Brandon Barbello is co-founder and COO of Archetype AI, with over 10 years of product leadership across Fortune 500 and startups.gettyIf you were to picture physical AI, most people imagine the same thing: a humanoid robot walking through a warehouse, folding laundry, working a line.​I spent almost a decade building hardware, software and AI products at Google, and I'm now the cofounder and COO of Archetype AI. The physical AI conversation has taken off over the past couple of years, but the association with humanoid robots is only a sliver of what physical AI can truly do. It isn't just about building new embodiments via robots; it's about giving a brain to everything the world has already built.​Why We're Looking In The Wrong Place​Software and hardware don't move at the same pace. You can push a software update to a billion phones overnight. You can't do that with a power plant; it has to be certified and rolled out carefully on-site. Hardware and industrial equipment need to be manufactured, certified, serviced and trusted into workplaces one unit at a time.​Industrial robotics has been scaling since GM installed its first robotic arm in 1961, and it's still scaling gradually rather than following the explosive curve of a digital product. Even in manufacturing, the industry most associated with automation, fully autonomous "lights-out" factories remain rare; most facilities still depend on human oversight and likely will for years.​That's not bad news. It just means we should be looking more broadly for physical AI's breakout moment. Indeed, exciting things are happening where physical AI meets the fixed infrastructure the world has already built.​The Infrastructure Is Already There​The world is projected to spend $106 trillion building and maintaining infrastructure, from factories to power grids to ports, by 2040. Much of it is already instrumented: utility meters, building management systems, fleet telematics, SCADA networks on factory floors. The volume of data it generates is staggering—a single autonomous vehicle produces around 25 gigabytes a day; a modern aircraft generates roughly 20 terabytes of engine data per hour. ​In other words, we already have visibility into how our infrastructure operates. But this vast stream of data hasn’t been translated into intelligence. Most of these physical assets still run on fixed schedules and manual thresholds set years ago, leaving big gaps: The thermostat that doesn't know the building is half-empty on a Friday afternoon or the maintenance plan built around the calendar instead of the equipment's actual condition. This is where the forgotten opportunity lies: making existing assets more responsive.​What That Looks Like Today​The same pattern shows up across industries: The data exists, the control systems exist, and the two barely talk to each other.​Data centers already track power draw, cooling load and rack-level temperature, but cooling and power allocation are typically set against static baselines rather than adjusted in real time as compute load shifts. Manufacturing and energy equipment generate a steady stream of operational signals, but most of it is reviewed only after something breaks, not used to anticipate what is about to. Transportation and logistics networks run on onboard software and routing data, but that software optimizes for a preset plan, not for what's changing around it in real time.​The Shift That Actually Matters​This isn't about adding more sensors. Most of these environments are already over-instrumented relative to the data getting used. What's missing is a layer of intelligence general enough to make sense of the many very different kinds of physical data.​Much of that data speaks in languages a camera can't pick up: temperature, vibration, pressure, voltage. That's why language and vision models haven't already solved this. They weren't built to read these signals. A wind farm and a hospital HVAC system don't share a sensor stack, but they share the same problem: lots of signal, no system translating it into decisions.​What scales is a general-purpose model that applies across asset types without custom training for every site, closer to how a single language model can read contracts, emails and code rather than needing a different configuration for each. ​Already In The Field​This kind of intelligence is already being applied to real equipment. Data centers that give cooling control to a learning model, for example, have cut energy use dramatically, simply by making an already-instrumented system responsive instead of static. In our own work at Archetype AI, we met a wind energy operator who had spent two years hand-building a model to detect turbine anomalies. Our physical AI foundation model, Newton, applied to the same sensor feeds, replicated that detection capability out of the box—and surfaced nine additional anomaly types the hand-built system had missed, because it could analyze multiple sensors at once and find patterns no single signal reveals alone.​Two years of specialized work, matched and exceeded on day one. That's the gap between hand-tuning signal by signal, use case by use case versus a model that generalizes to all signals and all use cases. None of it requires sending sensitive data to a distant cloud, either. These models can run on-site, which matters where connectivity is limited and data is proprietary.​The Smarter Bet​You don't need to wait for humanoid robots to become commercially viable to start capturing value. You already have physical AI-ready assets on your balance sheet.Before investing in new hardware, audit what you already have. Most organizations generate far more operational data than they use, and making that infrastructure dynamic is faster, lower-risk, lower hanging fruit than buying new equipment.​Physical AI looks different when you think outside the humanoid. Our turbines, conveyors, compressors and pumps can be roboticized embodiments too. Physical AI should inhabit the assets the world already has, a transformation that is already quietly underway inside the buildings, machines and vehicles the world built long ago.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?