Data centre investment drives 92% of American growth as other sectors flatline
Unprecedented economic concentration raises questions about sustainability as AI infrastructure buildout masks broader stagnation
America's economy appears remarkably robust in 2025—at least on paper. GDP growth figures suggest continued expansion, unemployment remains manageable, and stock markets hover near record highs. Yet beneath these reassuring headlines lurks an uncomfortable reality that economist Jason Furman recently quantified with striking precision: virtually all American economic growth in the first half of 2025 came from a single source.
Investment in data centres and information processing equipment accounted for 92% of GDP growth during that period, according to Furman's analysis. Strip out this technology infrastructure spending, and the American economy barely expanded at all—managing anaemic growth of just 0.1% on an annualised basis. For context, information processing investment represents only 4% of the overall economy, yet somehow generated nearly all the growth.
This represents an extraordinary concentration of economic activity in a single sector. Manufacturing contributed little. Retail and services sectors added almost nothing. Real estate investment remained tepid. The entire edifice of American economic expansion rested on one phenomenon: technology giants pouring hundreds of billions of dollars into AI infrastructure.
The monoculture risk
Economic concentration of this magnitude creates profound vulnerability. When 92% of growth originates from 4% of the economy, you've created what might be called an economic monoculture—a system lacking the resilience that comes from diversification.
The comparison to agricultural monocultures proves instructive. Farmers who plant single crops across vast acreages achieve impressive short-term yields but face catastrophic risk if disease or climate shocks strike. The genetic uniformity that enables mechanisation and efficiency simultaneously creates existential fragility. One successful pathogen can devastate entire regions.
American economic growth now exhibits similar characteristics. The technology sector's investment frenzy generates impressive GDP figures, yet dependence on this singular engine leaves the broader economy extraordinarily exposed. Should AI investment slow for any reason—regulatory intervention, technological disappointment, or simple financial exhaustion—there's virtually no other growth source to cushion the impact.
Furman acknowledges this reality whilst attempting optimism. He suggests that without the AI boom, lower interest rates and electricity prices might have stimulated growth elsewhere, potentially offsetting "about half" of what AI investment provided. Yet this admission actually strengthens the concern: even in the most generous interpretation, removing AI infrastructure spending would leave America with growth barely reaching 1%—dangerously close to recessionary territory.
Torsten Sløk, chief economist at Apollo Global Management, has observed the puzzling disconnect between economic forecasts and reality. For nine consecutive months through October 2025, economists predicted slowdown. Each month proved them wrong. The mystery resolves once you understand the composition: technology infrastructure investment overwhelmed every other trend. What appeared to be broad-based economic strength was actually narrow-based technological construction.
The circular money machine
The sustainability of this growth engine faces serious questions, particularly given the unusual financial structures supporting it. In September 2025, Nvidia announced plans to invest up to $100 billion in OpenAI to support massive data centre buildouts. OpenAI would then use these funds to purchase Nvidia chips and systems—creating what analysts immediately recognised as circular financing.
The pattern extends beyond this single deal. Nvidia holds a 7% stake in CoreWeave, valued at approximately $3 billion. CoreWeave has spent roughly $7.5 billion purchasing Nvidia GPUs. Nvidia has also committed $6.3 billion to purchase cloud capacity from CoreWeave that it cannot sell elsewhere, whilst simultaneously spending $1.3 billion over four years renting CoreWeave's services.
Similar arrangements exist with Lambda and other "neo-cloud" providers. In each case, Nvidia invests capital in companies that then spend substantial portions of those funds buying Nvidia products. The money flows in circles, inflating revenues and valuations for all participants whilst creating limited external economic value.
Financial analysts quickly drew comparisons to the late 1990s telecommunications bubble. During that era, equipment maker Lucent Technologies provided vendor financing to telecom companies so they could purchase Lucent's products. Revenue soared—Lucent peaked at $37.92 billion in 1999—until the bubble burst. By 2002, revenue had crashed 69% to just $11.80 billion. The company never recovered, eventually merging with Alcatel.
Jay Goldberg, an analyst with Seaport Global Securities, describes current AI deals as having "a whiff of circular financing" emblematic of "bubble-like behaviour". The concern isn't merely about whether AI technology has value—it clearly does—but whether current investment levels and valuations reflect genuine external demand or simply money chasing itself in ever-tightening circles.
The benefits extend beyond simple financing. Nvidia's equity stakes enable companies like OpenAI and CoreWeave to access debt markets at significantly lower rates than they could independently. Whilst startups previously borrowed at rates as high as 15%, Nvidia's backing allows them to secure funding at 6% to 9%—rates typically reserved for established giants like Microsoft or Google. This credit enhancement amplifies the circular nature of the arrangements, as cheap debt enables more chip purchases, which generates more revenue, which justifies higher valuations, which enables more investment.
History's uncomfortable precedents
The pattern has historical echoes. In the 1840s, Britain experienced Railway Mania—a speculative bubble that saw railway investment surge to 7% of GDP, representing half of all investment in the economy at the peak. Parliament authorised 9,500 miles of new track in 1846 alone. Share prices roughly doubled in just a few years as middle-class investors piled into railway securities, often purchasing with just 10% down.
The parallels to 2025 are striking. Revolutionary technology promising to transform commerce and society. Massive capital requirements creating opportunities for spectacular returns. Promotional schemes promising riches to investors. Government enthusiastically approving projects with minimal coordination. And, critically, a boom driven by falling interest rates—the Bank of England had cut rates in the early 1840s, creating what one contemporary described as "fertile ground for the bubble to come".
The Railway Mania eventually collapsed. By 1848, railway share values had fallen to the initial capital invested; they continued declining until the 1860s. Nearly a third of authorised lines were never built. Many investors faced financial ruin, particularly when companies called the remaining 90% due on partly-paid shares.
Yet the infrastructure endured. Unlike purely financial bubbles that leave nothing but losses, Railway Mania bequeathed Britain one of the world's most advanced rail networks. The railways continued operating, transforming commerce, enabling industrialisation, and serving the nation for nearly two centuries since. Contemporary observers of AI investment frequently cite this precedent: even if current valuations prove unsustainable and investors suffer losses, the data centres and computing infrastructure will remain, powering future economic activity.
The telecommunications bubble of the late 1990s followed similar patterns. Massive overinvestment in fibre-optic networks, vendor financing schemes, circular deals between equipment makers and service providers. The crash devastated investors and bankrupted companies including Global Crossing and WorldCom. Yet the fibre-optic infrastructure survived, eventually enabling the broadband internet revolution.
These historical examples offer both reassurance and warning. Yes, infrastructure outlasts bubbles. Yes, transformative technologies create genuine long-term value despite short-term speculative excess. But they also demonstrate that "eventually valuable" doesn't prevent devastating near-term losses for investors caught holding overpriced assets when markets correct.
The depreciation trap
AI infrastructure faces a challenge that railways largely avoided: rapid obsolescence. Victorian-era track and locomotives, once built, served for decades. Modern data centres confront a different reality. Hardware depreciates swiftly—often within months as new generations of chips arrive. Software advances can make existing installations inefficient or outdated. The economic model requires continuous investment just to maintain capability.
Harris Kupperman, chief investment officer at Praetorian Capital, has calculated the concerning mathematics. New data centres built in 2025 will suffer approximately $40 billion in annual depreciation whilst generating only $15 to $20 billion in revenue. For data centre investments to achieve returns comparable to similar-sized business ventures, they would need to generate roughly $480 billion in 2025 alone—a figure that seems wildly out of reach given current usage patterns.
The spending trajectory makes the challenge acute. Capital expenditures on AI infrastructure are projected to increase from $375 billion in 2025 to $500 billion in 2026, according to McKinsey analysis. Without corresponding revenue growth, losses will compound. A new data centre quickly becomes a Theseus's ship made from expensive components requiring constant replacement. If facilities don't generate substantial cash flows immediately, the cost of maintaining aging hardware will overtake whatever revenue they produce.
This differs fundamentally from railway economics. Once track was laid and locomotives purchased, operating costs remained manageable relative to revenues. Railways didn't require continuous reconstruction to remain functional. Data centres, by contrast, face perpetual upgrade cycles driven by Moore's Law and competitive pressure. Standing still means falling behind.
The power requirements compound the difficulty. Goldman Sachs Research estimates that data centre power demand will increase 165% by 2030 compared to 2023 levels. This expansion requires approximately $720 billion in grid infrastructure spending through 2030 just to enable the data centres to operate. Many regions already face power availability constraints, with interconnection timelines for new facilities extending beyond the development cycle of the data centres themselves.
Deloitte's 2025 AI Infrastructure Survey found that 72% of respondents consider power and grid capacity either very or extremely challenging for data centre buildouts. Supply chain disruptions concern 65% of respondents. These constraints exist before considering whether the facilities, once built, will generate sufficient revenue to justify their costs.
What comes after
Economic history suggests bubbles don't announce themselves until after they burst. During Railway Mania, contemporaries described the railway as "the wonder of the world"—much as AI is lionised today. The transformation appeared so profound, the commercial potential so obvious, that questioning the investment mania seemed naive.
Yet questioning proves essential. The concentration statistics are stark: 92% of American GDP growth deriving from 4% of the economy represents unprecedented dependence on a single narrow sector. The circular investment patterns between Nvidia, OpenAI, Microsoft, and various cloud providers create revenue and valuation growth that may not reflect genuine external demand. The rapid depreciation and continuous upgrade requirements of AI infrastructure differ markedly from past technology investments that left durable assets.
None of this means AI technology lacks value or that data centres won't play crucial roles in future economic activity. The internet survived the dot-com crash. Fibre-optic networks outlasted the telecommunications bubble. Railways remained central to commerce despite devastating investors during Railway Mania.
But "eventually valuable" doesn't prevent "immediately devastating" for those holding overpriced assets when reality intrudes. And when 92% of economic growth depends on continued investment in a sector exhibiting bubble characteristics, the potential consequences extend far beyond individual portfolio losses.
America's economy in 2025 increasingly resembles a construction site where frantic building activity creates the illusion of broad prosperity whilst the surrounding neighbourhood stagnates. The data centres rising across the landscape will likely prove useful in coming decades. The question is whether current investment levels—and the near-total dependence of economic growth on this single source—can continue without correction.
Jeff Bezos, whose Amazon pours billions into data centre infrastructure, has described the current environment as an "industrial bubble" rather than a financial one. The distinction matters. Industrial bubbles leave physical infrastructure behind; financial bubbles leave only losses. Yet both types of bubbles eventually burst, and both inflict substantial damage when they do.
The American economy has transformed itself into a high-tech monoculture, sacrificing resilience for the impressive short-term yields that concentrated investment provides. Whether this gamble succeeds depends on questions that remain genuinely unanswered: Will AI applications generate sufficient revenue to justify the massive capital expenditures? Can the technology advance fast enough to overcome rapid hardware obsolescence? Will external demand materialise at the scale required to support current valuations?
For now, the data centres rise, the investment continues, and GDP growth depends almost entirely on this single remarkable bet. The infrastructure being built will outlast whatever comes next. But for the investors, companies, and economy currently riding this wave, the sustainability of 92% of growth deriving from one sector's construction boom represents not triumph but precarity.