Daytona Raises $60M at a $1B Valuation to Sell Every Agent Its Own Computer

TL;DR

Daytona raised $60 million at a $1 billion valuation, led by FirstMark Growth; it reports $91 million raised in total. (Source: Fortune — funding; TechCrunch — deal)
Its headline operating metric is roughly 135% net revenue retention per month—spectacular if broad and repeatable, but potentially inflated by a small, scaling customer base. (Source: Daytona — Series B announcement)
Sandboxes already serve training and evaluation as well as production agents. That mix makes workload durability the central investment question. (Source: Fortune — funding; Daytona — Series B announcement)

Why agents need a cloud of their own

A model can generate code or a plan; an agent that acts needs somewhere to run programs, browse, handle files, and use software. Daytona sells isolated, persistent sandboxes—computers agents can operate, rather than another model API. As products move from chat demos into tool-using workflows, the execution environment becomes its own infrastructure layer, with costs and operational demands distinct from inference. (Source: Fortune — funding; Daytona — Series B announcement)

Founded in 2023 by Ivan Burazin, Goran Draganić, and Vedran Jukić, Daytona began by automating enterprise development environments. Burazin says the team handed customers to competitors, rebuilt the product, and relaunched with about three months of revenue left. The reported headquarters conflict: Fortune says San Francisco, TechCrunch New York. (Source: Fortune — funding; TechCrunch — deal)

A $1 billion bet on a new infrastructure layer

FirstMark Growth led the $60 million Series B; roughly $50 million came from existing investors, Burazin told Fortune. Backers include angels Patrick Collison, Tobi Lütke, Aravind Srinivas, Eric Yuan, Olivier Pomel, Adam D’Angelo, and Brendan Foody, plus Qualcomm Ventures, Samsung Next, and Okta Ventures. Pace Capital, E2VC, and Upfront Ventures are existing investors. Daytona says its $91 million total includes a $24 million Series A in February 2026, also led by FirstMark. The valuation prices expected category growth; no revenue multiple has been disclosed. (Source: Fortune — funding; TechCrunch — deal; Daytona — Series B announcement)

Why milliseconds matter

Daytona says its VMs provision in about 60 milliseconds, versus tens of seconds to minutes for conventional cloud VMs. It achieves this by distributing configured machine snapshots across server disks, then attaching CPU and memory when a sandbox is requested. Agents can clone a running machine to try alternatives in parallel. That fast, repeatable startup can turn a single task into many experiments without waiting for each environment to boot. (Source: Fortune — funding)

The company says it supports Linux, Windows, macOS, and GPU sandboxes on one platform. Its contrast with GPU-heavy “neoclouds” such as CoreWeave and Nebius: inference clouds are where an agent thinks; Daytona is where it works. Agent execution needs CPU and varied software environments, not only accelerators. In a market also served by E2B and Modal, Daytona’s differentiation is therefore provision speed and OS breadth, not an assumed price advantage. (Source: Daytona — Series B announcement; Fortune — funding; Upfront Ventures — Daytona)

Retention is the metric to interrogate

Daytona says annualized revenue run-rate rose about 25-fold from January to September 2026, with average monthly growth of 50.9%. More unusual is average monthly net revenue retention (NRR) of about 135%. If that rate compounded for twelve months, 1.35^12 is approximately 36.6: the same cohort would spend over 36 times as much. Daytona contrasts its figure with Snowflake’s roughly 160% annual NRR at IPO and says customers from six months earlier now spend nearly three times as much in aggregate. Monthly and annual NRR, however, are not directly comparable. (Source: Daytona — Series B announcement; Fortune — funding)

The key question is whether expansion reflects more production workflows per customer, or a few customers scaling temporary training and evaluation workloads. Without cohort sizes, customer-level distribution, gross retention, and revenue mix, 135% monthly NRR is a striking signal—not a durable growth law. (Source: Daytona — Series B announcement; Fortune — funding)

Usage is proven; production durability is not

Fortune names Anthropic, OpenAI, Nvidia, Cursor, Cognition, Meta, and Cohere as customers; Daytona also lists Writer, Clay, n8n, Postman, Mintlify, and bolt.new. Snorkel has provisioned over one million sandboxes, the company says. Writer, after running its own platform beyond 20,000 sandboxes a day, moved to Daytona and provisioned 300,000 in one month. AfterQuery says it replaced an internal GKE cluster in a day. These examples show operational pain and real usage, but not the proportion that is steady-state production. (Source: Fortune — funding; Daytona — Series B announcement)

The critical nuance is that many sandboxes support reinforcement learning, model tests, and evaluations—not just agents serving end users. Training can generate intense bursts of compute demand, then contract as research plans or budgets change. Production agents could create more repeatable consumption, but that market is still emerging. Upfront Ventures argues sandboxes and inference clouds are distinct markets and that compute demand is hard to overstate; that does not eliminate cyclical exposure if early revenue is experiment-heavy. (Source: Upfront Ventures — Daytona; Fortune — funding)

What the raise does not prove

The round validates investor conviction and observed demand, not that Daytona can preserve its speed advantage at scale, that the agent cloud becomes a durable category, or that today’s workloads recur. The strongest evidence to watch is production revenue and retention across a broad customer cohort after training experiments end. If sandboxes become agents’ default execution substrate, the opportunity is substantial; if the largest workloads remain episodic model development, growth could track volatile compute cycles. (Source: Fortune — funding; Daytona — Series B announcement; Upfront Ventures — Daytona)

FAQ

What does Daytona sell?

Persistent, isolated computers where agents can run code and use software. Daytona says these sandboxes start in about 60 milliseconds. (Source: Fortune — funding)

Does 135% monthly NRR mean 36x annual growth?

Only if it compounds unchanged for twelve months. That is an illustration, not a forecast; cohort composition and workload mix matter. (Source: Daytona — Series B announcement)

Why not run agents on a hyperscaler?

Daytona’s case is that fast provisioning, persistent state, and parallel clones suit agent workloads better than general-purpose VMs. Whether those benefits warrant a separate platform depends on workload and operational needs. (Source: Fortune — funding; Daytona — Series B announcement)