Photograph: Getty Images
Investors have spent much of the past three years or so acting as if they have discovered fire. The American economy has avoided a widely predicted recession in part because it is being propped up by a private-sector stimulus programme of staggering proportions. This is the era of artificial intelligence, where capital spending on data centres, chips and power infrastructure has become a main engine of growth. By some early estimates, these investments accounted for nearly half of the expansion in American gross domestic product in 2025. This surge in spending helps explain why corporate earnings have stayed resilient even as interest rates stayed higher for longer than many had anticipated.
The gravitational pull of this trend is most visible in the stock market. A mere seven companies—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla—now account for roughly one-third of the value of the S&P 500. Since 2023 more than half of the gains in the American equity market have been generated by that narrow group. Nvidia, which makes the essential chips powering the AI revolution, has become the leader of the boom. Unlike the speculative dot-com startups of the 1990s, which soared on hype and collapsed, Nvidia consistently turns a big profit and dominates its market.
This concentration of wealth and power is being reinforced by a shift in the political weather. The Trump administration has made AI dominance a cornerstone of its economic and geopolitical strategy. Regulators have been instructed to clear the path for deployment, sending a clear signal to the markets that antitrust concerns or environmental hurdles for data centres will not be allowed to slow the pace of progress. For investors this creates a sense of “permissionless growth”, where the main constraint is the physical limits of the electrical grid rather than regulation.
Yet the sheer scale of the commitment is beginning to make even the most optimistic observers twitchy. Big Tech firms have pledged hundreds of billions of dollars to data-centre construction. Oracle recently committed roughly $250bn to its infrastructure build-out, a figure that briefly rattled the markets. The danger here isn’t that these companies will go bankrupt—they have the finances of small countries—but that they could waste huge amounts of money on the wrong projects. If the demand for AI services does not eventually match the cost of the hardware needed to run them, the resulting write-downs will be painful.
The paradox of the current moment is best embodied by OpenAI, which released ChatGPT in 2022. It boasts hundreds of millions of users and revenue growth that would be the envy of any company in history. Its private valuation has soared towards $500bn. However, its infrastructure plans run into the trillions, and its path to steady profits is blocked by the enormous cost of computing power. Investors are currently pricing the company as a future monopoly and a driver of global productivity all at once. It is a heavy burden for any single firm to carry.
While the money flows freely, the evidence of a productivity miracle in the broader economy remains elusive. Recent studies from researchers at the Massachusetts Institute of Technology suggest that a vast majority of AI deployments have so far failed to meaningfully boost bottom-line profits. AI models can write elegant code or draft emails with startling speed, but they still suffer from a “reliability gap”. In high-stakes business environments, the risk of “hallucinations” or errors means that human experts must spend as much time auditing the AI as they would have spent doing the work themselves. Gartner, a consultancy, suggests that the technology has entered the “trough of disillusionment”.
Historians of technology would argue that this lag is entirely normal. They point to the “productivity J-curve”, a phenomenon where transformative technologies initially lower productivity because they require companies to completely redesign their workflows and retrain their staff. It took thirty years for the electric motor to really transform the factory floor as managers had to learn that they no longer needed to cluster machines around a single steam engine. The AI optimists believe this transition will happen much faster, but for now the data is not on their side.
The binding constraint may soon prove to be physical rather than intellectual. The amount of electricity needed to power American data centres is projected to more than double by 2030. This creates a new set of tensions. Soaring power prices could trigger a backlash from consumers, and the political consensus for deregulation may fray if the local utility bill starts to climb. In this sense, energy, rather than software, may become the true limit on the speed of the intelligence revolution.
There is also a darker possibility: a slow-motion bubble. If companies believe that AI is making them more productive, they may begin to cut staff or suppress hiring before the technology is actually ready to fill the gap. This would create a toxic combination of higher unemployment and stagnant productivity even as capital spending continues to climb. Because AI investment is now as significant to the macroeconomy as telecommunications spending was at the peak of the dot-com boom, a sudden collapse in confidence would be far more damaging than the crash of 2000.
For now the rally seems likely to persist into 2026. Demand for chips still outstrips supply and the cash-rich “hyperscalers” can afford to keep spending. There is also no obvious alternative engine for global growth, making AI the only game in town for those seeking high returns. Most investors would rather risk being wrong in a crowd than being left behind by what might yet be the biggest inflection point since the industrial revolution.
Whether the current prices reflect a beginning or an end depends on the speed with which silicon can be turned into tangible economic output. The technology is real and its potential is vast, but the economics is still a work in progress. The most important lesson may be that a transformation can be both genuine and overhyped at the same time. Investment bankers would be well advised to keep their eyes on the power grid: it may tell them more about the future of the S&P 500 than any earnings report from Silicon Valley. ■
