Upwind Raises $400 Million at $3.8 Billion Valuation for Cloud and AI Security

Upwind Security has closed a $400 million Series C at a $3.8 billion valuation, securing fresh capital to expand its cloud and AI security platform as enterprises deploy more AI applications and autonomous agents. Bessemer Venture Partners led the round, which will fund product development and go-to-market expansion.

Canapi Ventures participated alongside existing investors Cyberstarts, Greylock Partners, Craft Ventures, Leaders Fund, Salesforce Ventures, TCV, Alta Park Capital, Penny Jar Capital, Cerca Partners, Swish Ventures, and Picture Capital.

The company says it has quadrupled revenue in 2026, although the announcement does not disclose an absolute revenue figure. Its new valuation compares with $1.5 billion at its January 2026 Series B. Founded in 2022, Upwind is building around a central premise: security teams need to understand what is happening inside running systems, alongside what scans and configuration reviews tell them could happen.

Runtime Context Becomes the Investment Thesis

Cloud security can produce long lists of vulnerable software, risky permissions, and configuration problems. The operational challenge is deciding which combinations create an exposure that an attacker can actually reach—and what to address first.

Upwind combines information about cloud workloads, APIs, and AI agents with runtime intelligence to help security, development, and DevOps teams make those decisions. Instead of treating each alert as a separate problem, the platform aims to connect the affected workload with its activity, dependencies, permissions, and surrounding infrastructure.

“Context is key to managing internal and external risks, security teams need to understand how cloud, and AI interact in real time,” said Amiram Shachar, Upwind’s CEO and co-founder, in the funding announcement.

That approach becomes more consequential when software agents can call tools, use credentials, and change infrastructure. A vulnerability inventory alone cannot describe everything an agent is doing. Live activity provides another layer of evidence for distinguishing a theoretical weakness from an active incident.

How Upwind Collects Signals From Running Workloads

Upwind’s sensor documentation explains an important part of the technology behind that context. Its runtime sensor uses eBPF to observe activity at the Linux kernel level, including process execution, system calls, network activity, and file access. The documentation describes filtering and enforcement in the kernel, while the sensor itself runs as a user-mode process.

For Kubernetes deployments, the Upwind Operator installs the sensor as a DaemonSet, providing node-level visibility into containers. A separate cluster manager collects cluster metadata and helps enrich the signals collected by the sensors. The company also describes preventative capabilities such as blocking malicious processes or stopping compromised containers.

Its cloud security platform combines runtime sensors with agentless cloud scanners. That distinction matters: runtime observation adds evidence about live behavior, while scanning helps establish the broader inventory and security posture. Upwind brings those approaches together across application security, posture management, and detection and response.

Securing AI Agents—and Using Agents for Security

There are two sides to Upwind’s AI strategy. One is protecting the models, applications, tools, and agents that customers deploy. The other is using specialized agents to help security teams investigate and address findings.

On its AI security platform page, Upwind describes discovering AI resources and unsanctioned tools, mapping the components behind AI applications, and identifying the sensitive data and permissions those systems can reach. It also lists inspection of agent instructions and Model Context Protocol connections for risky configurations and prompt-injection paths. MCP connects AI applications to external tools and data, making those connections part of the security boundary.

The company’s Agentic Pack separates security work among named agents. Blue investigates incidents and correlates runtime, cloud, and identity signals. Red validates exploitable attack paths. Green focuses on remediation, while Choppy serves as a conversational control plane that coordinates the agents and carries evidence between stages.

These are Upwind’s descriptions of its product capabilities. Autonomous investigation and remediation remain distinct from the human decisions about response priorities and acceptable operational changes. The funding announcement positions the agents as a way to reduce triage work so teams can concentrate on judgment and response.

Expanding Across Enterprise Cloud Environments

Upwind supports Amazon Web Services, Microsoft Azure, Google Cloud Platform, and Oracle Cloud Infrastructure. According to the announcement, its AWS Security Hub integration also allows customers to access Upwind within the AWS console, with onboarding and billing through AWS.

The company names Siemens, Roku, Carvana, Nubank, Supabase, and ClickUp among its customers. The $400 million round will support further platform development and commercial expansion as the company pursues enterprises managing increasingly interconnected cloud and AI systems.

The investment sits alongside a broader push to secure agents’ access to enterprise infrastructure, including Infisical’s recent funding for credential and access infrastructure. Upwind’s focus is the activity surrounding that access: what is running, what it can reach, which exposures matter, and how security teams respond when behavior changes.