AI Data Centers Fuel U.S. Growth Amid Bubble Concerns

Is America’s Economic Growth Now Driven by AI Data Centers? Recently, Harvard economist Jason Furman sparked widespread attention across Wall Street and Silicon Valley with a set of data posted on X (formerly Twitter). Furman, who served as chairman of the White House Council of Economic Advisers under President Obama, pointed out that nearly all of the U.S. economic growth in the first half of 2025 stemmed from investments in artificial intelligence infrastructure. If investments in information processing equipment and software were excluded, the real GDP growth rate would have been just 0.1%. In other words, without AI-related capital spending, traditional economic sectors—manufacturing, retail, services, real estate—showed virtually no growth during this period. The Financial Times’ Unhedged column, led by reporter Robert Armstrong, described the phenomenon aptly as “Silicon Mountain”—a towering structure now bearing the weight of the nation’s economic expansion. But how solid is this foundation? A $40 Billion Bet To grasp the scale of the shift, consider the numbers. According to Morgan Stanley Wealth Management’s chief investment officer Lisa Shalett, the annual capital expenditure by hyperscale cloud companies on data centers and related projects is approaching $400 billion—four times what it was just a few years ago. These companies—Microsoft, Google, Amazon, Meta, and NVIDIA—now account for nearly one-third of all U.S. corporate capital spending. $400 billion is a staggering sum—roughly equivalent to Denmark’s entire annual GDP, or more than double the total EU budget for 2024. And this capital is overwhelmingly funneled into one goal: building the computational infrastructure needed to run next-generation AI models. A report from Renaissance Macro Research in August 2025 revealed a historic milestone: AI data center construction contributed more to U.S. GDP growth than consumer spending did in the first half of the year. For context, consumer spending typically accounts for two-thirds of U.S. GDP and has long been the primary engine of growth. Yet in 2025, a category that was nearly negligible in economic statistics just five years ago now outpaced the total spending of American households on shopping, travel, dining, and entertainment. The speed of this transformation is remarkable. From 2015 to 2020, tech investment hovered steadily between 3.7% and 4% of GDP. But starting in 2023, the curve began a steep upward climb. By the first half of 2025, the share had surpassed 4.5%, with AI hyperscale capital spending alone reaching 1.2% of nominal GDP—up from near zero just a few years prior. Where Is the Money Going? Tracking the actual flow of funds reveals a complex web of commitments. This year, we’ve seen record-breaking deals: NVIDIA pledged up to $100 billion in investments to OpenAI, OpenAI committed $300 billion to Oracle for computing capacity, and Oracle announced a $40 billion chip purchase from NVIDIA. These agreements are not simple transactions. NVIDIA’s $100 billion commitment, for instance, is staged: $10 billion upfront to build 1 gigawatt of data center capacity, with the full amount only achievable if the project expands to 10 gigawatts. Oracle’s $300 billion deal with OpenAI won’t begin until 2027 and will be spread over five years. Yet markets reacted instantly. Oracle’s stock surged after the announcement, despite weak quarterly results. Investors weren’t focused on current revenue or profits—they were betting on “remaining performance obligations,” the future income locked in by signed contracts. Oracle’s backlog grew by 359% year-on-year, with expected future revenue reaching $455 billion—most from OpenAI. Similar dynamics occurred with AMD. After announcing a chip supply deal with OpenAI, its stock jumped 35%. As part of the agreement, OpenAI gained the right to purchase up to 10% of AMD’s shares. The timing was strategic: OpenAI acquired the stake before the news, then the announcement boosted the stock, increasing the value of its holdings. Together, these deals have created a vast network of commitments. OpenAI’s total AI infrastructure and computing agreements now exceed $1 trillion. For a private company with annual revenue of around $10 billion and a valuation of $500 billion, this is wildly disproportionate. But these promises depend on conditions: OpenAI must grow its revenue to pay for these costs; Oracle must secure enough power—estimated at 4.5 gigawatts, equivalent to two Hoover Dams—to operate its data centers. And all of this must eventually deliver real commercial returns. Skepticism and Warnings David Einhorn, founder of Greenlight Capital, has been vocal in his skepticism. In a late September discussion, he called the scale of AI infrastructure spending “extreme to the point of incomprehensibility,” warning of massive capital losses. Einhorn doesn’t deny AI’s long-term potential—he believes it will surpass even the most optimistic forecasts. But he questions whether such massive investments will yield returns. “Will spending $1 trillion or $500 billion produce good results?” he asked. The core issue is timing. AI’s transformative impact may take a decade or more to materialize, but financial markets demand returns in the short term. And the data isn’t promising. A widely cited MIT survey found that 95% of companies using generative AI tools saw zero return on investment. Despite training employees and reengineering workflows, AI failed to boost productivity or revenue. Physical constraints are also real. OpenAI’s expansion would require power equivalent to New York City and San Diego combined—power that doesn’t yet exist. Building new power plants takes time, involves environmental reviews, and requires grid upgrades. Meanwhile, data center construction is outpacing energy supply. Highs and Lows of the System Peter Atwater of William & Mary sees this as a systemic risk. He compares today’s AI investment boom to the pre-2008 housing market, where capital flowed through a “conveyor belt” of interdependent players. If OpenAI fails to meet its obligations to Oracle, Oracle’s purchases from NVIDIA suffer, which affects NVIDIA’s ability to fund other AI ventures. One broken link can trigger a chain reaction. “At the peak of the housing bubble, the most fragile commitments were the first to be abandoned,” Atwater noted. “In the AI era, promises are the first thing that vanish.” The gap between input and output is widening. In the past two years, Microsoft, Meta, Tesla, Amazon, and Google invested about $560 billion in AI infrastructure, generating only $350 billion in related revenue—a 16:1 input-to-output ratio. If this were a startup, no venture capitalist would continue funding it. But because these are the world’s largest, wealthiest companies, markets assume they know what they’re doing. The Financial Amplifier The real danger lies not just in tech spending, but in how the financial system amplifies it. As The Information noted in mid-October, if there’s an AI bubble, Wall Street is at least half responsible. Highs in the market are soaring: Goldman Sachs reported its best year ever in investment banking and markets. BlackRock’s assets under management grew nearly $1 trillion in one quarter. Part of this is due to AI stocks dominating major indices—BlackRock manages index funds that automatically overweight companies like NVIDIA and Microsoft, pushing their prices higher and inflating fund sizes. Meanwhile, the cost of capital has dropped. The spread between government and corporate bonds is at its narrowest in years, meaning investors accept minimal risk premiums. Even high-risk AI data center firms like CoreWeave can issue bonds at 8.5% and find eager buyers. Private debt funds now manage $3 trillion—up 50% in five years. Their mission is to find higher returns than traditional bonds. With that pressure, lending standards are loosening. Business models, cash flow, and profitability matter less—just label it “AI,” and capital flows in. Private equity holds $2.5 trillion in dry powder—uninvested capital that must be deployed or risk investor dissatisfaction. This fuels record-breaking deals: Electronic Arts was acquired for $55 billion by a consortium including PIF and Silver Lake, setting a new record. Historically, such leveraged buyouts occur near market peaks. Index funds now control over half of all fund assets. They don’t pick stocks—they follow market weights. As AI giants grow in index weight, every dollar flowing into these funds automatically boosts them, creating a self-reinforcing cycle. AI companies announce massive investments → stock prices rise → valuations increase → funding capacity grows → larger investments announced. The system thrives on confidence. But if real results fall short—or if markets suddenly turn cautious—this cycle could collapse. Different Perspectives Not all experts are pessimistic. TS Lombard’s chief economist Dario Perkins argues that AI capital spending isn’t preventing a recession. He points out that the labor market in 2025 showed no signs of a downturn—low unemployment, no mass layoffs. The growth isn’t due to AI investment, he says. He also notes that many data center components are imported, meaning GDP accounting sees offsetting negative entries. So the net impact of AI investment may be smaller than the headline numbers suggest. Perkins also distinguishes between leveraged bubbles (like 2008) and today’s AI spending, which is largely funded by free cash flow, not debt. “There’s no massive leverage in this AI investment bubble,” he wrote. But he warns: “If it grows unchecked, it could become dangerous.” RBC Capital Markets’ Rishi Jaluria defends the spending from a technological perspective. If these investments unlock a world with fewer capacity constraints, they could accelerate AI development and enable applications currently limited by hardware. “If this delivers real ROI—cost savings, new revenue—then it creates net GDP benefit,” he said. The key word: “if.” History offers a cautionary tale. The dot-com bubble burst in 2000, destroying trillions in value. But it left behind a foundation—fiber optics, cloud infrastructure, software frameworks—that powered the next two decades of digital growth. Even if today’s AI investment proves excessive, the infrastructure built could still have lasting value. The Big Question Apollo Global Management’s chief economist Torsten Sløk admitted in early October that economists had predicted a slowdown for nine consecutive months—but the economy kept growing. He urged his peers to “look in the mirror.” Furman’s data—0.1% growth without tech investment—is a statistical fact, but its meaning remains contested. Is it a sign of a fragile economy masked by AI spending? Or is AI infrastructure itself a legitimate part of real economic activity? Excluding tech investment is like ignoring railroads in the 19th century or automobiles after WWII. These were real drivers of growth. The real issue isn’t the scale of investment, but its quality and sustainability. If it leads to productivity gains, new services, and consumer demand, it’s a sign of progress. If it results in idle data centers and excess capacity, it’s a massive misallocation of capital. As Morgan Stanley’s Michael Gapen noted, the puzzle is strong spending but weak job growth. His explanation: companies absorbed tariff costs, lowering unit labor costs and profits instead of raising prices or cutting jobs. But this isn’t sustainable. Atwater warns: “Today’s AI players act as if they have forever to figure out monetization. But markets don’t wait forever. When confidence drops, they’ll demand real results in a short time.” The system is tightly coupled. AI, credit markets, and stock markets are now one beast. If the foundation cracks, the entire structure could fall. In the end, Furman’s data stands: without AI data center investment, U.S. growth in 2025 was nearly flat. The question remains: is this the dawn of a new economic era—or a house of cards built on confidence? When the tide recedes, we’ll see who’s truly swimming.