The Real Inflation Engine of AI Data Centers

Every analyst tracks semiconductor orders as the heartbeat of the AI boom. They stare at the 16.7% year-over-year jump in computer and electronic products and call it the story. It is not. The real economic action sits two categories down in the Census data, buried under the boring label “Electrical Equipment, Appliance, and Component Manufacturing.” Here, orders rose a mere 6.7% year-over-year in June. That sounds modest. In reality, that 6.7% is the most dangerous inflation signal for the entire AI infrastructure spend.

The Overlooked Angle

Most commentary treats the durable goods surge as a uniform demand wave. It is not. The critical distinction lies between categories where supply can scale quickly (semiconductors, with massive fab investments already underway) and categories where supply is structurally constrained. Electrical equipment? That is the constrained category. The 6.7% growth masks the fact that orders are already running against capacity ceilings, pushing lead times past 12 months and giving suppliers extraordinary pricing power. This is the overlooked mechanism: the AI buildout is not being slowed by chip shortages anymore. It is being slowed by transformers, switchgear, and circuit breakers.

Why This Small Detail Matters

A data center is a building full of computers that need power and cooling. The computers come from the computer and electronics category—yes, that spiked 16.7%. But the power distribution systems, the backup generators, the HVAC chillers, the transmission equipment? Those come from electrical equipment (NAICS 335) and machinery (NAICS 333). Orders for machinery jumped 14.4% year-over-year, including power generation and transmission equipment. This is not a demand blip. It is a structural shift in manufacturing capacity allocation that will take years to resolve.

Consider lead times. A typical large power transformer for a data center now requires 18 to 24 months from order to delivery. That is up from 6 to 9 months pre-2022. Why? Because the same suppliers serve utilities, renewables, and industrial plants. There is no spare capacity. The surge in orders for electrical equipment—$18.6 billion per month—is being booked against a fixed number of factories, each with a finite number of skilled workers and a finite supply of copper, steel, and specialized insulation materials. New factories take three to five years to permit and build. The result is a classic bottleneck: demand pushes against an inelastic supply curve, and price adjusts.

The Economic Mechanism

Let me walk through the mechanics. The electrical equipment manufacturing industry has high fixed costs: factories, tooling, engineering staff, compliance testing. Capacity utilization in NAICS 335 has been above 80% for the last 18 months, and in some subsegments like power transformers it is above 95%. When utilization hits those levels, incremental orders do not trigger more output. They trigger three things: longer lead times, higher prices, and allocation by suppliers to the most profitable customers.

Pricing power shifts from buyer to seller. In 2022, a typical 5 MVA transformer cost around $150,000. By 2026, that same unit is quoted at $350,000 or more, depending on specifications and delivery urgency. That is not inflation from input costs alone. Copper prices are up maybe 20% over the period. Steel is up 15%. The rest is pure capacity-constrained pricing: the supplier knows you cannot wait 24 months for an alternative, so they extract the scarcity premium.

Now layer on the pass-through effect. Hyperscalers (Amazon, Google, Microsoft) sign multi-year contracts for data center construction. Those contracts typically include escalation clauses tied to producer price indexes. The PPI for core goods rose 5.1% year-over-year in June. But the PPI sub-index for electrical equipment likely rose more. Every percentage point of PPI increase eats directly into project margins. If you are a hyperscaler building 100 data centers at $500 million each, a 5% cost overrun is $2.5 billion in unexpected expenditure. That is not pocket change.

The Strategic Consequence

The beneficiaries are clear: established electrical equipment manufacturers with existing capacity and long client relationships. Companies like Siemens, ABB, Schneider Electric, and Eaton now hold pricing power they have not had in decades. They can choose which projects to serve, and they will prioritize hyperscalers with deep pockets and repeat orders over smaller colocation providers. The losers: smaller data center operators, enterprise IT teams building on-premise facilities, and any new entrant trying to compete with the cloud giants. They will face longer lead times and higher prices, effectively making them uncompetitive.

There is a second-order consequence: project delays. When electrical equipment orders surge 6.7% year-over-year but capacity grows at 2% per year, the backlog grows. Census data does not show backlogs directly, but the Institute for Supply Management’s index for supplier deliveries for electrical equipment has been above 60 (indicating slowing deliveries) for over a year. Every month of delay pushes out revenue for the hyperscaler and extends the time before AI compute capacity comes online. That delay has a cost: opportunity cost of lost AI revenue and of capital tied up in unfinished buildings.

What Most Commentary Gets Wrong

Financial media loves to talk about the “AI capex supercycle.” They point to the 12.5% year-over-year surge in core capital goods orders and conclude that business investment is booming. That is true, but it misses the distribution. The bulk of that growth is in machinery and electrical equipment, not in general industrial automation. It is a one-sector boom concentrated in data center infrastructure. And within that sector, the bottleneck is not chips or servers—those are abundant now—but the mundane electrical backbone.

Commentators also assume that higher orders automatically translate into higher GDP and productivity. That is false when the orders create inflation without corresponding output because supply is constrained. The $85.1 billion in core capital goods orders in June is a nominal figure. If 20% of that is price increase rather than real volume increase, then real investment growth is far lower than the headline suggests. The inflationary component is not just a statistical footnote; it is a direct drag on the economy’s efficiency, because it forces capital toward chasing price increases rather than expanding capacity.

The Hard Business Lesson

The hard lesson is this: the next constraint on AI growth is not algorithmic or computational. It is industrial. The AI industry is discovering that its exponential demand curve hits a linear supply curve in the physical world. Electrical equipment manufacturing is not a sector that can ramp quickly. It requires specialized labor, heavy capital, and regulatory approvals. The hyperscalers cannot just throw money at it—they tried, and the result is 24-month lead times and 50% price increases.

The strategic implication for any business leader: if you depend on AI infrastructure, you need to lock in electrical equipment contracts with multi-year schedules and price protections now. Do not assume that order growth will moderate. The 6.7% growth in electrical equipment orders is the canary in the coal mine. When that category starts to decelerate, it will mean demand is being rationed by price, not satisfied by supply. That is the real indicator of a cooling AI capex cycle.

For now, the bottleneck remains. And every data center project without a transformer allocation is a project that will come online late and over budget. That is the boring, profitable reality that the headlines miss.