Upwind Raises $400m Series C at $3.8bn Valuation

Upwind, a cloud and AI security start-up, has closed a $400m Series C round led by Bessemer Venture Partners. Co-founder and chief executive Amiram Shachar announced the round on 28 October in a post on Upwind’s website and on LinkedIn, saying it brings the company’s total funding to “over $800M”. The company told reporters the round values it at $3.8bn, SiliconANGLE and Ctech reported.

The round is bigger than first reported. In early September, reports put it at $300m at the same $3.8bn valuation, led by Bessemer and TCV, according to Ctech. Upwind has now added another $100m. It comes less than nine months after a $250m Series B, also led by Bessemer, that valued the company at $1.5bn in January.

In brief

Upwind has closed a $400m Series C led by Bessemer Venture Partners, up from $300m reported in September.
The company says it is valued at $3.8bn, up from $1.5bn in January, and that its revenue has quadrupled in 2026.
Upwind says it will spend the money on product and engineering, including AI agents that investigate and fix security problems.

Who put in the money

Upwind’s post says every existing investor took part: Greylock Partners, Cyberstarts, Craft, Salesforce Ventures, Alta Park, Picture Capital, Swish Ventures, Cerca Partners, TCV, Penny Jar and Leaders Fund. Canapi Ventures joins as a new investor. SiliconANGLE put the total raised to date at about $830m.

Bessemer has been active in large late-stage rounds this month. It also backed Legora’s $550m Series E and Bedrock Robotics’ $600m Series C.

The close follows a report by The Information, cited by Israeli business daily Globes, that Upwind turned down a takeover offer of about $4bn from monitoring company Datadog, SiliconANGLE said. Upwind’s announcement does not mention it.

What Upwind sells

Upwind was founded in 2022 by Shachar and former colleagues from Spot, a cloud cost start-up that NetApp bought in 2020. It is San Francisco-based, with Israeli roots. Its platform watches cloud systems while they run, rather than relying only on periodic scans. The aim is to tell security teams which weaknesses can actually be reached and exploited. SiliconANGLE reported that the company names Siemens, Roku and Carvana among its customers, and that AWS customers can open Upwind from the AWS console.

Shachar says the company is now building “AI Sensors for endpoints, Agents and Skills scanners”. He describes the change as a shift from “Code to Cloud” to “Endpoint to Code to Cloud”. Upwind also sells three AI agents: a Blue Agent that investigates attacks across log sources, a Red Agent that tests whether exposures can actually be exploited, and a Green Agent that fixes findings. It has also set up an AI Labs unit for cybersecurity research.

“Security needs to keep pace with environments that AI agents can change at a speed and scale that human teams alone cannot match,” Shachar said in comments carried by TNW. Upwind’s researchers are already visible in incident response: one opened the GitHub report on the compromised LiteLLM release on PyPI last week, and the team has posted analysis of the PHP2Shell flaws.

A $100m top-up in eight weeks

Adding $100m to a round between the first reports and the close, at an unchanged valuation, points to more investor demand than the original round could absorb. Upwind names Canapi as its only new investor, so most of the extra money appears to have come from backers who already know the company’s numbers. That is a vote of confidence, but it also means few newcomers have tested the price.

The revenue claim deserves care. Upwind says revenue has quadrupled in 2026 but has not published a figure. Ctech separately reported annual recurring revenue above $50m, compared with about $20m at the end of 2025. That is roughly two and a half times, not four, though the two numbers may measure different things. Until Upwind gives a figure, the valuation rests on growth that outsiders cannot check.

Runtime security is now an AI agent pitch

Upwind’s argument is that AI agents change cloud systems too quickly for people to keep up, so security tools need agents of their own. Attacks seen this month back up the first half of that. In the Gheist campaign reported by Wiz, stolen GitHub tokens were used to pull secrets out of build pipelines. In LiteLLM’s case, a poisoned package ran on Python start-up. Both moved through build and release systems, not through a static misconfiguration of the kind a periodic scan looks for.

The second half, letting agents fix production systems, is harder. Upwind’s own research, published on LinkedIn this month, found that the remediation plans of 8 of 12 AI models tested would have broken production when asked to fix deliberate misconfigurations in a test AWS account. That is the gap Upwind says its Green Agent is built to close. Buyers will want evidence that it does before letting it make changes unsupervised.