America’s 3.4% Boom and the Cost of Waiting

That difference in timing is becoming central to understanding the American economy.

The third-quarter GDP report shows an economy capable of financing substantial investment and sustaining demand. It also raises a harder question: how widely, and how quickly, will the benefits reach the people and businesses carrying the costs?

My reading is that America’s main vulnerability lies in the gap between the immediate demands of building its next economic model and the eventual income that model must generate.

The advance estimate puts real GDP growth at 3.4% annualised in the third quarter, following 2.2% in the second. That translates into approximately 0.84% growth over the preceding three months. Private domestic final demand increased at a 4.9% annualised rate.

Those are meaningful signs of strength.

Consumer spending expanded at 3.8% annualised. Business investment excluding housing rose 12.7%, with information-processing equipment particularly strong. Residential investment, however, contracted 3.3%.

The contrast matters. America is simultaneously expanding productive capacity in some areas and struggling to sustain investment in others.

It also matters how we interpret the accounting.

Reuters reports that inventory rebuilding contributed 1.61 percentage points to quarterly growth, while imports subtracted 2.64 points. These figures help explain the headline, but they cannot be treated as independent estimates of what the economy would have done without inventories or trade.

Imported equipment can appear first as investment expenditure. Imports are then subtracted to exclude foreign production. The same principle applies to imported goods held in inventories.

An imported server can therefore support the expansion of an American business without its entire purchase price representing American production.

This makes two popular calculations unreliable: adding global technology-company capital budgets and calling the sum a contribution to US GDP; or subtracting an estimated AI investment category and calling the remainder the economy “without AI”.

Both overlook connections among investment, imports, suppliers and subsequent production.

The economic contribution of AI requires careful measurement. Information-processing equipment, software and research spending are useful indicators, but none is a clean national-accounts category labelled artificial intelligence.

The stronger question is what the investment will produce after installation.

A capital budget creates commitments. An operating business must convert those commitments into useful output, paying customers and cash remaining after electricity, staffing, financing and equipment replacement.

GDP measures production. It does not validate a financing structure or establish that an investment will earn more than its cost of capital.

There is also a distinction between gross investment and the accumulation of economically useful capital. Equipment that becomes obsolete quickly may require repeated replacement. A large spending programme can therefore coexist with a less impressive increase in durable productive capacity.

The practical issue is the relationship between an asset’s useful life, its revenue-generating life and the maturity of the obligations used to finance it.

A building, its power infrastructure and its computing equipment do not necessarily share the same economic lifespan. Financing them as though they did can conceal risk until utilisation or pricing disappoints.

None of this requires the technology to fail.

AI could become widely useful while competition pushes prices down and transfers much of the benefit to customers. Society could gain through cheaper services and higher productivity while some infrastructure investors earn disappointing returns.

That possibility deserves more attention than the assumption that technological importance automatically confers pricing power.

For households, the distinction between activity and financial improvement is already visible.

In September, real disposable personal income fell 0.3%, real consumption was essentially unchanged, and the personal saving rate stood at 3.8%. The quarter’s strong consumption growth therefore ended with weaker monthly momentum.

One month cannot establish a sustained downturn. It can identify a question that a quarterly average conceals: how much future spending can be supported by recurring purchasing power?

BLS reported that real average hourly earnings were 0.7% below their level a year earlier. Real average weekly earnings were down just 0.1%, reflecting the contribution of longer working hours.

These are averages, not measures of the median household. They nevertheless demonstrate why higher aggregate spending need not mean workers can purchase more with an hour of labour.

Households can maintain expenditure through several channels: employment income, transfers, investment income, borrowing or a reduction in saving. Those channels carry different implications for resilience.

Consumption financed by durable real-income growth is easier to sustain than consumption requiring progressively smaller financial buffers.

The aggregate saving rate also cannot reveal whose buffer is shrinking. A comfortable national balance sheet can coexist with serious constraints among particular borrowers.

The credit evidence calls for discrimination rather than alarmism. The New York Fed’s second-quarter report showed broadly stable delinquency patterns across most products, with continuing pressure in auto loans and credit cards. That is a picture of uneven vulnerability, rather than evidence of an economy-wide credit collapse.

The July bank-lending survey was similarly mixed: commercial and industrial lending standards were broadly unchanged, while credit-card standards tightened.

An analysis that describes every household as insolvent or every business as starved of credit would be as misleading as an analysis that treats GDP growth as universal prosperity.

The labour market reinforces that point.

September payroll employment increased by 29,000, while unemployment remained at 4.2%. Hiring was modest, but unemployment had stayed within a narrow range since March.

The relationship between employment and output depends on hours, participation, labour-force growth and productivity. A smaller monthly payroll gain does not carry the same meaning under every demographic and economic condition.

Nor can strong GDP combined with weak payroll growth, by itself, prove that AI has delivered a productivity breakthrough. GDP and payroll employment cover different economic populations, and productivity requires an appropriate measure of hours worked.

The scheduled November 5 productivity release will provide more direct evidence for the third quarter. Even then, identifying how much improvement is caused by AI will require more than one aggregate observation.

Inflation adds another layer.

September’s PCE inflation rate was 3.6% year on year, with core inflation at 3.1%. The quarterly PCE measure, however, increased at a 2.1% annualised pace, with core at 2.5%.

These measures answer different questions. One compares September with the same month a year earlier; another compares quarterly averages and annualises the change.

A disciplined assessment has to consider both recent momentum and the accumulated annual increase. Selecting whichever window best supports a bullish or bearish argument produces a weaker analysis.

It is equally important to avoid making AI responsible for every inflationary pressure.

The San Francisco Fed’s October analysis identified technology-equipment price pressures associated with the investment boom, while finding their direct contribution to aggregate inflation relatively limited. Energy and tariff-related goods pressures remained important.

AI can influence prices in opposing directions. Infrastructure construction increases demand for resources immediately. Successful adoption can subsequently reduce the resources required to produce a given output.

The timing of those effects is uncertain. Monetary policy must operate while that uncertainty remains unresolved.

On October 28, the day before the GDP and personal-income releases, the Federal Reserve held its target range at 3.75–4.00%. Two members preferred a quarter-point increase.

That decision leaves the next move open.

The distribution of monetary restraint is particularly important here. A company able to fund investment internally may tolerate conditions that make a mortgage or a smaller firm’s expansion uneconomic. A highly leveraged infrastructure developer may be much more exposed than a technology company with substantial existing operating income.

The relevant dividing line is financing capacity, rather than simply “technology” versus “the rest”.

PIMCO’s recent discussion of whether AI investment is displacing other activity identifies a plausible mechanism: an investment boom can compete for resources and influence financing conditions elsewhere.

But demonstrating that mechanism in principle is different from measuring its actual contribution to weak housing or business activity. Tariffs, uncertainty, energy costs and sector-specific conditions also matter.

The proposition worth testing is that parts of the economy could bear substantial adjustment costs before the benefits of new productive capacity spread widely.

This is where the fiscal position becomes relevant.

The federal deficit was approximately $2 trillion in fiscal 2026, the year ending September 30. Treasury figures presented by the Joint Economic Committee put net interest spending at approximately $1.052 trillion against receipts of $5.403 trillion.

That means net interest absorbed roughly 19.5% of federal receipts.

Those are annual fiscal figures, not third-quarter GDP components. They describe the government’s room to respond if the expansion weakens.

Strong growth improves the capacity to support public debt. It does not eliminate the importance of the primary balance, refinancing costs and the maturity profile of outstanding obligations.

The Economist’s October analysis of bond markets usefully shifts attention towards this interaction between investment demand, borrowing costs and sovereign finances.

There is an important qualification. A fall in AI investment could reduce inflationary pressure and borrowing demand, helping government bonds. It could also weaken tax receipts and increase demands for fiscal support.

The net outcome would depend on the scale of the slowdown and the reaction of interest rates. Public-debt stress and private-investment stress should not simply be added together as though their effects were independent.

For the wider world, this American expansion can transmit benefits and costs simultaneously. Equipment suppliers can gain orders, while businesses dependent on dollar financing face difficult borrowing conditions.

A country can therefore benefit commercially from US investment without enjoying easier financial conditions.

For boards and lenders, that makes the financing assumptions behind growth plans unusually important. A project should be tested against lower utilisation, weaker pricing, higher refinancing costs and earlier equipment replacement.

Long-term purchase commitments, lease payments and guarantees deserve attention alongside conventional borrowing. The economic obligation matters even when its accounting label differs.

Three outcomes are plausible from here.

In the favourable case, AI adoption spreads across ordinary businesses. Better processes lift output per hour, reduce costs and support real incomes. The initial construction programme then develops into a broader improvement in productive capacity.

In a less comfortable case, investment remains vigorous but the gains spread slowly. Financing and resource pressures persist, households remain constrained, and weaker sectors continue to absorb much of the adjustment. GDP can keep growing under those conditions.

In the adverse case, revenue expectations fail to justify continued expansion. Investment plans are cut, suppliers lose orders and lenders reassess exposure. The damage depends on leverage, contractual commitments and who ultimately bears the losses.

These are scenarios, not forecasts. The present data do not settle which will prevail.

The most useful evidence over the coming months will be real household income, the breadth of employment gains, productivity beyond the technology sector, investment utilisation, cash generation after capital expenditure and the cost of refinancing.

Each tests a different part of the bridge between current spending and future benefits.

America has the capacity to undertake a remarkable technological investment cycle. Its third-quarter performance shows that capacity in action.

The harder economic achievement will be to make the resulting output valuable enough, widely enough and soon enough to justify the commitments being made today.

The bill for building the future arrives before the benefits. The strength of this expansion will depend on who can finance that interval.