The AI Capex Supercycle: What’s Really at Stake
The AI capital expenditure boom is a rapid surge in spending on data centers, chips, and related infrastructure that is projected to lift AI capex from a small fraction of economic output to around 3% of overall GDP within a decade, sparking heated debate over whether this is a productive investment wave or a dangerous tech spending bubble with serious market correction risks. The key takeaway is blunt: the AI buildout now looks less like normal tech investment and more like a macro event. Hyperscaler capex has already moved from 0.3% of GDP in 2019 toward an expected 1.4% in 2025, with consensus pointing to roughly 3% between 2027 and 2029. At that pace, AI infrastructure costs stop being a niche concern and become central to how future recessions, jobs, and profits play out.
Slok and Burry: Warning Signs of a Tech Spending Bubble
Torsten Slok, chief economist at a major investment firm, is blunt: the AI infrastructure boom is “building at a pace unmatched by previous investment cycles” and could become a macroeconomic risk if AI demand disappoints. Hyperscaler capex is expected to rise from 0.6% of GDP in 2023 to about 3.1% in 2027, a 2.5‑point jump that outpaces both the late‑1990s telecom buildout and the 2000s housing surge. His key fear is the unwind: “A cycle that builds at 0.85 percentage points a year can unwind at a similar pace.” Michael Burry, famous for calling the 2008 housing crash, has seized on Slok’s “three great charts” that compare the AI data center trajectory to housing’s boom‑and‑bust, arguing that AI capital expenditure is now large and fast enough that a sharp reversal could hit the broader economy, not just chip stocks.
Jensen Huang’s Profitability Pitch: Bubble Talk Is Premature
Nvidia’s Jensen Huang rejects the bubble narrative on the grounds that AI is already paying for itself. Asked when his finance team might worry about the AI party ending, he “answered with an invoice,” pointing to the rise of coding agents that are “incredibly profitable” and doing “useful work for very high‑paying jobs.” He says many companies, including his own, are happy to spend hundreds of millions annually on AI services to augment their software development, because the productivity gains justify the AI infrastructure costs. Huang also stresses a circular but, for now, self‑reinforcing loop: labs buy Nvidia chips, run profitable AI agents, sell those agents back to firms like Nvidia, and then use that revenue to buy more chips. In his view, this is not a speculative mania but a new, profitable computing platform. The problem, critics would note, is that profitable early adopters do not guarantee broad, durable demand once the hype fades.
Jamie Dimon: AI Spending Is Fuel, Not Fragility
Jamie Dimon, running one of the world’s largest banks, takes the opposite side of the tech spending bubble story. He told a recent interviewer that he is “optimistic that the country's AI spending will pay off” and that he believes the billions flowing into AI infrastructure will “ultimately play out and pay out.” Dimon argues the boom is directly lifting economic growth, estimating that AI‑related spending has already added about 1 percentage point to GDP and will add another similar boost next year. He also points to real‑economy spillovers: “You’ve got to get steel and cement and all these things to build the data centers,” which means jobs for construction, materials, and engineering. His own bank is behaving like a believer, raising its technology budget to nearly RM19.8 billion this year, with AI one part of that push. Dimon admits there are many things to worry about in the economy, but a cooling AI market is “not high on the list.”
Is a 2008-Style Collapse Likely—or a Hard but Healthy Correction?
The uncomfortable truth is that both sides are partially right. Slok and Burry are correct that AI capital expenditure now matters at a system level: moving from 0.3% of GDP to near 3% within a decade makes any future cut‑back a macro story, not a niche tech slump. Their warning that a cycle building at 0.85 percentage points of GDP per year can unwind just as fast is not fearmongering; it is arithmetic. At the same time, Huang and Dimon are right that AI infrastructure costs are not blindly speculative. They already support profitable services, real productivity gains, and tangible jobs from steel to cement. The likely outcome is not a clean repeat of 2008 but a rougher path: overbuild, disappointment, and a market correction that exposes weak business models while leaving enduring infrastructure and winners in place. Investors and policymakers should treat AI capex as a powerful growth engine—but one that needs better demand testing and less faith in straight‑line forecasts.
