Baseten Acquires Blaxel to Build the Infrastructure for AI Agents in Production

The acquisition expands Baseten’s ability to deliver perpetual sandbox execution in the same production platform customers already use for inference, training, and post-training

SAN FRANCISCO--(BUSINESS WIRE)--Baseten, the AI inference company powering the world's most advanced AI applications, today announced that it has acquired Blaxel, an infrastructure company building secure, persistent execution environments, or sandboxes, for AI agents. With the acquisition, Blaxel joins Baseten, bringing deep expertise and technology for agentic tools and infrastructure. Together with Blaxel, Baseten further expands its platform toward delivering the fastest, most reliable, secure and cost-effective infrastructure agents need to run code, use tools and maintain state.

"AI applications have moved from models that generate answers to agents that actually do things," said Amir Haghighat, co-founder and CTO of Baseten. "Agents need models, but they also need a secure environment to run in that's fast, scalable, and reliable enough for production. Adding Blaxel to Baseten will accelerate our vision to provide the infrastructure and tooling for agents to reliably operate, with great performance, at scale."

AI applications are increasingly taking actions instead of just generating responses, creating new infrastructure requirements. For security and efficiency, these agentic workloads require isolated execution environments that can persist across long-running tasks without creating unnecessary overhead.

Blaxel has built infrastructure specifically for this emerging class of applications, including secure microVM-based sandboxes that spin up and resume up to 5x faster than competitive sandbox solutions. The platform natively supports agent code execution, tool use, MCP servers, and other workloads that require secure, stateful compute.

By bringing Blaxel’s execution infrastructure together with Baseten’s inference platform, Baseten is building toward a more complete runtime for production AI applications. Developers will be able to run models and the environments where those models act on infrastructure designed to work together, cutting down the latency and complexity that comes from moving state and data between separate systems.

Paul Sinaï, co-founder and CEO of Blaxel, said: “We started Blaxel because agents need environments designed to work alongside the models powering them. Bringing Blaxel together with Baseten means we can build for where we think the infrastructure is headed for agents in production: a model and execution environment in a single system.”

In the near term, Blaxel’s technology expands Baseten’s foundation for secure code execution, tool use and session persistence. Over time, these capabilities will be integrated more deeply with Baseten’s inference, storage, observability and enterprise infrastructure to strengthen its support for agentic workflows.

The acquisition follows a period of significant expansion for Baseten, including a Series F fundraise with a $13 billion valuation and the acquisition of Parsed, the reinforcement learning startup specialized in post-training and continual learning.

About Baseten

Baseten is the inference company behind a new generation of AI products. The company builds systems software that runs the entire workload for AI applications—from GPUs and autoscaling to observability, billing, and developer tools—so teams can focus on models and user experience instead of infrastructure. Baseten’s customers include leading AI companies such as Abridge, Clay, Cursor, Lovable, Mercor, and OpenEvidence and others building specialized models for their domains. Founded in 2019 and based in San Francisco, Baseten has raised over $2 billion to date from investors including Altimeter Capital, Battery Ventures, Blackbird, BOND, CapitalG, Conviction, Durable Capital Partners, Greylock, IVP, NVIDIA, Sands Capital, Spark Capital, Verified Capital, Wellington Management, 01A and others.