Double. That’s the size one of America’s oldest manufacturers expects to reach, according to its CEO, and the reason isn’t a new product line or an overseas expansion. It’s the AI boom. The same boom that fills my review queue with agent frameworks and prompt orchestration dashboards is now doubling the footprint of a company that was making physical things in this country long before anyone typed a system prompt.
I review AI tools for a living. I install them, break them, and tell you whether they earn their subscription. That job has always felt weightless, a matter of tokens and API keys and monthly plans. Stories like this one are a reminder that the weightless part is an illusion sitting on top of something extremely heavy.
The part of the stack I never had to think about
When I benchmark an agent tool, my cost model has maybe four inputs. Subscription price. Token spend. Time to set up. Time lost to debugging. That’s it. Nothing in my spreadsheet accounts for land, water, or electricity, because none of it shows up on my invoice.
But the reporting piling up right now says those costs are real and landing somewhere. Consumer Reports has been looking at what AI data centers mean for electric bills and water. CNET framed it more bluntly, describing data centers coming for land, water, and power. Meanwhile Frontieras North America is publicly repositioning coal for what it calls the AI economy. Coal. In 2025 or thereabouts, a fuel source that most of the tech industry had mentally filed under history is being pitched as an input to chatbots.
That’s not a moral lecture, it’s a supply chain observation. Something has to be built, dug, cooled, and wired for my agent a PDF. The manufacturer doubling in size is on the winning end of that chain.
Why a reviewer should care about steel and turbines
Here is the practical reason this matters to anyone picking tools, not just anyone with opinions about energy policy.
- Pricing is not settled. If the physical layer under these tools is expanding this fast, the cost of running it is still being discovered. Vendors offering unlimited tiers today are making a bet on future compute prices, not stating a fact about them.
- Availability follows infrastructure. Rate limits, regional outages, and mysterious slowdowns are downstream of capacity. When you read that data centers are competing for power and land, you’re reading the root cause of the 429 error in your logs.
- Lock-in gets heavier. The more capital sunk into this buildout, the more incentive providers have to keep you inside their ecosystem. Portability is a feature worth paying for.
- Efficiency becomes a real differentiator. A tool that gets the same result with fewer calls used to be a nice-to-have. Under rising infrastructure costs, it’s a durable advantage.
The uncomfortable symmetry
Kai Williams put together 16 charts to explain the AI boom, and the existence of that piece tells you something on its own. Sixteen. It takes that many views to describe what’s happening, because this isn’t one story. It’s a software story, a capital story, an energy story, and now a heavy-manufacturing story, all wearing the same name.
The symmetry I keep circling is this. The AI tools I review are marketed as ways to do more with less. Less headcount, less time, less busywork. And the infrastructure underneath them is scaling by doing more with more. More land, more power, more factory floor. Both things are true at once, and only one of them shows up in the demo video.
What I’m changing about how I review
Not much, honestly, because I can’t audit a data center from my desk and I’m not going to pretend otherwise. What I can do is stop treating compute as free and infinite in my scoring. Three adjustments:
- I’ll note when a tool is wasteful by design, meaning it fires off dozens of calls to accomplish what a tighter prompt chain does in three.
- I’ll weight local and smaller-model options more generously than I used to. They’re often less impressive in a demo and far more predictable in a monthly bill.
- I’ll be more skeptical of pricing that looks too good. Somebody is subsidizing it, and subsidies end.
An old-line American manufacturer doubling its size because of AI demand is a genuinely interesting piece of news, and it cuts against the story the industry tells about itself as a purely digital phenomenon. The tools on my bench are the visible tip. The rest of it is being welded, poured, and plugged in somewhere you’ll never see, and eventually the cost of that finds its way back to your invoice.
Buy tools that would still make sense if compute got more expensive. That’s the whole review, compressed.
🕒 Published: