AI Investment Boom: Economic Catalyst or Bubble in the Making?

The A.I. Spending Frenzy Is Propping Up the Real Economy, Too

The trillions of dollars that tech companies are pouring into new data centers are starting to show up in economic growth. For now, at least. It's no secret by now, as investors await an earnings report on Wednesday by the chip behemoth Nvidia, that optimism around the windfall that artificial intelligence may generate is pumping up the stock market. But in recent months, it has also become clear that A.I. spending is lifting the real economy, too. It's not because of how companies are using the technology, at least not yet. Rather, the sheer amount of investment -- in data centers, semiconductor factories and power supply -- needed to build the computing power that A.I. demands is creating enough business activity to brighten readings on the entire domestic economy. Companies will spend $375 billion globally in 2025 on A.I. infrastructure, the investment bank UBS estimates. That is projected to rise to $500 billion next year. Investment in software and computer equipment, not counting the data center buildings, accounted for a quarter of all economic growth this past year, data from the Commerce Department shows. (Even that probably doesn't reflect the whole picture. Government data collectors have long had trouble capturing the economic value of semiconductors and computer equipment that large tech companies like Meta and Alphabet install for their own use, rather than farming out to contractors, so the total impact is likely to be higher.) The big tech companies are the largest financiers of the frenzy, but private equity firms have been pouring in capital, too. Brookfield Asset Management, which manages a vast real estate portfolio, estimates that A.I. infrastructure will sop up $7 trillion over the next 10 years. The torrent of cash comes as the effects from Biden-era infrastructure subsidies fade, erratic tariffs freeze corporate decision making and high borrowing costs deter less lucrative real estate projects such as housing and warehouses. In 2025, spending on data center construction -- not including the cost of all the technology they house -- will exceed investment in traditional office buildings, according to the Dodge Construction Network. "The expectations of very high returns in this industry are trumping the high interest rates that we are facing today," said Eugenio Alemán, chief economist with the financial services company Raymond James. Companies are promising even more spending, but their ability to deliver, he noted, depends on whether their expectations are fulfilled. Most A.I. tools are not profitable currently, and they will have to generate huge cash flows over time for the tech companies to recoup their investments. "There is always a risk that very little of what they say is going to pan out," Dr. Alemán said. "So whenever they figure out that it is not what they thought, there is going to be a large correction." For now, everyone wants a piece of the spending. To understand the excitement around booming A.I. use, it's helpful to take a spin through corporate quarterly earnings calls. Publicly traded construction firms, electricity providers and electronics manufacturers are eagerly telling investors they can get a piece of the action: * Duos Technologies, which provides analytics and imaging for railroads and other infrastructure, has recently expanded into building small data centers. "Our business is commercially and financially in a great position to take advantage of the superhot demand coming from the data center computing gold rush," said Charles Ferry, the company's chief executive. * "These data center managers and big A.I. providers need energy and energy storage in an insatiable way," said Dennis Calvert, the chief executive of BioLargo, an environmental services company that sells a large-scale battery system. * "With data center growth and climate mandates accelerating demand for clean, reliable baseload power, the opportunity for advanced nuclear has never been stronger," said James Walker, the chief executive of NANO Nuclear Energy, which makes small reactors. Data centers are also attractive to traditional construction companies, which see the opportunity to shift from their typical real estate development projects into a new asset class with ample capital behind it. Skanska, a large contracting firm, forecasts that data center construction will average 13.2 percent annual growth through 2029, a far speedier rate than any other sector it tracks. The American Cement Association, an industry group, estimated that the sector would require a million metric tons of cement over the next few years. One of North America's largest building materials suppliers, Amrize, developed an "A.I. optimized" concrete mix with Meta that has lower carbon emissions. Amrize said data center construction was a "bright spot" in its otherwise soft second quarter. The boom has also been good for electricians, engineers and heavy-equipment operators. Although data centers that are up and running typically employ only a small number of people, the construction phase can put thousands to work. That's part of the reason that U.S. construction employment has remained steady even as housing, office and warehouse projects have dried up. How long can the spending last? The intensity of the A.I. investment wave has raised uncomfortable parallels to the last time the tech industry funneled billions of dollars into infrastructure to support a new technology with high expectations of future profits. In 2001, after the stock market crash brought on by the collapse of speculative dot-com companies, the telecommunications sector crumpled, too: Companies that had taken on debt to build out fiber-optic networks failed, creating an implosion that rippled through the global economy. Already, there are a few signs of caution. The chief executive of OpenAI, Sam Altman, raised eyebrows this month with remarks that the sector is "overexcited" and that some players will lose a lot of money. UBS, while generally positive on the industry, wrote in a note to clients that there could be some "indigestion" over the capital expenditures underway. At the moment, investors are reassuring themselves that a pullback would not be catastrophic. For one thing, data centers are financed by a diverse group of lenders, reducing the exposure of any one part of the banking system. Leases generally have long terms with hard-to-escape contracts, which could insulate landlords even if their deep-pocketed tenants had to walk away. For another, even if A.I. use doesn't live up to the hype, the internet is expanding quickly. The flood of data center capacity in the pipeline is still likely to be absorbed, even if more slowly than it is now. Vacancy in leased data centers -- that is, those that aren't owned by their users -- is currently close to zero. Future developments are usually spoken for ahead of time, according to JLL, the real estate professional services firm. "If we think about just general data creation and data storage, that has been growing at a rapid pace for decades, and that will continue to grow," said Andrew Batson, JLL's head of data center research for the Americas. "At some point there will be some natural slowdown in demand, but that's not in our near-term forecast." He expects the sector will keep growing about 20 percent annually through at least 2030. In the coming years, the most significant constraint on data center growth is more likely to be supply: The energy, water, workers and technical equipment required to construct and run them are all getting more expensive. At the same time, local communities, once eager to attract data centers, have occasionally soured on them. In the latest example, the City Council of St. Charles, Mo., placed a one-year moratorium on new facilities over concerns about drinking-water contamination. "There's a ton of money going into it, but at some point the cost is going to bite," said Eric Gaus, chief economist with Dodge Construction Network, which closely tracks new developments. "You've got the locals who are saying, 'If you're going to put something here, you need to do more than just build it and walk away.'" Cade Metz and Ben Casselman contributed reporting.

Is the tide turning on the AI boom's myth of destiny?

For the last year and a half, AI hasn't just been a technology -- it's been a worldview. Nvidia's stock tore through Wall Street expectations and crowned the company more valuable than Microsoft and Apple. Microsoft pledged to spend like a sovereign wealth fund to bulk up Azure. Google rewired its entire roadmap around Gemini. Meta, never shy about a grand narrative, promised that "superintelligence" was within reach -- and CEO Mark Zuckerberg spent like it. The numbers matched the rhetoric: a trillion of market cap here, a trillion there, tens of billions in quarterly capital expenditures. The refrain was simple and contagious: inevitability. But inevitability can have a short shelf life in the world of technology. Over the past month, three jolts in particular have rattled the story. OpenAI CEO Sam Altman, who has made a career out of selling a version of the future, said the quiet part out loud: Asked if he thinks we're in an AI bubble, he said, "Yes." Meta, after months of splashy AI hiring and rhetoric, has reportedly frozen recruitment and chopped up its "super" lab. And MIT published research that went viral on LinkedIn, estimating that 95% of enterprise AI pilots return no business value. That trifecta -- a prophet hedging, a zealot pausing, and academics bringing receipts -- turned inevitability into a question. Markets noticed. Nvidia, the totem of the AI boom, will be treated less like a stock and more like a stress test for the entire economy when it reports its quarterly earnings on Wednesday. Its earnings call isn't just another quarterly update -- it's the hinge on which the hype rests. Wall Street expects another record, roughly $46 billion in revenue. But at a $4 trillion valuation, "better than expected" may not be good enough. If the golden child stumbles, even slightly, the talk on whether the tide is turning on AI gets louder. Already, cracks have shown. After a mid-August surge, tech stocks, including Nvidia and other AI-heavy firms, pulled back by about 1.6%, even as energy and real estate rose. Analysts warn that the Nasdaq's 41% gain since April may have inflated valuations -- a frothiness that makes every earnings print feel like a cliffhanger. And that's why Nvidia's earnings call has been elevated into a kind of secular rite. Take Humane's AI Pin. The $700 wearable was hyped as the next iPhone -- an AI-native device to liberate us from screens. It lasted less than a year before its assets were offloaded to HP in what amounted to a mercy sale. Or, take Microsoft's Recall, the feature billed as a photographic memory for your PC. Privacy watchdogs called it a surveillance nightmare, and the company had to walk back its rollout plans. For an industry that loves to declare "this changes everything," the first wave of consumer products has changed very little, except investor patience. The corporate numbers don't look much better. MIT's study put hard math on what many CIOs already suspected: Nearly all of those shiny AI pilots amount to little more than slideware. That finding rippled through boardrooms and social media feeds because it finally gave executives cover to say what they'd been whispering: AI demos are impressive, but they're not showing up in the P&L. Even among developers, the ground feels shaky. Stack Overflow's 2025 survey found that while 84% of coders now use AI tools, only 3% say they "highly trust" the outputs. Adoption is skyrocketing, but confidence is collapsing. The result is a paradox: AI is everywhere, and yet no one quite depends on it. Meta's reported recent pivot has only added to the sense of recalibration. After a year of Zuckerberg touting "superintelligence" and stuffing its payroll with AI hires for mind-boggling sums, the company suddenly froze recruitment, restricted transfers, and broke its mega-lab into four groups. The official line was focus. Some analysts called it discipline. But to most people watching, it looked like fatigue. For an industry that treats "more" as a strategy, a pause from one of the biggest spenders was its own kind of confession. Wall Street hasn't pulled the plug. Wedbush analyst (and raging tech bull) Dan Ives insists this is still "the second inning" of an AI bull market. But the market's edginess shows up in the tape: Palantir shares plunged more than 9% in a single session amid bubble chatter, while Nvidia dropped about 3.5% the same week. And Erik Gordon, a University of Michigan professor known for his bubble calls, warned Business Insider that the AI bust could prove even uglier than the dot-com collapse -- pointing to CoreWeave's stunning 33% valuation plunge, a $24 billion wipeout in just 48 hours, as a canary in the coal mine. Why? Because the infrastructure race is the only part of the AI boom that still feels like a sure thing. Nvidia's GPUs are the scarcest resource in tech. CoreWeave, a cloud startup that barely existed three years ago, is now buying up data centers as if they're beachfront property. Analysts may debate the future of copilots and chatbots, but no one questions the future of compute. The line from Big Tech is consistent: The returns are there in cloud, ads, and developer services; the spending is the bottleneck. That's why the market can wobble on sentiment and still finance another data center. But there's also concentration risk baked in. Tech giants now make up roughly 40% of the S&P 500. If AI sentiment turns, it's not just a few stocks at stake -- the entire market could feel it. There's a macro-version of this paradox, too. A John Thornhill column in the Financial Times argues that we're in Carlota Perez's "installation" phase -- manic investment, messy results -- before a "golden age" can materialize. Deutsche Bank analysts have echoed the concern, warning that the AI buildout mirrors past bubbles from 18th-century canals to the 1990s dot-com and telecom frenzies: vast overbuilding justified by the promise of transformation, only to pop when belief thinned. The New Yorker made a similar case this week, saying we're in an AI profit drought: vast spending, thin P&L evidence, long J-curve. All three narratives map cleanly onto what operators are saying privately. But the physical costs are coming due. AI workloads demand so much power that Google signed a deal with the Tennessee Valley Authority and a nuclear startup just to keep its Southeast data centers running. Researchers have quantified the water consumption of model training, showing that every breakthrough comes with an invisible utility bill. The "cloud," it turns out, is built of concrete, copper, and cooling towers. Regulators have woken up, too. On Aug. 2, the EU's AI Act began applying obligations to general-purpose models: transparency about training data, mandatory risk assessments, and new safety disclosures. The stricter rules will come in 2026, but the first bite is already here. In the U.S., agencies are circling corporate filings, probing whether companies are "AI-washing" their earnings calls. Copyright fights rage in the courts -- The New York Times is suing OpenAI -- even as other media groups cut licensing deals. And then there's China. After Washington's on-again-off-again export bans, Beijing has reportedly discouraged domestic firms from buying Nvidia's China-compliant H20 chip. That's not trivial; Nvidia makes roughly a quarter of its revenue in China. That's the kind of geopolitical tremor that could ripple through Wednesday's earnings call. In sentiment, maybe. Altman's bubble line broke the spell of inevitability. MIT's study turned hype into numbers, and the numbers were ugly. Meta's reported reorganization suggests that even the loudest boosters know when to pause. Developers are adopting but not trusting. Artists are suing. Regulators are writing (and rewriting) the rules. The "AI will change everything" pitch no longer lands as gospel. But in capital, not yet. The hyperscalers are still writing historic checks. NVDA is still the most important ticker in the market. If Wednesday's earnings blow past expectations again, the spending spree will look vindicated. But if they disappoint -- even slightly -- the bubble chorus will likely grow louder. Silicon Valley has always run on myth as much as math. The myth of inevitability allowed companies to raise obscene sums, to spend like nation-states, to paper over the fact that 95% of AI pilots go nowhere. The myth was strong enough to make a $4 trillion company out of Nvidia, to have Microsoft reimagine itself as an infrastructure empire, to send Zuckerberg chasing "superintelligence" like it's a Marvel subplot. But myths don't last forever. Eventually, someone reads the balance sheet. A year ago, AI was destiny. Chatbots were oracles, GPUs were holy relics, and anyone questioning the frenzy was accused of missing the future. Now? Its prophet mutters "bubble," and its supposed killer apps are retreating under the weight of their own demos. The question isn't whether or not AI is important -- it obviously is. The question is whether or not importance is enough to sustain trillion-dollar valuations and multitrillion-dollar infrastructure bets. If inevitability is gone, then AI, like every industry before it, will have to survive the harder test: proving it.