On June 30th, Schneider Electric, a global leader in energy management and automation, announced that it has entered into a definitive agreement to acquire Cognite. Cognite provides an Agentic AI Data Platform for the factory floor. This is a cash transaction valued at $3.1 billion.
The company, founded in 2017 and headquartered in Tempe, Arizona, reported revenue exceeding $170 million in 2025. That means the acquisition cost is 18.2 times revenues!
What could explain such a high price? Cognite is growing fast; Its annual recurring revenue bookings are growing 36%. But it is Cognite’s technology driving this growth. And Schneider Electric’s user base can help further accelerate growth.
Celanese Implements the Cognite Platform
I saw a presentation last year by Celanese, a global chemical manufacturer, about their use of Cognite. This is still the single best example I have come across of a company employing agentic AI and copilots at scale.
During COVID-19, Celanese began to consider the need for a digital transformation. When they began considering a platform to detect and respond to equipment anomalies, they realized those capabilities would support safety, improve product quality, and production optimization. The ROI associated with that could be massive.
If an asset issue was detected, resolving it could involve multiple applications used by multiple people, who see different information, enter different data, bounce emails and texts back and forth, and move information from one place to another. One event could create massive churn.
A person at a chemical manufacturing facility is involved in all kinds of processes. They prepare equipment for maintenance, perform isolation (disconnecting a piece of equipment from the chemical flow by closing valves), review quality or reliability metrics, and conduct rounds.
The director of digital manufacturing said, “We needed to model the data in a way that we can do simple searching. Can I have an industrial Google at a manufacturing facility? Why is it so hard for our people to find information in the right context? We spent hours and hours looking for data, whether it was for audits, compliance, or just basic troubleshooting. This became an investment priority.”
Celanese recognized a need to decouple its data from myriad factory applications. Data should be created once and move seamlessly in real time to where needed. This architecture was necessary to create a unified experience for users.
Celanese chose the Data Fusion platform from Cognite as its industrial data platform. But getting the context right is a difficult problem. Contextualization is the process of identifying and representing relationships between data to mirror the relationships that exist between data elements in the physical world.
Once the data was cleaned and contextualized, and appropriate data governance was implemented, they built a user interface using Generative AI. At Celanese, this GenAI UI is the sole point of interaction for workers. GenAI not only surfaces contextualized data, but it also creates new workflows where needed.
In one demo, a user asked to see a piping and instrumentation diagram. The worker then wanted to see the work orders for the vessel in question and the vessel’s tag. The UI provided these. The manager then assigned Fred the job of diagnosing the problem. The manager used the AI-generated applet to create a work order for Fred. Fred then uses the interface to diagnose the issue. The AI then asks, “Do you see this? Is this happening?” The generative AI even looked at a picture of the asset taken on Fred’s phone. Then, based on the picture and the answers, the AI suggested that corrosion might be the problem. Fred agrees. Then a new updated work order was created to swap out the asset.
Celanese did not discuss the ROI in detail, except to say it was significant and would continue to grow. Pre-trained AI agents were created. They focused on specific use cases for specific user personas. Use cases included: optimized operator rounds, data-driven checklist management, balanced workloads, and streamlined maintenance. Their ability to do effective preventative maintenance, for example, increased by 15%.
I talked to Girish Rishi, the CEO of Cognite, after the acquisition was announced. Cognite’s customer base includes large chemical and energy companies with plants that cost over a billion dollars. He said that every extra day a plant stays open, generates an incremental $11 million. With Cognite, an extra four days is very doable.
Celanese did say, however, that the UI did sometimes hallucinate, and they would work on solving that problem in the coming year. I mentioned this to Rishi, the CEO of Cognite. He said the talk on hallucinations has gone down; it is not the problem it was a few years ago. “That said, we operate in a mission-critical environment.” The answers have to be right. “The only way to reduce and eliminate hallucinations is with contextualized data. If the contextualized data is tested, and it is accurate, then the AI outcomes” don’t hallucinate. Further, they operate with a human in the decision-making loop, so there is always a check.
Knowledge Graphs Create Contextualization
Cognite’s knowledge graph creates this context. A knowledge graph creates relationships across previously siloed data sources. Knowledge graphs weave together a unified, seamless layer for data management and, by doing this, often uncover hidden patterns and relationships, patterns no human could detect. Answering a question like “Why has this piece of equipment gone down?” can require accessing many siloed data sets and then looking for relationships between the data sets and the event that occurred. Knowledge graphs can find relationships that no human could uncover.
Mr. Rishi pointed out that “for artificial intelligence to work and scale, the data has to be contextualized. The data has to be meaningful, concise, and comprehensive. Without that foundation, there is no AI.”
Mr. Rishi gave an analogy to explain how knowledge graphs work. “I’ve just been asked on July 8 to travel to San Francisco. So, Girish will travel on July 8 from Salt Lake City to San Francisco on the Delta flight at 3:30 in the afternoon. He’ll be in an economy seat at 1111 C. He will land at 4:40.” At that time, the traffic from the airport to San Francisco means the taxi will take about an hour to get to the destination. “You should have a meeting no earlier than 6:30. “That is context.” Compare that to the much less detailed data set, ‘Girish will travel to San Francisco the week of July 5th.’ That is factually true, but far less useful.
Mr. Rishi pointed out that contextualizing plant data is far more complex than contextualizing enterprise data. There is electromechanical equipment that has been in operation for decades. There are operations, engineering, supply chain, and enterprise systems. There are Word documents, images, graph data, voice prompts, and videos.
Incredibly detailed time-series data is generated as sensors continuously track equipment metrics such as temperature, pressure, and vibration. This data is collected and analyzed to prevent equipment failures, optimize manufacturing processes, improve product quality, and forecast energy demand.
Time-series data sets go a long way toward explaining why industrial data sets are so much larger than enterprise data sets. When Celanese implemented the Cognite platform, they had 2.5 trillion records from 47 data sources!
One of a knowledge graph’s superpowers is finding the root cause of a problem. “I grew up with six sigma, doing statistical process control, and diving down to zero defects,” Rishi said. “That was my Motorola pedigree. The way you did root cause analysis is you had a green or a black belt project. These took anywhere from 30 days to six months. ‘Why did the factory shut down for two days?” Now, the root cause is done on the Atlas AI knowledge graph. Root cause analysis is a favorite capability across their customer base. At their user conference last October, an energy company CEO stood up and said, ‘What took six weeks now takes six hours.’
GenAI Is Used to Create Workflows and Mini Apps
A manufacturer might have a third-tier supplier in China that provided a defective part. Cognite traverses the knowledge graph to determine whether the company has previously experienced this incident with this supplier. How many customers are out there with finished products that use this part? Atlas AI provides the results.
It then gives the user the option to use GenAI to create a workflow that sends a recall notice to all relevant customers. Customers can be provided with relevant context and receive ongoing status updates while the company works on the problem.
In using large language model AI to create workflows, users can select Anthropic, OpenAI, or Gemini. And there is a fourth option. Large language models don’t perform well on time-series data. “We have worked with Nvidia,” Rishi explained, “to have our own foundation model called Dark Horse. This is trained on industrial time-series data. We train it, make it smart, and the results are quantumly better than generic large language models.”
