyPilot also claims to reduce idle GPU time, which is money wasted in a tight market.
The rising cost of compute
GPU prices have surged over the past year, driven by the explosion of generative AI. Companies that once bought capacity on one cloud are now shopping around, but hitting the limits of manual multi-cloud management. SkyPilot’s $20 million raise — from investors including Lightspeed Venture Partners and others — signals that the market sees a growing need for a unified solution.
Neither SkyPilot nor the investors disclosed the company’s valuation. The funding will go toward expanding the engineering team and building out integrations with more cloud providers and GPU types.
Stoica’s history with Databricks, which built a billion-dollar business around simplifying data engineering, gives the startup credibility. But the competition is stiff: established players like Kubernetes and newer tools like Run:ai and Anyscale are also eyeing the same problem.
For now, SkyPilot’s pitch is simplicity. “We don’t want users to think about clouds,” Stoica said. “They should think about their models.”
The company plans to release a major update to its open-source platform later this quarter, adding support for more cloud regions and cheaper GPU spot instances.
