Sail Research closed $80 million Series A funding led by Andreessen Horowitz to build infrastructure for efficient long-running AI agents and next-generation inference workloads.
Sail Research, an infrastructure company built specifically for long-horizon AI agents, has secured $80 million in combined Seed and Series A funding at a $450 million valuation. Kleiner Perkins led the Series A, with Sequoia leading the Seed round. Additional investors include Redpoint Ventures, Theory Ventures, Vine Ventures, CRV, A*, and Abstract Ventures, alongside angel investors John Hennessy (chairman of Alphabet Inc.), Lip-Bu Tan (CEO of Intel), and Tri Dao (Chief Scientist at Together AI).
The AI landscape is shifting toward agents that operate autonomously on complex tasks over hours and days—a departure from the turn-by-turn interactions that shaped today's infrastructure. Existing systems were optimized for human interaction with a prompt, not the characteristics of AI agents, which improve with access to more compute and context. As global AI spending is projected to reach $2.5 trillion in 2026, organizations remain constrained in deploying ambitious agent workloads, hampered by both cost and the rate limits and scale ceilings of platforms never designed for extended operation.
Sail removes these constraints through two core offerings: an inference stack rebuilt for throughput and efficiency to handle agents spending billions of tokens on a single task, and Sailboxes—sandbox environments designed to run for days that charge only for active work time. Together, these components enable teams to build maximally ambitious agents while maintaining favorable economics.
"Sail exists to make intelligence abundant," said Neil Movva, Sail's co-founder and CEO. "Every decision we make, from the chip level to the API, is about giving teams the tokens, the scale, and the runtime to build agents without limits."
Sail's efficiency gains come from proprietary infrastructure optimizations, including deep customization of open-source inference engines to maximize GPU performance, intelligent workload distribution across providers for resilience, and strategic use of underutilized compute. In a recent benchmark on BrowseComp-Plus, Sail's inference achieved 90.72% accuracy at up to 10 times lower cost than competing solutions.
"Most inference infrastructure was designed to minimize latency on single requests, but agents need sustained throughput across thousands of concurrent calls over hours," said Samir Menon, co-founder and CTO. "We rebuilt the stack around that constraint, and the efficiency gains compound across every layer."
Movva previously led GPU performance optimization at NVIDIA before building infrastructure expertise at Apple and Together AI. Menon also comes from Apple, where he built systems at scale.
Sail's infrastructure currently powers workflows at Parallel Web Systems, Jack and Jill, and Detail.dev. Detail.dev, a California-based code review platform, uses Sail to power agents that analyze pull requests and codebases at scale and depth previously requiring significant human engineering effort. "Building on Sail lets us ship long-horizon agents with great economics. Trillions of tokens and counting, we're happy customers," said Dan Robinson, Detail.dev's CEO.
The company's API is compatible with existing OpenAI-based workflows and supports leading open-source models including DeepSeek, Gemma, GLM, Kimi, and Nemotron. More information is available at sailresearch.com.