AI is not floating in a cloud somewhere. It is a physical industry, and the clearest proof is that one of America’s oldest manufacturers says the AI boom has it on a path to double in size, according to its CEO in a recent Fox News report.
I spend most of my time here explaining AI agents in plain language: what they do, how they decide, why they sometimes get things wrong. But a story like this one deserves attention from anyone trying to understand this technology, because it points at something the chat window hides. Every agent you ask an email or book a flight is running on machines that had to be built, powered, and cooled by companies with loading docks.
Old-line manufacturing meets new-line software
The detail worth sitting with is the age of the company. An old American manufacturer doubling in size is not the kind of headline the last two decades of tech coverage trained us to expect. The story we were told was that software would eat the world and the factories would keep shrinking. The AI buildout is running the other direction. Data centers need steel, concrete, transformers, cooling equipment, electrical gear, and the industrial capacity to make all of it at volume.
That demand does not care whether a supplier is a startup or a century old. It cares whether the supplier can produce. Which is why a manufacturer that has been around longer than the transistor can suddenly find itself with a growth curve that looks like a tech company’s.
You can see the same reframing happening in stranger corners. Frontieras North America, per a TradingView piece, is pitching a rethink of coal for what it calls the AI economy. I am not endorsing that pitch, and I do not have enough detail on it to evaluate the claims. What I notice is the positioning: an energy-adjacent company deciding that “AI” is the frame that makes its business legible to investors right now. When companies start renaming their purpose around your technology, your technology has left the software category.
The part that shows up on your bill
The other half of this story is less flattering, and it is being covered carefully by outlets that do not normally write about AI at all. Consumer Reports has looked at how big tech data centers affect electric bills and water use. CNET has framed it more bluntly, arguing AI data centers are coming for land, water, and power.
These are consumer publications. They cover washing machines and cell plans. When they start writing about the resource footprint of AI infrastructure, it is a signal that the costs have stopped being abstract and started showing up in household budgets and local zoning fights.
For readers of this site, that reframes a question I get a lot. People ask whether AI agents are worth using, and they usually mean: will this save me time? The manufacturing and utility angles add a second version of the question. What does it cost, and who pays it? A single agent request is trivially cheap to you. Multiplied across millions of users, it becomes a substation, a cooling system, and a line item on a regional grid operator’s forecast.
Why the charts matter
Kai Williams put together a set of 16 charts at understandingai.org explaining the AI boom, and that format is telling in itself. The boom has become large enough and complicated enough that you need a dozen and a half different measurements to describe it. Not one metric. Not one story. Compute spending, model capability, energy draw, capital flows, adoption rates. They do not all move together, and some of them are in tension.
That is the honest state of things. The boom is real enough to reshape an old manufacturer’s growth plans and real enough to worry consumer advocates about utility rates, at the same time, from the same underlying cause.
What to take from this
If you are a non-technical reader trying to build an accurate mental model of AI agents, add these pieces to it:
- Agents run on physical infrastructure that someone builds and someone powers.
- That infrastructure is creating real industrial demand, including for companies that predate computing.
- It also creates real local costs in electricity, water, and land that are now being tracked by mainstream consumer press.
- “Is AI a bubble?” and “is AI physically expensive?” are separate questions with separate answers.
You do not need to resolve the debate to use these tools well. You just need to stop picturing AI as weightless. The next time an agent handles a task for you in two seconds, know that the reason it can is partly a factory somewhere, running a second shift.
🕒 Published: