We hear that question from prospective customers constantly. Over eight years and Xemelgo deployments across 100+ manufacturing facilities, it's come up more than any other. We like the question. Not because there's an easy answer, but because it points to something bigger than our company.
Every technology revolution follows the same arc. It starts with a breakthrough in infrastructure. Personal computers became affordable. The internet connected the world. Cloud made compute available on demand. GPUs made large scale AI practical.
Then the first generation of companies tries to build directly on top of that infrastructure. It doesn't look that hard at first. Eventually, everyone discovers they're solving the same problems over and over. That's when a new layer emerges.
Windows abstracted away the differences between hardware. Cloud platforms abstracted away servers, storage, and networking. Foundation models now abstract away much of the complexity of AI. Every time, software makes the underlying infrastructure usable at scale.
We call the equivalent layer in the physical world Operational State. It's the real time picture of what's happening on a shop floor, in a warehouse, or across a supply chain, distinct from what an ERP or WMS assumes is happening. Building and maintaining that layer is the next platform opportunity, the same way Windows and cloud were before it.
Most enterprises already own an ERP and a WMS. They're starting to invest in sensors, hoping to plug them directly into those systems of record. So, the objection makes sense: "We already have the hardware. We already have SAP. Do we need another layer in between?"
They're not wrong to ask, and they're not wrong that they can build a lab version of this. We've watched Fortune 50 companies attempt it. Some of them are our customers today, after their homegrown builds hit a wall in production. The pattern repeats too often to treat each case as a one off.
The build usually starts with a simple goal: read data from a few sensors and send it to an ERP. Then reality shows up.
Readers are noisy. Most of the data is irrelevant. Business processes change. New hardware models get introduced. Someone asks for a new use case. Another site is configured differently than the last one.
Before long, the team isn't maintaining integrations anymore. They're maintaining the operational understanding of an entire facility, building a digital representation of every workflow and every site. That isn't integration. That's a Platform, whether your company planned to build one or not.
The shape of Physical AI changes by vertical, but the underlying gap is the same everywhere: what the digital system says, versus what's actually true on the ground.
In aerospace, that gap shows up as a compliance risk. SEKISUI Aerospace replaced clipboards and paper travelers with a connected factory, giving them full digital traceability from raw material to finished part at audit time. AS9100 audits stop being a source of dread when the paper trail is automatic instead of manual.
In retail, the gap shows up as inventory that's wrong the moment it's counted. Retailers use RFID to automate transactions and rapidly inventory entire stores, cutting a task that used to take hours down to minutes. Shelves reflect what's actually there instead of what the system assumed from last quarter's inventory audit, resulting in higher sales and increased customer satisfaction.
In manufacturing, the gap shows up as lost time hunting for work orders. Curtiss-Wright went from unable to locate open work orders to finding them in minutes, and cut production lead time by 85% with real time WIP tracking.
Three industries, three different symptoms, one shared root cause.
The more mature these platforms get, the less visible they become:
Nobody opens Windows and thinks about device drivers.
Nobody spins up a cloud server and thinks about distributed storage.
Nobody calls an AI API and wonders how GPUs are being scheduled.
With every revolution, the complexity of the technology itself has disappeared behind a clean abstraction. That's exactly what will happen with Physical AI.
There will be companies building sensors, companies building robots, and companies building AI applications. Between all of them, there will be a platform responsible for turning billions of noisy observations into something every downstream system can trust.
With 43 billion RFID tags shipped last year alone, the raw data volume already exists in our realm. What's missing in most operations isn't the sensors. It's the layer that makes their output trustworthy.
So when customers ask whether they should build it themselves, our answer is usually different from what they expect...
The better question is whether building and continuously evolving that platform is where you want your time and resources to be spent over the next five years.
History suggests that every major computing revolution eventually rewards specialization.
Physical AI won't be any different.