Zendoric
← Back to the day · October 24, 2026

Daytona, a unicorn at three years old: cloud built for agents is no longer a niche, it's the new infrastructure layer

🕒 Published on Zendoric: October 24, 2026 · 00:20

✨ AI-generated · how it's made

Daytona raises $60M at a $1 billion valuation to give every AI agent its own isolated computer. With Modal valued at $15.75 billion, capital is targeting the layer where agents run, not just the models.

By El Ecosistema Startup · October 23, 2026.

Daytona Platforms has closed a US$60 million Series B that values it at US$1 billion, the company announced. Its total funding now stands at US$91 million. FirstMark led the round, as it did the US$24 million Series A in February. Investors include well-known business angels such as Patrick Collison (Stripe), Tobi Lütke (Shopify) and Aravind Srinivas (Perplexity), along with strategic funds such as Qualcomm Ventures, Samsung Next and Okta Ventures.

The company's thesis is simple. According to founder Ivan Burazin, the classic cloud was designed for long-running applications operated by people, while agents "generate thousands of short, stateful jobs that nobody planned in advance." Daytona's answer is to give each agent its own isolated computer (a "sandbox": a closed environment where the agent runs code without putting the rest of the system at risk), with Linux, Windows, macOS and GPUs. According to Pulse 2.0, these environments boot in milliseconds, keep their state when paused, fork into parallel copies and disappear once the task is done.

For now, the figures behind the valuation are those the press reports from the company itself. Tech.eu reports a roughly 25-fold increase in revenue run rate (annualized recurring revenue) between January and September. Pulse 2.0 puts monthly net revenue retention at around 135%. Crypto Briefing estimates the team at about 30 employees. We have no absolute figures for current revenue, and without them the multiple should be taken with caution: growing 25-fold from a small base is far less than it sounds.

The most significant data point concerns distribution. According to InfoWorld, OpenAI included Daytona among the sandbox providers supported by its new Agents API, alongside Blaxel, Cloudflare, DigitalOcean, E2B, Modal, Oracle, Runloop and Vercel. Customers include Meta, Cohere, Writer and n8n. According to Pulse 2.0, Writer went from an in-house platform that topped out at 20,000 sandboxes a day to provisioning 300,000 in a month after migrating.

For industry context, SiliconANGLE reported this month that Modal Labs raised US$750 million at a US$15.75 billion valuation, and that nearly a third of its annualized revenue (more than US$1 billion) comes from its sandbox service.

Our take: value is shifting from "who has the best model" to "who controls the place where the agent acts." It is the same pattern we see in the battle over agent plumbing: whoever becomes the integration standard wins. OpenAI offering nine providers at once is good news for developers, because portability reduces the risk of lock-in. But it is also a sign that this layer still has no owner, and rounds in the hundreds of millions point to a consolidation that will leave few survivors. A 30-person unicorn with no dedicated sales team, according to Crypto Briefing, also illustrates the underlying thesis: AI lets very small teams do a great deal.

There is a security angle the article barely touches on, and it matters. Isolating agents is the first practical barrier against the harm they could cause if they fail or are manipulated. Turning isolation into a commercial product is good news, as long as the containment is auditable and not just marketing. In the short term, the risk is the usual one with an infrastructure bubble: valuations that price in a market still taking shape. In the long term, this layer is part of the foundation that will allow reliable agents to take on heavy tasks, from scientific research to engineering, and bring us closer to the abundance we are aiming for.

🔗 Related on Zendoric

Sources & references