Distributed computing startup Taho Inc. is nothing if not ambitious, setting its sights on replacing Kubernetes as the main orchestration and scheduling layer that sits at the heart of artificial intelligence workloads after raising $3.5 million in seed capital today.

The round was led by “strategic angel investors and industry insiders,” though the company, which was built by infrastructure veterans of Meta Platforms Inc., Google LLC and Snap Inc., didn’t mention any specific names. What it did talk about was specific improvements in terms of AI performance, saying its compute framework can boost workload processing times by up to 100%, while reducing costs to a small fraction of the usual amount.

Taho says it has developed a high-performance federated compute framework that can improve workload efficiency by doubling the effective output of AI hardware such as graphics processing units. The platform does this by replacing bloated infrastructure software and complex runtimes with what the startup says is a faster, simpler and lower-cost way to deploy workloads in any computing environment.

The company takes aim at Kubernetes, the popular open-source software that’s used to orchestrate containers that host the components of modern applications. Kubernetes has become critical to AI applications, thanks to its flexibility, its ability to scale up, and its extensive automation capabilities. It enables more efficient resource allocation for AI applications across distributed environments, while accelerating deployment times.

However, Taho co-founder and Chief Executive Todd Smith says Kubernetes wasn’t actually designed for the demands of modern AI workloads running at large scale, so there’s major room for improvement. He said the startup is not aiming to replace Kubernetes entirely, but rather just its orchestration and scheduling capabilities. Taho runs below the application layer and can plug directly into Kubernetes clusters, containers and the continuous development/continuous integration pipelines they’re already using, he explained.

Taho transforms distributed cloud and edge-based infrastructure into a tightly interwoven fabric that acts like a single intelligent supercomputer that dynamically shares resources in perfect coordination. It’s purpose-built for high-throughput, always-on workloads such as AI training and inference jobs, large language models and simulation pipelines, the startup said.

These workloads are essentially decomposed into discrete tasks that can be distributed across an organization’s available compute resources, no matter where they’re located, then reassembled in real time. The results of its computations persist globally, eliminating redundant work.

“What we do is replace the traditional, centralized orchestration layer for the workloads you put on Taho,” Smith said. “Instead of a single control plane scheduling whole containers, Taho acts as a federated runtime that spreads work as fine-grained tasks across your fleet including cloud, on-prem, and edge. In many deployments Kubernetes becomes plumbing and packaging, while Taho becomes the primary execution fabric that delivers the big performance and cost gains.”

Those performance gains and cost reductions can be significant, with some compute jobs finishing up to 10 times faster and costs reduced by up to 90% in some workloads. What’s more, teams needn’t put in much work to achieve these gains, for Taho is compatible with existing software frameworks and continuous integration/continuous deployment workflows.

Smith said the company’s goal is to introduce a new paradigm for infrastructure compute to support the “parabolic demand growth” that’s driven by the demands of AI and other high performance workloads. “We all see AI workloads are exploding, but infrastructure buildouts cannot keep pace,” he said. “We started Taho because the world needs a better way to compute – one that’s simple, fast and affordable enough for every AI-driven company to grow profitably.”

Taho said the capital from today’s round will be used to hire new staff in key areas and expand its early customer deployments ahead of a broader rollout of its platform next year.

Image: Taho