AI cloud startup Modal raises $750M at $15.75B valuation

Startup Modal, a provider of artificial intelligence infrastructure, has secured $750 million in funding, led by Accel and joined by Kleiner Perkins and numerous existing investors. The Series D round has increased Modal's valuation to $15.75 billion, more than tripling its worth from the May funding round. The company reports annualized revenue of over $1 billion, marking a more than threefold growth since the May round.

A significant portion of Modal's revenue, approximately one-third, comes from Sandboxes, a cloud service offering optimized containers for AI workloads. These containers facilitate model training, AI agent development, and execution of generated code. The company's custom infrastructure management system, or scheduler, enables the provisioning of tens of thousands of containers per second, as demonstrated in a recent internal test where Modal ran 1 million containers concurrently.

Sandboxes that host AI agents may include sidecars, containers equipped with auxiliary software, such as tools that detect and filter hallucinations. The sidecars operate on the same physical server to minimize latency. Modal's secret injection feature allows AI agents to use application login credentials without exposing them, enhancing security against rogue agents and cyberattacks.

In addition to Sandboxes, Modal provides a cloud service optimized for AI training workloads, enabling developers to deploy graphics card clusters with simple scripts. Data is exchanged between the GPU clusters using RDMA, a network protocol that shortens packet travel routes. Modal competes in the inference market with Endpoints, a service that enables developers to deploy AI models on hosted infrastructure and make them accessible through an application infrastructure.

If an enterprise prefers not to develop its own model, it can utilize Modal's library of open-source algorithms. The services run on hardware rented from over 20 cloud providers, ensuring continuous operation even if one cloud experiences technical issues. A significant portion of the newly raised funding will be allocated to product development, with plans to introduce services focused on reinforcement learning and post-training phases of the AI development workflow, aimed at improving AI reasoning, safety, and cost-efficiency.