The American economy is running on two engines. One is enormous. The other is sputtering.

THE AMERICAN economy's second-quarter growth figure was 1.5%, an annualised rate that sounded alarming until the details were examined. Domestic final sales — the part of GDP that excludes imports and inventories, and therefore best reflects home-grown demand — grew by 3.9%. The headline was dragged down by a surge in imports, a fall in government spending and an inventory correction. The real story is not that demand has collapsed. It is that it has split.

One part of the economy is booming. The other is running out of runway. What happens to the gap between them will determine whether the third quarter brings relief or reveals a deeper vulnerability.

The AI engine

In the second quarter of 2026, private business investment in AI-related categories reached $1.5 trillion on an annualised basis, up $300 billion, or 25%, from a year earlier, according to the Kobeissi Letter. Over the preceding two years AI-related spending has risen by $500 billion, or 50%. That is not a side show. Direct AI investment accounted for between a quarter and a third of recent GDP growth, according to the same analysis.

The Federal Reserve's own researchers, writing in a July Feds Note, put the economy in a "buildout phase" rather than a productivity-payoff one. Investment in data centres, semiconductor equipment and software is "showing no signs of slowing", the FOMC minutes from June noted. Equipment spending alone grew by 15.2% in the second quarter, the fastest pace in categories where AI hardware dominates.

To be sure, the scale of this buildout invites scepticism. Not every dollar of capital spending becomes tomorrow's productive capacity; some becomes excess. The Fed's note observes that, as of mid-2026, there is no large aggregate productivity signal. Micro-level experiments show gains, but they have not yet added up to economy-wide improvement. Historically, the productivity benefits of a new general-purpose technology lag its investment by years. The Luddites were eventually proved wrong about the direction of history, but right about the timing of the pain.

Yet even if the AI boom turns out to be a bust in the long run, its short-run contribution to demand is real. Factories are being built. Servers are being ordered. Engineers are being hired. Those are all things that show up in GDP. The risk is not that AI investment is an illusion. It is that it is highly concentrated: a handful of hyperscalers, a few chipmakers and a cluster of data-centre operators account for the lion's share.

The consumer engine

On the other side of the ledger, the consumer — which accounts for more than two-thirds of American economic activity — is showing cracks. Personal savings have fallen to 2.6%, matching the low seen during the 2022 energy shock, according to TD Economics and Bank of America's Consumer Checkpoint. That is a two-handle savings rate, something not seen since early 2008, when households were sitting on record levels of leverage.

Consumer spending itself is not collapsing. In the second quarter it rose by 3.2%, an acceleration from the first quarter's meagre 0.5% gain. Personal spending in June added a further $65.2 billion, according to BEA data. But the margin for error is vanishing. The larger tax refunds that provided a near-term cushion are estimated to be exhausted by households by September, TD Economics warns. Real disposable income grew by only 0.3% in June.

The savings rate matters because it measures the difference between income and spending. When it reaches historic lows, it means households are financing current consumption by running down buffers. That is sustainable only as long as income keeps rising and shocks do not arrive. It is no accident that the labour market, while still decent, is cooling: the unemployment rate has drifted to between 4.2% and 4.5%, and forecasters surveyed by the Philadelphia Fed expect monthly job gains to slow to an annual average of just 34,600 for 2026, down sharply from their earlier estimate of 48,500.

The incentive structure is clear. Households want to maintain their standard of living. With inflation still above the Fed's 2% target and wages not keeping pace across all segments, the choice is to spend savings or cut back. Many are choosing to spend. That keeps GDP figures respectable for now but leaves little cushion for what comes next.

The tariff tax

Overlaying these two stories is a policy shock that is doing double duty: raising costs and distorting trade flows. The Congressional Budget Office projects that customs duties will rise from 0.6% of GDP in 2025 to 1.3% in 2026, as the Trump administration's expanding tariff schedule takes effect. J.P. Morgan estimates the average effective tariff rate has settled in the mid-to-high teens, with further sectoral tariffs on Canada, Brazil and Vietnam coming into force in August.

Tariffs promise dignity to workers and deliver invoices to consumers. Their political appeal is obvious: they make protection visible and costs diffuse. The trouble is that they rarely rebuild the industries they claim to save. The Tax Foundation and the Yale Budget Lab estimate that current policies reduce the level of long-run output by 0.4% to 0.6%. In the near term, J.P. Morgan calculates that tariffs add 0.2 to 0.4 percentage points to the PCE price level, with near-complete pass-through to domestic buyers.

The trade data confirm the drag. The trade deficit surged to $103.65 billion in May, its second-highest level since the so-called "Liberation Day" tariffs were first announced, according to Forbes. Imports of capital goods hit a record, partly driven by AI-related equipment purchases, but the overall effect on GDP arithmetic is negative: imports are subtracted from GDP. In the second quarter, international trade shaved a full percentage point off growth. Structures spending, which includes buildings and infrastructure, fell for the tenth consecutive quarter.

The Fed's impossible arithmetic

This is where the Federal Reserve finds itself. Under newly appointed chairman Kevin Warsh, the FOMC kept the federal funds rate at 3.5% to 3.75% at its July meeting, removing the previous "easing bias" from its statement. Core PCE inflation stood at 3.4% in the second quarter, with headline PCE running at 3.8% in April and 4.1% in May, according to the June FOMC minutes. Inflation remains well above the 2% objective.

But the labour market is weakening and growth has slowed to 1.5%. A rate hike would punish consumers who are already running down savings. A rate cut would signal surrender to inflation that is partly tariff-driven and partly energy-driven. The FOMC is stuck between a growth slowdown that is real and a price problem that is also real.

The June dot plot reveals the division. Some participants expect the appropriate rate to be within or slightly below the current range by year-end. Many others think it should be above. The median projected rate for end-2026 is 3.8%, higher than the 3.4% projected in March, as tariff inflation and AI-driven cost pressures have shifted the committee's outlook.

Prediction markets priced in over three-quarters probability of a hold at the July meeting, with virtually no chance of a cut and roughly a quarter odds of a hike. Mr Warsh, a Trump appointee, is not pushing for the latter. The political incentive is clear: a rate hike during a slowdown would provoke the White House. The institutional incentive is also clear: the Fed wants to see whether tariff-driven inflation is transitory before tightening further.

The forecast

Professional forecasters surveyed by the Philadelphia Fed in May predict real GDP will grow at an annual rate of 2.2% in the third quarter, a downward revision from their earlier estimate of 2.3%. The FOMC's own median projection for full-year 2026 is also 2.2%, but that is a quarterly-to-quarterly annual average, not a reading of the second half in isolation. Polymarket traders, by contrast, are more optimistic: they assign a 44% probability to Q3 growth of 3% or more and a 76% chance of at least 2.5%.

The divergence between the forecasters and the prediction market is itself telling. The former are worrying about the consumer, tariffs and the fading of the tax-refund buffer. The latter are betting on AI investment momentum to carry the quarter. Both could be right about their respective drivers. The question is which force will dominate when the numbers arrive on October 29th, when the BEA releases its advance Q3 estimate.

The stronger case is that the third quarter will show a modest rebound from the 1.5% low of Q2, but that the rebound will be narrow and fragile. AI capex will keep pushing equipment and software spending higher. Consumer spending will hold up as long as the labour market continues to add jobs, even slowly. But the personal savings rate is at levels that have not been seen since the last recession. Tariffs are raising costs without rebuilding supply chains. The Fed is constrained on both sides.

The broader lesson is not that the American economy is on the brink of recession. It is that it is growing on a foundation that more resembles a scaffolding than a structure. When the AI buildout slows, when households exhaust their surplus, or when tariff costs bite deeper into real incomes, the 1.5% headline of the second quarter will look less like a statistical artefact and more like a warning.

The better answer for policymakers is not to pick between inflation and growth. It is to reduce the tariff overhang, which is doing both harm at once, and to give the Fed room to respond to whichever problem proves more persistent. Keeping rates steady while tariffs push up prices and suppress investment is the worst of both worlds. Consumers pay first. The rest of the economy follows.