Tractian hits $1.4 billion as factories pay to predict machine failures

Tractian, an Atlanta-based startup that pairs industrial sensors with AI, has raised $130 million in a Series D round at a $1.4 billion valuation, The Information reported. Sapphire Ventures led the round, which makes Tractian a unicorn.

The company confirmed the round in its own press release on October 22, 2026. M&G’s Catalyst co-led. Also taking part were N47, General Catalyst, Bosch Ventures, 53 Stations, Monashees, Endeavor Catalyst, Norte Ventures and Y Combinator, according to the release.

Sensors, a failure dataset and an agent layer

Tractian sells what it calls physical AI: sensors fitted to critical machines, plus software that reads their condition. According to the company, its platform combines vibration, ultrasound and other sensor data into a single view of each machine. On top sits an agentic AI layer that Tractian says was trained on one of the largest industrial failure datasets. The company says that layer diagnoses root causes and recommends actions.

The release says the new money will help scale the platform and speed up a shift from predicting failures to protecting machine health more broadly. It does not break down how the $130 million will be spent. Founder and CEO Igor Marinelli framed the goal as to “close the alert-to-action gap and eliminate unplanned downtime and lost revenue.”

Tractian names Eli Lilly, McKesson, Whirlpool, Coca-Cola bottling facilities, Cummins and Embraer among its customers, along with unnamed data center operators and government infrastructure. It says more than 2,000 critical facilities worldwide use its system. The company has more than 700 employees, with headquarters in Atlanta and offices in Monterrey, São Paulo and Poland.

Anders Ranum, a partner at Sapphire Ventures, tied the investment to the buildout under way in AI infrastructure: “The AI boom is driving a massive build-out of physical infrastructure.”

Growth figures behind the valuation jump

The release does not disclose revenue. Bloomberg Línea, which covered the round the same day, reported that Tractian’s annualized revenue passed $100 million in September 2026, up from $40 million in 2024. The outlet also put the Series C valuation at $800 million, which would make the new figure a 75% increase. Another database, multiples.vc, lists that December 2024 round at $720 million, so the earlier valuation is reported inconsistently. Tractian’s own release does not state it.

Bloomberg Línea reported that headcount grew from about 400 in April 2025 to more than 700 now, and that half of revenue comes from outside Brazil, mainly the US and Mexico. It describes Tractian as the region’s second AI unicorn after legal-tech company Enter. The founders are Brazilian, and Bloomberg Línea reported that research and development is based in São Paulo.

Those figures come from the outlet’s reporting, not from audited accounts, and Tractian has not published them itself.

Why unplanned downtime is the pitch

The company’s argument rests on the cost of stopped machines. Its release cites a Siemens estimate that unplanned downtime costs the world’s 500 largest industrial companies about $1.4 trillion a year. That is a market estimate, not a Tractian result, and the release does not say how much of it the company’s customers have avoided.

For plant operators, the practical change is that condition monitoring is moving from vibration-only sensing, where Tractian started in 2019 according to Bloomberg Línea, toward multiple sensor types feeding a system that proposes a fix. For data center operators, a named customer category, the appeal is similar: more machines to watch as capacity grows.

Bloomberg Línea also reported that Tractian plans to use the funds for new sensors, product development and training AI models on a fleet of Nvidia GPUs, and to keep growing in the US while accelerating in Europe. The company’s release does not confirm those plans.

What remains unverified

Several claims rest on the company alone: the failure-dataset claim, the customer list and the 2,000-facility count. The release also gives a customer figure of more than 2,000 in its boilerplate, so the count may be measured two ways. No independent figures on how accurate Tractian’s failure predictions are have been published in the sources reviewed. The round is also not the same as availability: Tractian already sells its system, and nothing in the announcement introduces a new product, price or plan.

The Information’s full article is behind a paywall. Its public summary describes Tractian as a seven-year-old company that provides sensors to assess machine condition and predict how long until they fail.