There’s a paradox developing on Wall Street: the fog of uncertainty surrounding artificial intelligence’s ability to deliver sky-high profits is thickening, yet AI bulls and bears alike seem to be doubling down on their convictions.
The bigger the numbers involved — whether for spending, profits or returns — the more entrenched advocates on both sides of the argument become. And many of these AI numbers are astronomical.
Wall Street research highlights this growing entrenchment, and the latest Bank of America fund manager survey speaks to it. A record 82 per cent of respondents think the AI trade is the most crowded, but roughly half still say we’re not in a bubble.
This curious conundrum could set the stage for a more volatile second half of the year and a particularly bumpy few weeks as the second-quarter U.S. earnings season kicks off, with investors on both sides looking for evidence that they are correct.
The bulls’ case is relatively straightforward. Trillions of dollars of AI-related capex in the coming years will fuel growth, profit and productivity on a scale that justifies the stunning surge in stock prices and signals that this rally has legs. Until now, earnings for AI beneficiaries have exceeded all expectations, so the bar for even higher profits is being raised. And rightly so.
The counterargument is that the cost of the AI buildout has grown so high that companies simply can’t generate the lofty returns investors expect. Hyperscalers have burned through their cash piles and are now relying on external debt and equity financing to fund this mountainous spending.
Meanwhile, it’s becoming clear, bears argue, that the cost of compute is too high. Demand for the latest, most expensive AI models will fall or be diverted toward cheaper, open-source alternatives, most likely from China. Either way, the return on this unprecedented investment is likely to disappoint.
Bank of America strategists neatly captured the increasing polarization between the AI camps in a chart last week that showed a “generational transfer” of free cash flows from hyperscalers to chip companies. The vast sums hyperscalers are spending on AI infrastructure and capex are essentially flowing to semiconductor companies, which will enjoy an increasing share of future AI profits.
They note that the “Magnificent Seven” hyperscalers have spent $234 billion in capex this year but their stocks have barely risen, as investors anticipate that these firms’ free cash flow will turn negative for the first time in at least two decades.
Bears will point to those figures, while bulls may highlight the semiconductor side of the equation and argue that the transformational nature of AI will justify this short-term “wealth transfer.”
Both sides make compelling arguments based on the available information, so how can either camp increasingly be sure it is right? One word: uncertainty.
Although parallels can be drawn between AI and other transformative inventions like railroads, computers and the internet, investors have no real AI playbook. Nobody knows how it will ultimately impact businesses, the workplace, jobs and the economy. So even though there is less “evidence” to support either viewpoint, there is also less to refute either one, making it easier for everyone to convince themselves that they are right.
But with deeper entrenchment comes greater risk. The U.S. stock market is increasingly dependent on the bullish AI narrative, and so is the wider U.S. economy.
Compute capex in the first half of the year accounted for a larger share of GDP than at any point in history, and overall tech-related investment was up 30 per cent from a year ago, according to analysts at Carlyle. At the same time, all other capex fell, resulting in a record divergence. AI now accounts for almost all U.S. net investment.
Ultimately, investors will have to come to terms with the fact that the AI buildout might be a zero-sum game, meaning something has to give. Either hyperscalers’ free cash flow suddenly bounces back or chipmakers’ growth slows sharply. There’s certainly more room for chip stocks to extend last month’s retracement, with the “SOX” semiconductor index still up 75 per cent so far this year.
Zoom in, and there is already a lot of skittishness in the market. Volatility in the stocks of some major U.S. tech companies central to the AI story, like Intel, Qualcomm, and Oracle, has exploded to historic levels recently.
Noah Weisberger, head of equities at BCA Research, says clients are increasingly concerned with single-stock volatility, adding that last month’s AI wobble suggested investors had gotten “a little over their skis.”
If investors in U.S. tech want an idea of how volatile AI-related stocks can get, they should look at South Korea’s KOSPI, an index even more concentrated in AI names than the S&P 500 .SPX. The KOSPI’s three biggest drops since the Lehman Brothers bankruptcy in September 2008 have all come this year: a 12 per cent slump in March, a 10 per cent slide in June and a 9 per cent tumble on Monday driven by a steep fall in semiconductor darling SK Hynix.
Whether you view that as a buying opportunity or a sign of bigger declines ahead largely depends on which side of the AI divide you are on.