The most revealing Honeywell product may still be the one that started the story. In 1885, Albert Butz patented a furnace regulator that could adjust a damper without a person trudging downstairs. It sensed a physical condition, compared that reading with a desired state and made a small correction. Honeywell’s new corporate shape is built around the same loop, only now the furnace is a refinery, a hospital campus, a data center or a power grid - and the correction may come from software trained on decades of operating data.
On June 29, Honeywell Aerospace separated into its own public company. Eight months earlier, advanced materials had become Solstice Advanced Materials. Honeywell Technologies kept the familiar HON ticker, the Charlotte headquarters and a deliberately narrower job description: automation for buildings, industrial operations and process plants. The separation converted one of America’s remaining industrial conglomerates into three cleaner stories. Honeywell’s is the one about making physical systems more aware of themselves.
This is not the household thermostat business many people associate with the name; that operation was spun into Resideo in 2018 and licenses the Honeywell Home brand. Today’s Honeywell sells the nervous systems behind commercial buildings and industrial sites: sensors to notice, controllers to decide, software to interpret, actuators and equipment to respond, and service teams to keep the loop working years after installation.
The dull machines are where the drama lives
A chiller does not trend on social media when it runs properly. Neither does a gas detector, fire panel, catalyst bed or distributed control system. Their importance becomes obvious only when they fail. Honeywell operates in that inverse-attention economy. Its customers pay to reduce unplanned downtime, energy use, safety incidents, bad batches and the time it takes an inexperienced operator to understand an alarm.
The customer list follows the infrastructure. Building owners use Honeywell to coordinate heating, cooling, access, video, fire detection and energy management. Data centers need cooling and power controls that can absorb rising density without sacrificing uptime. Hospitals and hotels need comfortable rooms and auditable life-safety systems. Refineries, chemical plants and LNG facilities use Experion control systems, UOP process technology, catalysts, pumps, compressors and lifecycle services. Manufacturers and utilities buy sensing, measurement, worker-safety and automation products.
“The problem was never the data. It was that the data lacked context.”Honeywell’s explanation of Forge
That line contains the company’s software thesis. Industrial sites already produce torrents of temperatures, pressures, vibration readings, alarms and maintenance records. A generic analytics tool can graph them. Honeywell Forge tries to understand what each signal means for a specific pump, room, production line or process. It is the intelligence layer spread across more than 32,000 customers, 324,000 sites and roughly 6.4 million connected assets as of March 2026.
Three businesses, one increasingly talkative layer
The current portfolio divides into three segments. Building Automation covers controls, fire and life safety, security and energy management. Process Automation and Technology combines plant controls and services with process licenses, modular equipment, catalysts and purification technology. Industrial Automation covers sensing, measurement, safety and productivity systems used by factories, utilities and other industrial customers.
Rooms to campuses
Comfort, access, fire, security, energy and uptime.
Molecules to margins
Control, catalysts, licenses, pumps and plant optimization.
Signals to action
Sensing, measurement, safety and production workflows.
Context across all three
Asset data, domain models, AI applications and services.
The organizing idea is that Forge can sit across all three. Vanderbilt University, for example, connected more than 100 campus buildings and over eight million square feet into a hardware-agnostic interface. A facilities team gets one operating view instead of a scavenger hunt through proprietary dashboards. In a process plant, the same logic becomes more consequential: combine live readings, historical behavior and operating constraints to warn a person before an abnormal condition turns into a shutdown.
Experion Operations Assistant, commercially launched in March, was piloted by Chevron and TotalEnergies. Experion Cognition goes a step further. In a proof of concept at Borouge International’s Ruwais complex, AI agents made recommendations and selected automated decisions inside a control-system environment. Honeywell says multiple pilots predicted selected alarm incidents an average of five to ten minutes early. That is not enough time for a coffee, but it can be enough time to avoid a costly surprise.
Why the old hardware matters to the new AI
Honeywell’s distinction is not a larger language model. Siemens, Schneider Electric, ABB, Emerson, Rockwell Automation, Johnson Controls and specialist software vendors all want parts of the same budget. The differentiator Honeywell claims is the combination of domain knowledge, installed equipment and operational history. More than 7,000 engineers work in the current company, which holds over 25,000 patents and applies for roughly 650 more each year.
In industrial settings, expertise is not decorative. An AI system can be probabilistic when it drafts an email. A control system governing pressure or combustible gas needs deterministic boundaries: known rules, safe states and predictable responses. Honeywell’s approach layers AI recommendations around controls that still respect physical constraints. It is less magical than the usual AI pitch and more useful for a plant manager who must explain every incident.
The moat is not merely data. It is knowing which reading matters, what “normal” means and when software must hand the decision back to a human.
The approach also addresses a demographic problem. Experienced operators are retiring while facilities grow more complex. Capturing their process knowledge in models, playbooks and explainable recommendations can shorten the path to competence for newer workers. It does not remove accountability. It changes the operator’s job from watching every dial to supervising exceptions, verifying advice and intervening when the system reaches the edge of what it knows.
The money is in the long relationship
Honeywell earns money several ways at once. It sells devices and engineered equipment. It designs and integrates projects. It licenses process technology and supplies catalysts. It charges for software, connected services, maintenance, upgrades and aftermarket support. A controller may remain in service for decades, creating a customer relationship longer than most software companies have existed. Forge can turn that installed base into a recurring digital business without asking customers to replace every machine first.
The post-spin company produced $5.187 billion in second-quarter sales, with orders up 16 percent and backlog near $20 billion. Management now expects $19.8 billion to $20 billion in 2026 sales. Those figures describe the focused automation perimeter. The $37.442 billion number attached to Honeywell’s 2025 results included businesses that have since left, so it is a poor shorthand for the company standing today.
The portfolio is still moving. Honeywell completed its £1.325 billion purchase of Johnson Matthey’s Catalyst Technologies business in July, adding process catalysts and purification capabilities. It had already bought Sundyne’s pumps and compressors, Carrier’s access-solutions business and Air Products’ LNG process technology. At the same time, it agreed to sell its warehouse and workflow operation and its productivity-solutions business. This is not subtraction for tidiness alone. Honeywell is clustering around places where process knowledge, equipment, controls and software can be sold together.
A place in the market between software and steel
Honeywell occupies an awkward, valuable middle. It is too physical to be treated as a conventional software company and increasingly too digital to be judged as a box maker. Its competitors range from diversified automation giants to cybersecurity firms and narrow sensor specialists. Customers can assemble a patchwork of them, keep older control systems running, or build integration themselves. Honeywell’s wager is that one accountable supplier across more of the stack can reduce the seams where information and responsibility get lost.
There are limits to that promise. Old equipment is stubborn. Industrial upgrades move through budgets, shutdown windows and safety reviews. Connecting operational technology increases the importance of cybersecurity. An autonomous recommendation is useful only if a customer trusts the underlying data, understands the reasoning and knows who remains responsible. Honeywell’s scale helps it enter the room; it also raises the cost of getting those details wrong.
Still, the company has landed on a timely definition of progress. The future of industrial AI will not arrive as a humanoid strolling across every factory floor. It will show up as a building that trims demand before an energy peak, a technician who sees the right procedure beside the right machine, and a control room that spots trouble five minutes sooner. The machines do not need personalities. They need context, boundaries and a useful next move.
That returns Honeywell to Butz’s basement, where a mechanical regulator quietly made a judgment that once belonged to a person. The scale has changed. The loop has not.