Modal Labs raises $80M to simplify cloud AI infrastructure with programmable building blocks – News

Cloud computing startup Modal Labs Inc. has raised $80 million in Series B funding and achieved “unicorn” status as it strives to become the infrastructure platform of choice for every stage of artificial intelligence development.

Today’s round was led by Lux Capital and saw the participation of several existing investors, bringing Modal’s total amount raised to $111 million and its valuation to $1.1 billion.

The startup has built a serverless AI infrastructure platform that’s designed specifically for building and scaling AI models, applications and services. Because it’s serverless, developers don’t have to worry about managing the underlying servers, virtual machines, virtualization environments and everything else. Instead, they can access fast and reliable high-performance graphics processing units from Oracle Cloud Infrastructure and benefit from instant autoscaling, where the platform automatically ensures every workload has adequate compute resources.

In a blog post announcing today’s round, Modal co-founders Erik Berhandsson and Askshat Bubna, who serve as Chief Executive and Chief Technology Officer respectively, said cloud-based compute infrastructure is currently undergoing a significant transformation as it evolves to better suit the needs of AI developers. They explain that legacy clouds were not designed for modern AI, leading to problems in terms of GPU capacity management, highly variable demand that makes it expensive to reserve instances, and many more.

“As developers, we saw a massive opportunity in rethinking infrastructure for these new-age needs,” the executives wrote.

Modal’s aim is to make the developer experience as simple as possible, and one way it does this is by allowing all of their training and inference infrastructure requirements to be defined using programmable building blocks for storage, compute and networking, so there’s no need to mess around with complex configuration files. Users can aggregate both GPU and central processing unit resources around the world, so that they never have to concern themselves with capacity management. It also supports sub-second startup times, enabling “fast cold starts” for any workload.

Besides providing infrastructure, Modal’s platform offers services including Sandboxes for developers to quickly spin up secure and dynamically-defined environments to test new AI models and agents. It supports batch workloads for customers that want to run massive parallel AI applications such as batch transcription, protein folding and weather forecasting. Moreover, there are notebooks for developers to collaborate on exploratory data analysis.

“Building this wasn’t easy, and we had to go very deep in order to lay the foundation we needed,” said Berhandsson. “This included building our own file system, container runtime, scheduler, and much more. But as a result, we now have thousands of customers running complex applications who agree that we offer an unparalleled experience.”

Those customers include some of the leading lights in the AI industry, such as Meta Platforms Inc. and the data labeling specialist Scale AI Inc., which was recently acquired by the social media giant. Meta notably used Modal’s infrastructure to run its new “neural debugger” tool, Code World Model, spinning up thousands of concurrent sandboxed environments for reinforcement learning. Meanwhile, Scale AI relies on Modal to handle the enormous spikes in volume it sees, as well as its Model Context Protocol servers that orchestrate AI agents.

“Everyone here loves Modal because it helps us move so much faster,” said Scale AI Vice President of Engineering Aakash Sabharwal. “Whenever a team asks about compute, we always tell them to use Modal.”

Berhandsson said his company has already made “tremendous progress” over the last year, and expects that to accelerate now it has the funding it needs to scale its business. “Our goal is to be the infrastructure provider for every single part of developing and running AI in production,” he explained. “The infrastructure demands of AI are only becoming more complex over time.”

Images: Modal Labs