Sail Research, a startup backed by Sequoia Capital, has raised $80 million to build infrastructure purpose-built for AI agents that operate autonomously over extended periods, not the brief back-and-forth interactions that today's AI systems were designed for. The funding round, which valued the company at $450 million, reflects a growing recognition among venture investors that the next wave of AI productivity will require fundamentally different technical infrastructure.

Why Are Long-Horizon AI Agents Different from Today's AI?

Most AI infrastructure today was optimized for a specific use case: a person types a prompt, the AI responds quickly, and the interaction ends. But AI agents are different. They work autonomously on complex tasks over hours or days, making decisions, gathering information, and refining their work without human intervention at each step. This shift requires rethinking how AI systems consume computing resources and how much they cost to run.

Sail Research's platform addresses this gap with two core components. First, the company rebuilt its inference stack, which is the computational engine that powers AI models, from the ground up to prioritize throughput and efficiency rather than speed on individual requests. Second, it created Sailboxes, a sandbox environment where AI agents can run for hours or days while only being charged for the time they're actively working, not idle time.

The efficiency gains are substantial. In a recent benchmark called BrowseComp-Plus, which tests deep research capabilities, Sail's infrastructure achieved 90.72% accuracy while delivering up to 10 times lower cost per token than leading alternatives. To put this in perspective, tokens are the basic units of text that AI models process; lower cost per token means organizations can run more ambitious AI agents without breaking their budgets.

Who Is Behind Sail Research and Why Does That Matter?

The founding team brings rare expertise in building systems at massive scale. CEO Neil Movva previously worked at Nvidia, where he pushed GPU (graphics processing unit) performance to its limits, and later built infrastructure at Apple and Together AI. CTO Samir Menon also comes from Apple, where he built systems designed to handle enormous computational loads.

The investor backing reflects confidence in both the founders and the market opportunity. Beyond Sequoia's seed investment and Kleiner Perkins' series A leadership, the round included participation from several other prominent venture firms and angel investors with deep tech credentials. Those angels include John Hennessy, chairman of Alphabet Inc.; Lip-Bu Tan, CEO of Intel; and Tri Dao, Chief Scientist at Together AI.

"Sail exists to make intelligence abundant. 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," said Neil Movva, co-founder and CEO of Sail Research.

Neil Movva, Co-founder and CEO of Sail Research

How to Evaluate AI Infrastructure for Long-Horizon Agent Workloads

• Cost Efficiency: Look for platforms that charge based on actual compute time rather than flat rates, and measure cost per token across realistic workloads to understand true economics.
• Throughput Optimization: Evaluate whether the infrastructure is designed for sustained performance across thousands of concurrent requests over hours, not just single-request latency.
• Compatibility and Flexibility: Ensure the platform supports multiple open-source models and is compatible with existing workflows, reducing switching costs and vendor lock-in.
• Benchmark Performance: Review independent benchmarks that measure both accuracy and cost on tasks relevant to your use case, not just raw speed metrics.

Sail's infrastructure is already powering production workloads at several companies. Parallel Web Systems, which builds background agents for web research, uses Sail to handle the inference layer of its platform. Detail.dev, a code review platform based in California, uses Sail to power agents that analyze pull requests and codebases at a depth and scale that would previously have required 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, CEO of Detail.dev.

Dan Robinson, CEO of Detail.dev

The timing of Sail's funding reflects broader market dynamics. Global AI spending is projected to reach $2.5 trillion in 2026, yet most organizations remain constrained by the cost and rate limits of platforms never designed for long-horizon agent workloads. Sequoia's investment in Sail, alongside its backing of other AI infrastructure plays, signals that venture capital is increasingly focused on the foundational layers that will enable the next generation of AI applications.

Sail's API is compatible with existing OpenAI-based workflows and supports leading open-source models including DeepSeek, Gemma, GLM, Kimi, and Nemotron, making it accessible to teams already invested in those ecosystems. This compatibility reduces friction for adoption and allows organizations to experiment with long-horizon agents without completely rebuilding their AI stacks.

"The infrastructure layer for the agent era is one of the most important bets in AI right now, and Neil and Samir are exactly the founders to build it. They bring a rare combination of deep compute expertise and systems rigor that only comes from having built at the limits of scale," said Aditya Naganath, partner at Kleiner Perkins.

Aditya Naganath, Partner at Kleiner Perkins

The $80 million funding round underscores a shift in how venture investors are thinking about AI infrastructure. Rather than betting on faster inference or larger models, firms like Sequoia and Kleiner Perkins are backing companies that solve the economic and operational constraints that prevent organizations from building truly ambitious AI agents. As AI agents become more central to business operations, the infrastructure that powers them will likely become as important as the models themselves.