Fly.io Raises $25M Series D for Agent-First Cloud Infrastructure

Fly.io has raised a $25M Series D to push deeper into infrastructure for AI agents. Dell Technologies Capital and Intel Capital co-led the round, with Andreessen Horowitz, EQT, Geodesic, and Y Combinator participating.

The financing arrives with a strategic and leadership reset. Fly.io is focusing the company on persistent "computers for agents," Scott Johnston is joining as CEO, and founder Kurt Mackey is moving into an advisory role while remaining on the board. The broader implication is straightforward: as AI agents move from short tasks into production work, infrastructure built around disposable sandboxes may not be enough.

What Happened

The $25M Series D announcement was published on July 24, 2026. Dell Technologies Capital and Intel Capital co-led the financing, while Andreessen Horowitz, EQT, Geodesic, and Y Combinator participated. Martin Casado, a general partner at Andreessen Horowitz, and Daniel Docter, a managing director at Dell Technologies Capital, will join Fly.io's board.

Fly.io also named former Docker CEO Scott Johnston as CEO and a board member. Johnston has held leadership roles at Docker, Puppet, Loudcloud, and Netscape, giving the company an operator with direct experience scaling developer infrastructure through major platform transitions. In Fly.io's account of the transition, founder Kurt Mackey said he will remain on the board and serve as an advisor, while Fly.io's official team page continues to identify Jérôme Gravel-Niquet as CTO.

Why This Matters

Fly.io is not presenting the round as a simple expansion of its existing cloud. The company is organizing its next phase around "computers for agents," persistent, connected environments where software agents can maintain state, reach other systems, and continue working across sessions. Sprites, which began as a smaller internal effort, is moving toward the center of that strategy.

That distinction sounds technical because it is technical, but the business logic is straightforward. A temporary sandbox is useful when an agent needs to run a bounded task and disappear. A production agent that remembers context, uses tools, handles long-running work, and survives restarts needs durable storage, secure networking, and lifecycle control. Fly.io is betting those requirements will become their own infrastructure category rather than a feature buried inside someone else's cloud menu.

The Traction Behind the Round

Fly.io says it closed the strongest quarter in its history and now serves more than 37,000 customers. More than 8,000 of those customers are described as agent-native, and the company reports that revenue from its largest agent-native customers grew nearly 12x over the previous 12 months. Among Fly.io's largest customers overall, agent-native companies now account for roughly two-thirds of revenue.

Those figures are company-reported, not independently audited, but they help explain the timing. The market thesis is not being built solely on conference chatter or a slide with a heroic arrow pointing upward. Fly.io can point to customers including Firecrawl, Kilocode, and Plastic Labs using the platform for production AI applications, giving the strategic pivot a commercial foundation.

The Infrastructure Fly.io Already Built

Fly.io has been building a developer-focused public cloud since 2017. Its Fly Machines are fast-launching Firecracker microVMs that developers can manage through a REST API or command-line tools. The platform combines compute with persistent storage, private networking, global routing, and regional placement, all of which are useful building blocks for applications and especially relevant to agents that need persistent state.

The company's earlier pitch centered on giving developers the power of serious cloud infrastructure without forcing them to earn a minor degree in cloud bureaucracy. That still matters. Agent infrastructure will not win because it exposes the greatest number of configuration switches. It will win when developers can safely give software workers durable resources without turning every deployment into an archaeology project.

Market Context and Competitive Pressure

The first wave of agent tooling emphasized models, orchestration frameworks, and safe execution environments. The next layer is being shaped by what happens after the demo. Agents need identities, storage, networking, observability, permissions, recovery, and predictable economics. Traditional hyperscalers can provide each primitive, while newer platforms can package those primitives around the way agent builders actually work.

Fly.io's argument is that persistent computers should be a first-class unit rather than an assembly exercise. That puts the company in a crowded infrastructure market, but it also gives Fly.io a specific wedge. The company is not trying to out-hyperscale the hyperscalers across every workload. It is trying to make production agent compute feel native, durable, and accessible to developers already comfortable with its platform.

What the Leadership Change Signals

Bringing in Scott Johnston is part of the financing story, not a separate personnel footnote. Johnston helped lead Docker through its own developer-platform transition, and Fly.io now needs to turn a technically credible product direction into an operating category. Kurt Mackey's move to an advisory role while remaining on the board preserves founder and product continuity while giving the company a CEO recruited for scale.

The board additions reinforce that objective. Dell Technologies Capital and Intel Capital bring infrastructure and compute expertise, while Martin Casado and Daniel Docter add perspectives from developer platforms and enterprise technology. Capital can buy capacity and time, but the board and CEO changes suggest Fly.io understands that distribution, positioning, and execution will matter as much as the underlying technology.

The Bigger Industry Shift

The most important part of Fly.io's Series D is the claim embedded within it: the primary user of infrastructure may increasingly be software rather than a person. If that happens, cloud products will need to accommodate agents that run longer, maintain more state, use more tools, and operate with less direct supervision. The winning abstractions may look less like disposable functions and more like durable computers with clearly defined boundaries.

Fly.io has not proved that this category will belong to it, and the company did not disclose a valuation or detailed allocation of the proceeds. What it has shown is a coherent alignment of product, customer demand, capital, and leadership around the same thesis. The $25M Series D gives Fly.io more room to test whether computers for agents become a distinct platform market or simply the next expected layer of the cloud.