Cooler on paper, hot underneath

September 30, 2026

The Personal Consumption Expenditures (PCE) index, the Federal Reserve’s inflation target, rose 0.3% in August and 3.4% from a year ago. Both measures undershot expectations of 0.4% and 3.7%. Benchmark revisions shaved about 0.3 percentage points from annual inflation. The data cooled. The economy did not.

The reset centered on financial services and computer software and accessories. July headline inflation has been revised now to 3.4%, not 3.7%. The measuring stick moved. The inflation problem did not. This was about methodology but does not change the experience consumers are feeling in their wallets.

The Fed saw the revisions coming and expected more relief than markets did. It raised rates anyway. Revisions can cool one report. They cannot cure persistent inflation.

Core PCE rose 0.2% in August and 3.0% from a year ago. In comparison, core CPI was lower at only 2.4%. Core PCE has run hotter for 10 straight months, the longest inversion since 1983. The usual relationship flipped. Not in the Fed’s favor.

Super core services PCE, excluding energy and housing, jumped 0.4% in August after edging up 0.1% in July. The annual rate ticked up to 3.5% from 3.4%. Revisions shaved 0.4 percentage points from the annual rate. It is still 1.5% above its pre-pandemic pace. Sticky is an understatement.

Transportation services formed the outlier, up 1.4% in August. Financial services and food services and accommodations surged too. Health care was tamer, but still rose 3.2% from a year ago, well above its pre-pandemic pace.

San Francisco Fed analysis places much of healthcare in the noncyclical bucket. Administered payments and industry forces matter more than the business cycle. Rate hikes can crush housing. They cannot easily derail hospital prices.

Health contracts reset annually. More inflation is already baked into the cake. Preliminary KFF data point to a 14% jump in small-group premiums in January, after an 11% increase at the start of 2026. The pressure arrives with a lag. It does not vanish on command.

The Fed cannot reduce hospital charges. It must squeeze the rate-sensitive economy harder to offset the inflation it has less power to reach. That is the brutal trade-off.

September warning. August food and energy readings are already in the rearview mirror. Gasoline jumped from about $4.07 per gallon at the end of August to $4.48 by September 21. That double-digit surge will hit headline inflation directly.

Diesel surged from about $5.60 to $6.53. That is the larger second-round threat. It raises costs for trucking, rail, agriculture and construction. Those costs do not stay on loading docks. They migrate into food, goods and delivery fees.

Persistence is the larger threat. After five years, consumers and businesses have developed muscle memory for inflation. Pair it with robust growth and inflation becomes harder, not easier, to derail.

Consumers tap savings to spend

After-tax income rose 0.3% in August and 4.8% from a year ago. Inflation erased the monthly gain; real disposable income, which is net of taxes, was flat. Private wages and salaries and government social benefits did the lifting. Social Security is surging as the tail end of the baby boom retires.

Consumers spent anyway. Personal consumption expenditures surged 0.9% in August and 6.1% from a year ago. Real spending jumped 0.6% for the month and 2.6% over the year. Energy, vehicles and recreation led goods. Food services and accommodations led services. Savings kept the party going.

Spending grew three times faster than disposable income, even after a large upward revision to income. The saving rate fell to 4.1% from 4.6% in July. The cushion is larger than we thought. It is shrinking fast.

Benchmark revisions raised the history of personal and disposable income and, with it, saving. Better tax, business and government records found household resources the earlier estimates missed.

Compensation accounted for the largest upward revisions: wages, salaries and employer-paid benefits. Benefits are rising so fast they are crowding out wage gains. Proprietors’ income, interest and dividends, and government social benefits also moved up. Medicaid and Medicare count as income, even when households never see the cash.

Better AI data, stronger GDP

Real GDP grew at a revised 2.2% annualized pace in the second quarter, up from 1.5%. First-quarter growth rose to 2.5% from 2.1%. The revisions did more than lift two numbers. They exposed a stronger capital spending cycle.

Gains were broad-based. Stronger investment, consumer spending, inventories and government spending drove the second quarter upgrade. Real final sales to private domestic purchasers rose 4.6%, up from 4.2%. Private demand ran at more than twice the headline GDP rate. The economy had more torque than the first estimate showed.

Better source data made the difference. The Bureau of Economic Analysis (BEA) added the Census Bureau’s Annual Integrated Economic Survey, IRS records, federal budget data, National Science Foundation surveys and revised seasonal factors. The fog around capital spending began to lift.

Equipment had been undercounted. The revisions made the level of data center investment 34% higher by the second quarter than it was pre-revisions. Earlier surveys were voluntary; hyperscalers disclosed little about the timing and scale of their buildouts. The new data closed part of the gap. 

Structures likely remain undercounted because projects are leased, routed through special-purpose entities or recorded with long lags. The blind spot narrowed. It did not disappear.

Data centers are no longer a rounding error. AI-related construction, computing hardware and networking equipment reached about 0.8% of GDP in early 2026. The broader computing-infrastructure footprint was roughly 1.5%, more than double its 2015–2022 average. BEA does not publish a clean AI contribution. 

The five-year growth path changed less than the latest quarters. The upgrade was back-loaded and concentrated in the AI investment boom and wealth it generated. The expansion was not rewritten. Its engine was.

More capital, not yet a productivity miracle

The revisions suggest productivity was stronger in the first half, helping explain resilient profits. Tech firms captured much of the gain. Capital deepening is visible. Economy-wide transformation is not.

The problem is scale. AI productivity is showing up at individual firms, not yet across the economy. For now, the data-center boom adds more to demand and inflation than to economy-wide efficiency. We have counted the concrete, chips and servers. We have not yet proved a productivity dividend.