Technology / New York — The Daytona agent cloud $1 billion valuation rests on a simple bet: AI agents need computers of their own, built very differently from the cloud built for people.
Track it live: Live markets tracker
The Daytona agent cloud $1 billion valuation, confirmed this week after a $60 million Series B, would have sounded improbable three years ago for a company selling something as unglamorous as rented computers. Yet that is precisely the point. Daytona Platforms is not selling a chatbot, a model, or an application that end users ever see. It is selling the machines on which autonomous AI agents live while they work — writing code, browsing, testing software, and running tasks that can stretch for hours. TechCrunch reported on October 23 that the round was led by FirstMark Growth, and company announcements on October 20 put the valuation at $1 billion, making Daytona one of the newest unicorns in the fast-forming agent infrastructure sector.
The structure of the round tells its own story. FirstMark Growth also led Daytona's $24 million Series A in February, according to TechCrunch on October 23, and roughly $50 million of the new $60 million came from existing investors, the company said in its October 20 announcements. Total funding now stands at $91 million. When insiders provide most of a round at a sharply higher valuation, it can signal deep conviction from the people who know the numbers best. It can also signal concentration, a point sceptics return to below. Both readings are available from the same facts, and the difference will only be settled by revenue that Daytona, for now, is not disclosing in absolute terms.
A round built by insiders and industry names
Beyond FirstMark, the investor list reads as a map of the software industry Daytona hopes to supply. Company announcements on October 20 listed angel investors including Stripe's Patrick Collison, Shopify's Tobi Lütke, Perplexity's Aravind Srinivas, Zoom's Eric Yuan, Datadog's Olivier Pomel, Quora co-founder and OpenAI board member Adam D'Angelo, and Mercor's Brendan Foody. Strategic investors Qualcomm Ventures, Samsung Next, and Okta Ventures also joined, TechCrunch reported on October 23. Each name carries a slightly different meaning. Collison and Lütke run platforms where agents are expected to transact. Yuan and Pomel run infrastructure and communications businesses that agents will use. Srinivas builds an agent-adjacent product himself. Their presence is marketing, but it is also diligence by proximity: these are operators who can judge whether agent workloads are real.
Daytona is three years old and based in New York, founded by Ivan Burazin, Goran Draganić, and Vedran Jukić, according to company announcements on October 20. That age matters for context. A $1 billion valuation at three years old, on $91 million raised, implies investors are pricing not current profit but a position in a market that barely existed when the company started. Comparable patterns appeared in earlier infrastructure waves: developer tools and cloud observability firms also reached unicorn valuations ahead of disclosed revenue, on the argument that whoever supplies the default layer during a platform shift collects rent for a decade. Some of those bets paid off handsomely. Others were repriced sharply when growth slowed.
What an agent cloud actually does
Burazin's explanation, quoted in coverage of the October 20 announcement, is the clearest statement of the thesis: "Agents don't work like the long-lived services today's cloud was designed for. They spin up thousands of short, stateful jobs nobody planned in advance." A human-facing web service runs continuously, with predictable traffic. An agent, by contrast, may need a complete computer for ninety seconds, another for six hours, and then need both paused, copied, or rewound to an earlier state when an experiment fails. Daytona's Agent Cloud offers sandboxed computers running Linux, Windows, or macOS, with graphics processors available, that can be provisioned in milliseconds — around 60 milliseconds is the figure cited in coverage — and can pause, resume, and fork like documents rather than servers.
Pricing gives a sense of the commodity underneath the valuation. Analysis coverage of the announcement put list pricing at about $0.165 per hour for a sandbox with two virtual CPUs and four gibibytes of memory. At that price, a single agent working a full day costs only a few dollars in compute, which is cheap against a human salary and expensive only at the scale agents promise: thousands of parallel workers, running continuously, across every customer at once. The business therefore depends on volume and retention rather than margin per machine. It is a familiar cloud equation, but with a twist — the customer is software, and software does not sleep, does not negotiate, and can switch providers by changing a configuration line, which makes both loyalty and churn faster than in human markets.
Why this matters
Why this matters is that the agent economy is acquiring a physical layer, and the contest for that layer is happening now, before most people have used an agent at all. Every prior computing shift created a landlord. The web created hosting and then hyperscale cloud providers. Mobile created app stores that tax distribution. If agents become a primary way software is written and work is done, the company that rents agents their computers sits underneath every transaction, largely invisible and difficult to displace once workflows, snapshots, and tooling accumulate around it. That is the prize a $1 billion valuation is really pricing.
The reported growth figures, attributed to the company, explain why investors are willing to pay early. Daytona said its revenue run rate grew roughly 25 times between January and September 2026, and that average monthly net revenue retention was around 135 percent, both reported in coverage of the October 20 announcements as company figures. Retention above 100 percent means existing customers spend more over time without any new logos being added, a pattern investors prize because it compounds. Twenty-five-fold growth in nine months, however, starts from an undisclosed base. Twenty-five times a small number can still be a small number, and Daytona has not published absolute revenue, so the valuation rests on the durability of that growth rather than on a proven scale of sales today. That distinction is the honest centre of the story.
Context sharpens the stakes. Demand for agent compute is rising alongside demand for raw AI compute itself, a trend examined in China's push to double AI compute capacity, and alongside a parallel tightening of how frontier models are developed, as in OpenAI's decision to cut live internet access from frontier training. Sandboxed agent computers sit between those two stories: agents need isolation for safety, and they need reliable machines for usefulness. Daytona's pitch is that it provides both in one product, which is why reported customers and infrastructure users in coverage include AI labs and enterprises such as Anthropic, OpenAI, Nvidia, Cursor, and Cognition. If even a fraction of those relationships deepen, the retention figures become easier to believe.
The sceptics' case
Sceptics make three arguments, and none is trivial. First, the hyperscalers are not standing still. Amazon, Microsoft, and Google can add agent sandboxes to clouds their customers already use, bundle them cheaply, and absorb years of low margins. Second, sandboxing is a feature that may become standard rather than a company; open-source tools and rival startups are chasing the same milliseconds. Third, the round's insider weight — about $50 million of $60 million from existing investors, per the October 20 announcements — means the $1 billion price was set largely by people already holding the stock, not tested by a broad field of new buyers. None of that makes the valuation wrong. It makes it a forecast, and forecasts about agent demand have already swung wildly in both directions during 2026.
Supporters answer that focus is itself the moat. A general cloud optimises for its largest workloads, while Daytona optimises only for agent behaviour: instant starts, cheap pausing, forking, and state that persists between bursts. They point to the strategic investors — Qualcomm Ventures, Samsung Next, and Okta Ventures, named in the October 20 announcements — as evidence that device, chip, and identity companies also expect agents to need dedicated machines, and may help distribute the product. On that view, being early and specialised when a workload is strange is exactly how infrastructure categories are won.
What happens next
What happens next will be measured in disclosure as much as in product. Watch whether Daytona publishes absolute revenue or customer counts at its next raise, whether the reported 135 percent retention holds as early cohorts mature, and whether hyperscalers ship competing agent sandboxes with aggressive pricing. Watch also the founders' hiring and capacity choices after October 20: a company that has raised $91 million in total can build ahead of demand, but building ahead is only wise if the 25-fold growth was the start of a curve rather than a spike. For now, the fairest reading is balanced. A three-year-old New York company has persuaded some of software's best-known operators that agents need their own cloud, and has been valued at $1 billion for that belief. The machines are real, the pricing is public, and the workload is growing. Whether the rent holds is the billion-dollar question the valuation itself poses.