Journal
The Agentic Cloud: Why Daytona Raised $60 Million to Build Infrastructure for AI Agents
Oct 21, 2026 · Donoghteh

Daytona’s $60 million Series B and $1 billion valuation point to a new infrastructure layer: fast, isolated computers where AI agents can safely do real work.
AI agents need more than a model
The next wave of AI infrastructure may be less about making models think faster and more about giving them a safe place to act. When an agent writes code, runs a tool, browses a website, or operates business software, it needs a computer environment. That environment must be available quickly, isolated from sensitive systems, and easy to reproduce.
Daytona is betting that this need is large enough to become its own infrastructure category. On October 20, 2026, the company announced a $60 million Series B at a $1 billion valuation, led by FirstMark and joined by strategic and individual investors, according to [Fortune](https://fortune.com/2026/10/20/daytona-startup-building-the-agentic-cloud-valued-at-1-billion-in-latest-funding-round/). The round brings its reported total funding to about $91 million.
The headline is a funding milestone. The deeper signal is that the agent stack is expanding beyond models and GPUs.
From model inference to agent execution
A model can generate a plan or a line of code, but useful work happens when that output is executed. That execution creates operational questions: What software may the agent access? What data is in scope? Can it run without a person’s laptop being open? Can multiple attempts happen in parallel without sharing state?
Daytona’s product provides isolated sandboxes for running AI-generated code and agent workflows. The company’s own site describes the platform as secure, elastic infrastructure, with sandbox creation in under 90 milliseconds. Fortune reports Daytona says it can prepare a new sandbox in about 60 milliseconds by using pre-distributed machine snapshots. Those are company-reported performance figures, not independent benchmarks, but they make the product thesis clear: agent workloads need environments that feel nearly immediate.
Traditional cloud virtual machines can perform similar tasks, but an agent that waits tens of seconds or minutes for every environment loses the responsiveness expected of software. At high volume, setup time and operational overhead compound. Faster provisioning also makes it practical to run different approaches side by side, then compare their results.
Why isolation matters as much as speed
Speed alone does not make an agent safe. Agents can execute code, access tools, and make changes. Giving each task a dedicated environment can limit the damage from a mistake, reduce cross-task contamination, and constrain network or software access to what the job actually needs.
This is why the sandbox is not just a developer convenience. It can be a control boundary. Companies deploying agents will need to decide what each agent is allowed to see and do, how its work is logged, and how environments are reset. The infrastructure layer that hosts execution can become part of the security model.
That does not remove the need for identity controls, human approvals, or careful permissions. Instead, it gives teams a more concrete place to enforce and test them. For founders building agent products, execution environments, reproducibility, and least-privilege access are product requirements, not back-office details.
A new balance in the AI infrastructure stack
The current AI buildout is strongly associated with GPUs, which power model training and inference. Agentic products add a complementary demand: ordinary CPU capacity to run browsers, code, databases, and business applications. Daytona’s CEO argues that agent infrastructure will shift spending toward CPUs as agents do more than generate text. That forecast remains a thesis, but it identifies a real architectural distinction: the infrastructure where an agent reasons is not necessarily the infrastructure where it performs work.
The potential market is tied to usage, not simply the number of models. One person may run multiple agents; one agent may launch several short-lived environments; and evaluation or training systems may execute many tasks in parallel. If those patterns become common, infrastructure demand will include secure compute, orchestration, state management, and auditability.
Fortune reported that Daytona serves AI companies including Anthropic, OpenAI, Nvidia, Cursor, Cognition, Meta, and Cohere, and that customer Snorkel had provisioned more than one million sandboxes on the platform. These examples suggest that agent execution environments are already useful in both product workflows and model evaluation. They do not, by themselves, prove that one vendor will dominate the category.
What founders and investors should watch
For founders, the opportunity is not to add an “agent” label to existing cloud primitives. It is to make a task’s execution reliable: start quickly, provide only the needed tools, preserve state when useful, isolate parallel work, and return results in a form people can verify. The best infrastructure will disappear into the workflow while making failures easier to understand and contain.
For investors, the key questions are practical. How often do customers run environments, and how does that usage grow as agents move from pilots to production? What is the cost per successful task? Can a provider maintain security boundaries at scale? How hard is it for cloud incumbents to match the developer experience and latency? Revenue growth matters, but sustained workload depth and gross economics will say more about whether this becomes a durable layer.
The infrastructure behind the action
Daytona’s financing does not settle the debate over how much of the future cloud will be agent-native. It does show that investors are treating agent execution as a distinct infrastructure problem. As AI moves from answering questions to taking actions, the computer underneath the agent becomes part of the product.
The next infrastructure race may be decided not only by who can serve the smartest model, but by who can give millions of agents a fast, isolated, observable place to work.
**Sources**
- [Fortune: Daytona raises $60 million at a $1 billion valuation](https://fortune.com/2026/10/20/daytona-startup-building-the-agentic-cloud-valued-at-1-billion-in-latest-funding-round/)
- [Daytona: secure infrastructure for running AI-generated code](https://www.daytona.io/)