The Federal Reserve's monetary-policy report puts artificial intelligence on both sides of the price ledger. Building the systems creates present demand for electricity, chips and construction materials. Productivity may eventually reduce costs, but the report does not say when that dividend will arrive or how large it will be. [1]

That timing extends Thursday's account of Meta's 14-gigawatt target meeting a transformer shortage. The paper separated a corporate capacity target from equipment orders, interconnection and operating load. The Fed now supplies the macroeconomic version of the same sequence: physical inputs are bought before the promised efficiency is measured.

The distinction is easy to lose because AI investment is commonly described as a productivity story. That may prove correct. The report leaves the possibility open that new tools could let workers and companies produce more at lower unit cost. It also records the prior step, in which data centers compete for power, semiconductors and building materials. [1]

Those are not contradictory claims. They occur on different clocks. A transformer, chip tray or cooling system is paid for during construction. A productivity gain appears only after software is deployed, work changes and output can be compared with the labor and capital used to produce it. The report does not quantify either AI's share of current inflation or its future disinflationary effect. [1]

That limit matters. Calling AI permanently inflationary would turn a timing observation into a forecast the Fed did not make. Calling it already disinflationary would book a benefit before the report can identify its arrival. The defensible conclusion is narrower: the buildout creates demand now, while the offset remains uncertain.

AI discourse on X tends to collapse this interval. Product demonstrations become evidence of economy-wide efficiency, while infrastructure critics treat every power project as proof that the technology can never repay its cost. No verified topical X post surfaced in the research for this article, so neither camp enters as attributed evidence. Reuters' account provides the institutional record instead. [1]

The report is also not a rate decision. It gives Congress the Fed's account of inflation pressures and growth conditions; it does not assign votes or prescribe a permanent policy response to AI. Its value is diagnostic. It places the server campus inside the inflation discussion without pretending to know the date on which a better model will begin lowering prices.

The next evidence will need units. Power demand can be measured in megawatts, construction in dollars and chips in deliveries. Productivity requires output per unit of labor or capital over a stated period. Until both sides are measured, the honest order remains the one the Fed recorded Friday: costs first, possible gains later.

-- HENDRIK VAN DER BERG, Brussels