Sail Research has raised $80 million to build infrastructure specifically designed for AI agents that run continuously for hours, not seconds, processing thousands of simultaneous tasks. The startup's funding, led by Kleiner Perkins, signals a fundamental shift in how the AI industry thinks about compute infrastructure. Rather than optimizing for quick chatbot responses, the focus is now on persistent, sandboxed environments that keep autonomous software running reliably in production.

This distinction matters far beyond Sail's own product. The technical problems Sail is solving, efficient inference, persistent compute environments, and secure sandboxing, directly map onto the challenges facing decentralized AI infrastructure projects such as Akash, Render, and io.net. If Sail's claimed 12x cost-efficiency advantage holds up in real-world deployments, it sets a performance benchmark that decentralized compute networks will need to match to compete for the same workloads.

What Makes Long-Running AI Agents Different From Chatbots?

The AI agents powering the next generation of autonomous software operate fundamentally differently from the chatbots most people interact with today. Traditional chatbots respond to a single query and return an answer in seconds. Long-running agents, by contrast, need to maintain state, reason through complex workflows, and execute tasks over extended periods without human supervision.

Sail's core product combines high-efficiency open-source model serving with what the company calls "Sailboxes," persistent sandboxed environments designed to keep AI agents running continuously in the cloud. These Sailboxes support OpenAI-compatible APIs, meaning developers can build agents using familiar tools while benefiting from Sail's infrastructure optimizations. The company claims to have already processed trillions of tokens, with applications spanning cybersecurity analysis and automated code review.

Kleiner Perkins partner Aditya Naganath explained the market opportunity clearly.

"It felt obvious to both of us that you're going to need a different, specific inference platform built for these long-running agents," Naganath stated.

Aditya Naganath, Partner at Kleiner Perkins

Sail was co-founded by Neil Movva and Samir Menon. Movva's background includes roles at NVIDIA, Apple, and Together AI, plus a previous startup called Blyss, an encrypted AI platform built on advanced secure computing methodologies. This combination of infrastructure expertise and security focus directly informed Sail's product design.

How Are Crypto Exchanges Adapting to Agentic AI?

While Sail itself has no blockchain component or token, the broader crypto ecosystem is simultaneously grappling with how to integrate autonomous AI agents into trading infrastructure. Major crypto exchanges including Bybit, Coinbase, Binance, and traditional brokerages like Interactive Brokers have all shipped agent-access layers built on the Model Context Protocol (MCP), an open standard released by Anthropic in late 2024.

These implementations reveal how the industry is balancing autonomy with control. Rather than blocking AI agents from trading platforms, exchanges are building dedicated, walled infrastructure to let agents operate safely. The common approach involves isolated accounts, API-only permissions, and user-set risk caps that prevent agents from accessing main funds or executing unauthorized transactions.

Recent exchange launches demonstrate the rapid pace of adoption:

• Bybit (June 24, 2026): Launched AI subaccounts targeting developers and traders, offering API-only access with user-set leverage, allocation, and withdrawal caps completely isolated from main funds.
• Coinbase (June 11, 2026): Introduced Coinbase for Agents, allowing ChatGPT and Claude to connect to user accounts for spot and derivatives trading in natural language, with payments handled by its x402 machine-to-machine protocol.
• Binance: Expanded its AI Agent Skills Hub to 13 skills spanning futures, algorithmic execution, peer-to-peer trading, and simple earn products, enabling research-to-settlement workflows end-to-end.
• OKX: Transformed its developer platform into an autonomous-agent layer supporting 60+ blockchains and 500+ decentralized exchanges, handling 1.2 billion daily API calls and $300 million in daily volume.

According to an FM Intelligence study, at least ten retail brokers and platform vendors wired AI agents into live client accounts in the first half of 2026, with Anthropic's Claude named in nine of the ten launches.

Steps to Understand How AI Agent Infrastructure Works

• Persistent Compute: Unlike traditional APIs that handle single requests, persistent compute environments keep AI agents running continuously, maintaining context and state across multiple tasks without restarting between operations.
• Sandboxed Execution: Agents operate in isolated environments where their actions are restricted to predefined scopes, preventing unauthorized access to main accounts, funds, or sensitive data while still enabling autonomous decision-making.
• Cost Efficiency: Infrastructure optimized for long-running agents can reduce operational costs significantly, with Sail claiming 12x better efficiency than proprietary systems, a benchmark that will influence how decentralized compute networks compete.
• API Standardization: The Model Context Protocol provides a single interface that allows agents to work with multiple AI models (ChatGPT, Claude, etc.) without requiring custom integrations for each platform.

Why This Matters for Decentralized AI Infrastructure

Sail's $80 million raise represents the kind of concentrated early-stage bet that Kleiner Perkins typically reserves for companies it views as category-defining. The startup launched Sailboxes in April 2026 and is already processing trillions of tokens, demonstrating rapid product-market fit.

The implications for decentralized AI networks are significant. Decentralized compute platforms like Akash, Render, and io.net have been positioning themselves as alternatives to centralized cloud providers. However, they face a critical challenge: if Sail's cost-efficiency claims hold up in production environments, decentralized networks will need to demonstrate comparable performance benchmarks to compete for the same workloads. The market is no longer just about availability or decentralization as abstract values, it's about concrete performance metrics and cost per token processed.

The strategic argument articulated by Spotware CEO Ilia Larovitcyn captures the broader shift underway.

"The AI agent will become the primary distribution layer and the main point of interaction between traders and the market," Larovitcyn explained.

Ilia Larovitcyn, CEO at Spotware

This suggests that feature-rich trading apps and exchange software won't disappear, but their role will evolve into an execution and data layer with AI sitting on top, handling the interactions. For decentralized infrastructure providers, this means the competition isn't just about raw compute capacity, it's about building the persistent, efficient, secure environments that autonomous agents actually need to operate at scale.

The convergence of Sail's infrastructure innovation and crypto exchanges' rapid adoption of agent-access layers suggests that 2026 is the year when autonomous AI moves from experimental to operational. The winners will be those who can deliver the combination of cost efficiency, security, and persistent compute that long-running agents require.