The US economy would be in recession-territory if it were not for the enormous bet that has been placed that AI really will deliver productivity growth in short order, a bet that looks perilous.
Michael Roberts is an Economist in the City of London and a prolific blogger.
Cross-posted from Michael Roberts’ blog
The US stock market continues to hit new record highs; the bitcoin price is also close to highs and the gold price has rocketed to all-time highs.
Investors in financial assets (banks, insurance companies, pension funds, hedge funds etc) are wildly optimistic and confident about financial markets. As the chair of Rockefeller International, Ruchir Sharma put it: “Despite mounting threats to the US economy — from high tariffs to collapsing immigration, eroding institutions, rising debt and sticky inflation — large companies and investors seem unfazed. They are increasingly confident that artificial intelligence is such a big force, it can counter all the challenges.” AI companies have accounted for 80 per cent of the gains in US stocks so far in 2025. That is helping to fund and drive US growth, as the AI-driven stock market draws in money from all over the world. Foreigners poured a record $290bn into US stocks in the second quarter of 2025 and now own about 30% of the market — the highest share in post-second world war history. As Sharma comments, the US has become “one big bet on AI”.
The AI investment ‘bubble’ (as measured as the stock price relative to the ‘book value’ of a company) is 17 times the size of the dot-com frenzy of 2000 — and four times the subprime mortgage bubble of 2007. The ratio of the US stock market’s value to GDP (aka the “Buffett Indicator”) has moved up to a new record high at 217%, more than 2 standard deviations above the long-term trendline.
And it is not just corporate shares that are booming. There is a huge demand to hold the debt of US corporations, particularly the large tech and AI companies as per the so-called Magnificent Seven. The ‘spread’ of interest paid on corporate bonds compared to ‘safe’ government bonds has fallen to under 1% pt.
These bets on the future success of AI cover all bases, or another way to put it: all the eggs are in one basket: AI. Investors are betting that AI will eventually deliver huge returns on their stock and debt purchases, when the productivity of labour rises dramatically and with it, the profitability of AI companies. Matt Eagan, portfolio manager at Loomis Sayles, said that sky-high asset prices suggested investors were banking on “productivity gains of the kind we have never seen before” from AI. “It is the number one thing that could go wrong”.
Up to now, there is little sign that AI investment is delivering faster productivity. But ironically, the huge investment in AI data centers and infrastructure is holding up the US economy in the meantime. Almost 40% of the US real GDP growth last quarter was driven by tech capex and the bulk of that capex was in AI-related investments.
AI infrastructure has risen by $400 billion since 2022. A notable chunk of this spending has been focused on information processing equipment, which spiked at a 39% annualized rate in the first half of 2025. Harvard economist Jason Furman commented that investment in information processing equipment & software is equivalent to only 4% of US GDP, but was responsible for 92% of GDP growth in the first half of 2025. If you exclude these categories, the US economy grew at only a 0.1% annual rate in the first half.
So without tech spending, the US would have been close to, or in, a recession this year.
What that shows is the other side of the story: namely the stagnation of the rest of the US economy. US manufacturing has been in recession for over two years (ie any score in graph below that is lower than 50).
and now there are signs that the larger services sector is also in trouble. The ISM Services PMI (an economic survey indicator) fell to 50 in September 2025 from 52 in August and well below forecasts of 51.7, signalling the services sector has stalled.
The US labour market is also looking weak. Employment grew at an annualised rate of just 0.5% in the three months to July, according to official data. That is well below the rates seen in 2024. “You’re in a low-hire, low-fire economy,” said Federal Reserve chair Jay Powell last month.
Young workers in the US are being disproportionately affected by the current economic downturn. US youth unemployment has risen from 6.6% to 10.5% since April 2023. Wage growth for young workers has declined sharply. Job vacancies for career starters have fallen by more than 30%. Early-career workers in AI-exposed occupations have experienced a 13% relative decline in employment.
The only Americans spending big money are the top 20% of earners. These households have done well, and those in the top 3.3% of the distribution have done even better. The rest are tightening their belts and not buying more.
Retail sales (after removing price inflation) have been flat for over four years.
The graph above shows that inflation has eaten into the spending power of most Americans. The average inflation rate remains stuck at about 3% a year on official figures, well above the target rate of 2% a year set by the Federal Reserve. And that average rate hides much of the real hit to living standards and real wage increases. Food and energy prices are rising much faster. Electricity now costs 40% more than it did five years ago.
Indeed, electricity prices are being driven up even more by AI data centres. OpenAI uses as much electricity as New York City and San Diego combined, at the peak of the intense 2024 heat wave. Or as much as the total electricity demand of Switzerland and Portugal combined. That’s the electricity of roughly 20 million people. Google recently cancelled a planned $1bn data center in Indiana after residents protested that the data center would “jack up electricity prices” and “suck away untold gallons of water in an area already plagued with drought”.
And then there is the impact of Trump’s tariff tax on goods imports into the US. Despite denials by the Trump administration, import prices are rising and beginning to feed through to goods prices inside the US (and not just in energy and food).
So far, foreign companies, in aggregate, are not absorbing the costs of the tariffs. During the 2018 trade war, import prices were mainly discounted by foreign companies. This time import prices have not declined. American importers rather than foreigner exporters are paying the tariffs, with more pass-through likely ahead for consumers. As Fed chair put it, “the tariffs are mostly being paid by the companies that sit between the exporter and the consumer… All of those companies and entities in the middle will tell you that they have every intention of passing that through [to the consumer] in time.”
Importers, wholesalers, and retailers are paying higher costs upfront and hoping they can eventually raise prices enough to shift the burden. The problem is that consumers are already tapped out. Household budgets are under pressure from rising debt, delinquencies, and wages that do not stretch far enough. Trying to pass along tariff costs in this environment would push demand even lower.
Businesses know this, which is why many of them are absorbing the costs instead. But when they do that, their margins shrink, and it becomes harder to sustain operations without making cuts elsewhere. When profitability gets pressured, management has few options. They cannot control tariffs and they cannot force consumers to spend more. What they can control are expenses. That begins with slowing hiring and scaling back growth plans, then cutting hours and overtime. If tariffs remain in place and consumers stay weak, the ripple effects spread further into the labour market.
Then there is government spending. The current closedown of government departments being imposed by Congress has given the Trump administration a further opportunity to slash federal government employment in a vain attempt to reduce the budget deficit and rising government debt. It’s a vain attempt because Trump’s claim that increased tariff revenues will do the trick is not credible. Tariff revenues since January 2025 are still only 2.4% of the projected total federal revenue in fiscal year 2025 of $5.2 trillion.
And as for the claim that tariffs would eventually fix the US trade deficit with the rest of world, that too has proved nonsense, so far. In the first seven months of 2024, the deficit was $500 billion deficit; the first seven months of 2025, it was $654 billion, up 31% yoy to a record high.
Contrary to Trump’s claims, the tariff hikes on imports will do little to ‘Make America Great Again’ in manufacturing. Robert Lawrence of Harvard’s Kennedy School reckons that “closing the trade deficit would barely raise the share of US manufacturing employment”. The net value-added in the trade deficit in manufactured goods in 2024 was 21.5% of US output. This would be the increase in US value if the trade deficit were eliminated. How much employment would this produce? It would amount to 2.8m jobs, which would be a rise of only 1.7 percentage points in the share of manufacturing in US employment, to 9.7% of overall jobs. But the share of production workers in US manufacturing in this case is just 4.7%, the other 5 percentage points consisting of managers, accountants, engineers, drivers, sales people and so forth. The rise in employment of production workers would be just 1.3m, or just 0.9% of US employment.
The US economy is not yet on its knees and in a recession as business investment is still rising, if slowing in growth.
Corporate profits are still growing. Operating income for S&P 500 companies (excluding financials) grew at 9% in the most recent quarter, compared with the year before. Revenues rose 7% (before inflation). But that’s just for the top companies led by the Magnificent Seven. Overall, the US non-financial corporate sector is beginning to see profit growth disappear.
And the Fed is set to cut its policy interest rate some more over the next six months, reducing the cost of borrowing for those who want to speculate in those fictitious financial assets). So a recession has not emerged yet. But increasingly, everything depends on the AI boom delivering on productivity and profitability. If the returns on massive AI investments turn out to be low, that could cause a serious stock market correction.
It is true that the big tech companies have mostly financed their AI investments out of free cash flow. But the huge cash reserves of the Magnificent Seven are being drained and AI companies are increasingly turning to equity and debt issuance.
The AI companies are now signing contracts with each other to build revenues. This is a form of financial musical chairs. OpenAI has signed about $1tn in deals this year for computing power to run its artificial intelligence models, commitments that dwarf its revenue. OpenAI is burning through cash on infrastructure, chips and talent, with nowhere near the capital required to fund these grand plans. So to fund its expansion, OpenAI has raised huge amounts of equity and started to tap debt markets. It secured $4bn in bank debt last year and has raised about $47bn from venture capital deals in the past 12 months — though a significant chunk of that is contingent on Microsoft, its biggest backer staying in the mix. The credit monitor Moody’s has flagged that much of Oracle’s future data centre sales relies on OpenAI and its unproven path to profitability.
Much now depends on revenues for the likes of OpenAI rising sufficiently to start to cover the exponential rise in costs. Goldman Sachs economists claim that AI is already boosting the US economy by about $160 billion, or 0.7% of U.S. GDP in the four years since 2022, which translates to roughly 0.3 percentage points of annualized growth. But this is more a statistical trick than real growth in productivity from AI so far and there is little revenue boost for the AI sector. Indeed, returns from further AI development may be diminishing. The cost of launching ChatGPT-3 was $50 million, launching ChatGPT-4 cost $500 million, while the latest ChatGPT-5 cost $5 billion, and according to most users, wasn’t noticeably better than the last version. Meanwhile, the likes of China’s Deepseek and other much cheaper competitors are undermining potential revenues.
So a financial bust is on the cards. But when financial investment bubbles burst, the new technology does not disappear. Instead, it can be acquired at low prices by new players in what Austrian economist Joseph Schumpeter called ‘creative destruction. By the way, this is exactly the argument of this year’s winners of the so-called Nobel prize in economics, Philippe Aghion and Peter Howitt. Booms and slumps are inevitable but necessary to drive innovation.
So AI technology could eventually deliver higher productivity growth if it manages to shed human labour sufficiently. But that may materialise only after a financial crash and consequent slump in the US economy. And if the AI-driven US economy dives, so will the rest of major economies. Time is not on the side of the Magnificent Seven. Indeed, adoption of AI technology by companies remains low and is even falling among larger companies.
Meanwhile the spending on AI capacity goes on mounting; and investors keep piling cash into buying AI company stock and debt. It’s one big bet on AI for the US economy.
Be the first to comment