AI Is Carrying the Economy. That’s the Warning.
The artificial intelligence boom is no longer just a tech story. It is showing up in the real economy through job growth, factory orders, and construction spending. That is the good news. It is also the warning.
Over the past several weeks, the evidence has become harder to dismiss. AI-related investment is appearing in construction spending, factory orders, power equipment, cooling systems, networking gear, professional-services demand, and capital markets. Firms are hiring people to install, power, finance, and integrate the new technology.
This is what a genuine investment boom looks like. It is also, potentially, how an economy becomes increasingly dependent on a single source of growth.
The strongest version of the AI bull case is that this is America’s next great capital-deepening cycle. The United States did not become dangerously dependent on railroads, electrification, the automobile, or computing. It became richer because of them. Each of those technologies required enormous upfront infrastructure investment that looked, in the early stages, like a speculative boom before it resolved into structural growth. The productivity gains lagged the investment by years, sometimes a decade, but they arrived. The early over-investment became the foundation rather than the ruin.
There is real evidence for that view in the current data. AI is already producing measurable commercial demand. Businesses are using AI tools for coding, research, customer service, workflow automation, and data analysis, and the productivity gains in those specific applications are no longer speculative — they are being reported in firm-level data. If model capabilities continue compounding and utilization rates climb, today’s infrastructure buildout could underwrite two to three percent structural growth for years, widening the U.S. lead over Europe and China in the process. The financing loop, on this reading, is virtuous rather than vicious: capital flows to hyperscalers, hyperscalers build infrastructure, the infrastructure generates cloud revenues, cloud margins validate the spending, and the cycle sustains itself because marginal returns exceed the cost of capital. The hyperscalers’ own operating margins suggest that condition still holds, for now.
The problem is not that this scenario is implausible. It is that the same evidence that makes it plausible also makes the stakes of being wrong much higher.
AI Is Driving GDP Growth
Before 2025, private investment in information-processing equipment and software contributed an average of 0.3 percentage points to annual GDP growth — a consistent, modest, unremarkable number across more than a decade. In the first half of 2025, that same category accounted for close to all of the economy’s growth. Strip it out and the economy was barely growing. Some sectors expanded, others contracted. But excluding the AI-boom, there was very little net growth. In the first quarter of 2026, according to the BEA’s GDP estimate, computer and peripheral equipment investment grew at a 67.4 percent annualized rate. Strip out AI-driven categories, and headline GDP growth would likely fall from 1.6 percent to roughly 1.0 percent. The exact contribution is hard to pin down because imported chips and equipment offset some of the domestic investment boost, but the direction is unmistakable.
Even more concerning: if we also strip out the AI-adjacent investment and activity—such as the construction of data centers—GDP growth might vanish altogether. This suggests that AI might be crowding out other growth or concealing an economy struggling to grow elsewhere, which makes this a larger concern.
Our economy may have become so concentrated in AI investment that even a deceleration would trigger a serious economic slump.
Whenever we talk about bubbles, our minds go back to the housing bubble and the dot-com bubble. So let’s be clear: this does not look like the housing bubble of the mid-2000s. Housing became a danger to the economy because debt was tied to collateral that could fall sharply in value, and much of that risk sat on bank balance sheets or inside fragile financial structures. When home prices fell, collateral values fell, bank balance sheets contracted, credit tightened, and the economy went into crisis.
The better analogy for the AI downside risk is the U.S. energy boom of 2010 to 2014. Fracking was real, and it transformed the American energy position. But when oil prices collapsed, it produced a manufacturing recession and a regional employment shock that dragged on the broader economy for two years. Even today, there appears to be hesitancy for investors to support expanded extraction for fear of another price collapse and wasted capital cycle. The technology was not fraudulent. The overinvestment corrected anyway. Real technologies can arrive wrapped in investment cycles that eventually turn.
That brings us to the internet comparison. The internet was real. The cables crossing ocean floors and the fiber infrastructure were real. The transformation of how we communicate, shop, and work was real. But that didn’t mean there was not over-investment. Investors confused a great technology with a great investment at any price. The internet eventually remade the economy. Many internet investors still got wiped out.
The risk is that AI may follow a similar pattern. It may become one of the great technologies of the age while still producing an overextended investment cycle along the way. Those two outcomes are not mutually exclusive.
AI’s Financing Loop and the Coming IPO Mania
The danger is clearest in the financing loop now forming around the sector. Capital markets fund AI companies and hyperscalers. Those companies spend inside the AI ecosystem. The money flows to chipmakers, memory suppliers, construction firms, utilities, power-equipment makers, cooling-system producers, and data-center developers. Supplier revenues rise. Those revenues validate the AI thesis. The validated thesis attracts more capital. More capital funds more capex.
That loop is healthy as long as marginal returns exceed the cost of capital — which is precisely the bull case. It becomes fragile when the financing mix shifts. The early buildout was largely self-funded. The hyperscalers ran their capex from operating cash flow, which meant a pullback in spending, while painful for suppliers, would not trigger broader credit-market feedback. That structure is changing. Bridgewater recently found that aggregate capex for the Big 5 hyperscalers, after dividends and buybacks, now exceeds projected free cash flow — meaning outside capital is no longer optional. Industry-wide AI-related debt issuance is reportedly around $300 billion this year. When the engine driving half of recent GDP growth shifts from cash-funded to capital-markets-funded, a sustained risk-off move stops being an inconvenience for technology companies and starts being a threat to the growth numbers themselves.
The coming IPO wave adds another layer. OpenAI, Anthropic, SpaceX, Cerebras, and other AI-linked companies are expected to test public-market appetite for some of the largest valuations ever attached to young technology businesses. This matters less as a stock-market signal than as a capex-funding signal. Strong valuations extend the fundraising runway that keeps the investment loop running. Weak valuations compress it.
The Politics of AI Today and Tomorrow
Official Washington is also blessing the boom. The Trump administration’s new AI order frames artificial intelligence as a national-security and economic priority. Official approval can make booms more durable by lowering skepticism and turning an investment thesis into a patriotic obligation — which means the correction, when it comes, tends to come later and from a higher base.
There’s also another risk out there: rising AI backlash. Unlike the dot com bubble or the housing bubble, AI is generating political opposition. A nontrival share of younger people—high schoolers, college students, recent grads—feel a kind of moral revulsion when confronted with AI. Instead of the youngsters telling the older generations to get with the program, many youngsters are raising the alarm. The data centers are also becoming increasingly controversial, accused of damaging the environment, damaging health of nearby residents, consuming scarce water, circumventing local building restrictions, driving up electricity costs, and building ugly box buildings marring America’s beautiful countryside. Even if some of these concerns are unfounded or exaggerated, the possibility for anti-AI politics to become an electoral winner is rising. Very few investors or AI executives are taking this risk seriously right now.
The bull case and the risk case are not arguing about different facts. They disagree on one thing: whether returns will arrive fast enough and broadly enough to justify the concentration. If they do, this is electrification. If they don’t, or if they arrive more slowly than the financing cycle demands, the economy will have organized an unusual share of its growth around a thesis that needed more time.
AI may well deliver the productivity boom its champions expect. The risk is that the economy has already begun borrowing against that future before the returns have been earned. AI now runs through construction sites, factory floors, power grids, capital markets, and the GDP accounts. If the investment cycle turns, the damage will reach far beyond Silicon Valley.