Honeywell has begun testing a new AI platform called Forge Cognition that layers AI agents and digital twins over the sensors and control systems already running inside large buildings, with Nvidia and Dell as its first major partners. The project traces back to a challenge Nvidia's leadership put to Honeywell at CES in January 2026, after which Honeywell's operations team in Santa Clara had about 30 days to put together a pitch and roughly 90 days to show real results before Honeywell's June investor day.
The first real test ground was Nvidia's own 750,000 square foot Santa Clara headquarters, known internally as Voyager, where the system now watches close to 900 pieces of building equipment. Suresh Venkatarayalu, chief technology officer of Honeywell Technologies, said the software cut unnecessary building management alerts by as much as 90%, addressing a problem where historically 90% to 95% of alarms turned out to be false. Separately, the platform identified roughly 10% in potential energy savings by rethinking how heating and climate zones across the building are managed.
Forge Cognition runs on Nvidia's Nemotron model, trained on Honeywell's own engineering knowledge, paired with Nvidia's Omniverse software to build digital replicas of equipment and floors, and Dell hardware installed on site to run the system locally. The platform currently operates in manual or semi-autonomous modes, answering plain language questions through a chatbot interface and flagging likely equipment failures before they happen, but Venkatarayalu said full autonomy is being held back on purpose. He said he wants "10 out of 10 accuracy all the time" before letting the system act without a human in the loop, adding that he personally would not deploy such agents without a control system still in place, and that Honeywell expects to keep operating in this supervised mode for roughly the next two to three years.
