The Fed Now Points to Data Center Buildout as a Key Inflation Driver. Here’s What Comes Next.

The Federal Reserve this week placed the data-center buildout inside America’s inflation problem. 

Not at the edge of it. 

Not as a future concern.

Inside it now—alongside tariffs, energy shocks and the other forces keeping prices above the central bank’s 2 percent target.

That is the unmistakable message in the minutes of the Federal Open Market Committee’s July meeting. The Fed held its benchmark rate at 3.5 percent to 3.75 percent, but three members wanted an immediate quarter-point increase. Staff attributed elevated inflation partly to “the surge in demand related to the AI buildout.”

“Some participants observed that materials for data centers, such as chips and steel, had registered large price increases and that consumer items such as smartphones, computer equipment, software, and electricity had also been subject to price pressures,” the minutes state.

The data-center boom has usually been described as a race for chips, land and megawatts—or as the capital-investment engine holding up economic growth. The Fed is now describing that same race as a channel through which inflation can move from an industrial construction site to household budgets — and to include the cost of human labor.

The minutes are, as always, carefully crafted. Several officials said AI’s price effects had so far been confined to selected categories. Others said the investment wave was already pushing up demand more broadly, or would soon. Some believed AI productivity would eventually lower costs and expand supply. The disagreement was about timing: the factories, substations, transmission equipment and workers must come first. The productivity payoff comes later.

Labor is already flashing red. Fed officials cited “strong demand for skilled workers—including electricians, machinists, and engineers—leading to notable increases in their wages.” They are the people needed to wire server halls, manufacture electrical gear, expand the grid and build or restart nuclear plants. A shortage in that workforce can delay a campus and make every power project feeding it more expensive.

The hardware constraint is no better. Lead times for critical grid equipment now run two to four years, and some utilities report paying four to nine times what transformers once cost. Money can reserve a factory slot. It cannot instantly create one.

Now the cost of that money may rise. The minutes say investors were fully pricing a quarter-point increase by the Fed’s September meeting and another by the end of the first quarter next year. J.P. Morgan has pulled its own forecast forward and now expects a hike in December. Michael Feroli, the bank’s chief U.S. economist, wrote in a J.P. Morgan Global Research briefing that “hot inflation readings could result in a hike as early as September.”

The problem is not only whether rates move, but how violently markets move while waiting. Goldman Sachs chief U.S. economist Jan Hatzius warned in a research note quoted by Fortune that investors may have “less information and potentially more inaccurate beliefs.” J.P. Morgan’s Alex Wolf told the same publication that the bond market’s reaction reflected “newfound uncertainty.” For long-duration projects, uncertainty becomes another line in the financing model.

The Fed is watching the borrowing itself. Officials highlighted the “rapid buildout of AI-related infrastructure” and the growing use of debt, including credit from nonbank investors and regional banks. They warned that a downgrade in AI earnings expectations could reprice assets, tighten financial conditions and strain institutions exposed to the sector.

That does not mean the buildout stops. It means money becomes more discriminating. Campuses with contracted tenants, secured equipment and credible power will separate from projects built on speculative demand. Lenders will press harder on leverage, completion guarantees, refinancing risk and who ultimately stands behind the electricity bill. Nuclear and other capital-heavy generation face the same split: stronger demand can support long contracts, while higher rates punish projects that need years of construction before producing revenue.

The feedback loop is ugly. Data centers push up demand for scarce grid hardware, skilled labor and electricity. Those costs strengthen the inflation case for higher rates. Higher rates then raise the price of the campuses and power plants meant to relieve those shortages. Developers may respond by ordering earlier and hoarding equipment, tightening supply for utilities and manufacturers outside the AI race.

Power markets are already moving in that direction. PJM’s new large-load proposal would require big new loads to bring their own new capacity—building or securing generation—or see the unsupported portion of their demand placed earlier in line for emergency curtailment. The grid operator says data centers account for roughly 30 of the 32 gigawatts of demand growth it forecasts between 2024 and 2030.

That is what comes next: tougher financing, harder proof of power and more pressure to keep costs off ordinary customers. The buildout is still moving. But the Federal Reserve has now put a price on its speed—and every data-center developer, utility and reactor builder will have to show who pays it.