A stamping press on a plant floor in Ohio is shouting in a language nobody on the management floor speaks. It produces a part every 1.4 seconds and a data point every fraction of that. Most of those data points die where they were born. Sight Machine, very politely, refuses to let them.
The quiet operating system of the modern factory
Walk into a Global 500 plant in 2026 and look up. The cameras, sensors and PLCs that run the line are talking to something. More and more often, that something is Sight Machine. The company calls itself "the industrial AI company," which is the kind of label tech companies hand themselves without much thought. In Sight Machine's case, it actually earned the title before the words got fashionable.
Today Sight Machine sells what it calls a Manufacturing Data Platform. The unsexy name hides a fairly radical idea: that every machine, line and plant inside an enterprise should be modeled as a single, live, AI-ready dataset. Not a dashboard. Not a report. A digital twin of production itself, kept honest by real sensor data and ready for any AI agent the operator wants to point at it.
"We bring the complete industrial AI stack to the factory in weeks, not quarters."
- The company's own pitch, and one of the few they can defend with reference customers.Manufacturing's dirty secret
Manufacturing generates more data than almost any other industry. It uses less of it than almost any other industry. That is not a slogan; it is the operating reality of most plants on Earth. Data sits in PLCs, in MES systems, in spreadsheets, in operators' heads, and in the silent histories of machines that nobody bothers to read.
For two decades the industry's answer was a parade of acronyms - MES, SCADA, OEE, IIoT - each promising visibility, each delivering another silo. By the time the cloud era arrived, most factories had digital exhaust pouring out of every port and almost no way to turn it into a decision before the shift ended.
"Most plant data dies on the floor. The factory of 2030 is the one that catches it."
- Paraphrased, but only just.Sight Machine's bet was that the bottleneck was not sensors or storage. It was structure. Without a shared model of what a "line" or a "good part" or a "downtime event" actually means across machines, no algorithm - clever or otherwise - can earn its keep. So they decided to build that model first, then let the AI come.
From Slashdot to stamping presses
The founding team is, on paper, unlikely. Jon Sobel, the CEO, was an early Yahoo and Tesla executive. Nate Oostendorp, the CTO, co-founded Slashdot - that ur-text of internet nerd culture from the late nineties. Kurt DeMaagd, now Chief AI Officer, came out of academia and applied analytics. Anthony Oliver rounded out the original four.
They started the company in Michigan in 2011 - which is to say, they started it in Detroit's shadow, which is to say, on purpose. By 2012 they had added a San Francisco headquarters. The pitch, repeated often enough that investors started repeating it back, was that industrial software needed both: Silicon Valley's tolerance for software experimentation, and Detroit's intolerance for software that does not actually work.
One model. Every machine. Every line.
The Sight Machine platform does four things in roughly the order a factory needs them done. It connects to anything that emits data, from a 1980s PLC to a brand-new edge gateway. It structures that data against standardized schemas, which is the unglamorous step that makes everything afterwards possible. It analyzes the result, surfacing the patterns that operators care about - quality defects, hidden downtime, throughput drift. And it operationalizes, which now means handing the model to AI agents that can suggest, and sometimes execute, the next move.
Factory Connect
Enterprise data integration across MES, SCADA, historians, and edge devices.
Factory Analyze
Out-of-the-box analytics, custom dashboards, KPI monitoring across plants.
Factory Build
Plant Digital Twin authoring; standardized data modeling with Blueprint automation.
Factory Operate
Industrial AI agents for root-cause analysis, optimization, and 3D visualization on Omniverse.
"Connect. Structure. Analyze. Operate. In that order. The order matters."
- A summary of every Sight Machine sales call, condensed for the sake of patience.The customers that don't tweet about it
Sight Machine's customer logos are, by the standards of B2B SaaS, magnificently boring. They are not the kind of brands that get covered in the morning newsletters. They are, however, the brands that produce the cars, packaged goods, chemicals, food and clothing inside almost every house on the continent. Global 500 manufacturers, Fortune 50 references in automotive and apparel, and an investor roster that reads less like a venture syndicate and more like a hall pass into the industrial world: GE Ventures, Mercury Fund, Sorenson Capital, TeamViewer, E.ON, and most recently NVIDIA's NVentures.
Sight Machine by the numbers
Capital raised relative to revenue is the giveaway: this is a patient-money business, the kind that only pays off when the plants actually run differently.
NVIDIA on top, Microsoft underneath, factory in the middle
In 2025 Sight Machine pulled off the kind of integration story that competitors quietly resent. Its platform now sits between Microsoft Fabric's Real-Time Intelligence on the data side and NVIDIA Omniverse on the visualization side. Translated: structured factory data flows up to Microsoft's lakehouse, gets modeled by Sight Machine, and pops out the other end as a physically accurate, GPU-accelerated 3D twin running on NVIDIA's industrial stack.
"The Factory Operate solution is built on three companies' work, but the model in the middle is ours."
- The whole strategy, in one sentence.The win is not the demo. The win is that the demo runs on the same data the line ran on at 7:14am that morning.
Making manufacturing stronger, sustainable, resilient
Sight Machine's mission statement is the kind that sounds like a press release until you remember what manufacturing is actually for. Cars. Vaccines. Food. The materials of ordinary life. A factory that runs better is a supply chain that breaks less often, an emissions curve that flattens earlier, and a job that is safer than it was last quarter. The company's pitch is that all of that is, at root, a data problem with a model-shaped hole in the middle of it.
It also happens to be a pitch that fits the era. Industrial policy is back. Reshoring is back. Geopolitics has made the resilience of physical production something nobody in a boardroom can ignore. The companies that turn that pressure into faster, cleaner, smarter plants will need exactly the kind of plumbing Sight Machine has spent fifteen years installing.
The factory floor learns to read
The interesting question is not whether Sight Machine wins a single category. It is whether the model it built - one canonical, AI-ready twin per enterprise - becomes the default for industrial AI, the way Kubernetes became the default for cloud. There are signs that is happening. The NVIDIA and Microsoft stack-ups suggest that the platform layer is being treated, by both giants, as a peer. Customers continue to show up, slowly, the way industrial customers always do, with five-year horizons and very little tolerance for vapor.
Back to that stamping press in Ohio. It still produces a part every 1.4 seconds. It still produces a data point every fraction of that. The difference, in 2026, is that the data points no longer die where they were born. They move, structured and labeled, into a model that knows what a good part looks like, what a bad shift looks like, and what an unhappy bearing sounds like at 2:47am on a Tuesday. Somewhere upstairs, an AI agent reads that model the way you read this sentence. And the factory, almost imperceptibly, gets better.