AI capex spending bubble: what it is and why it matters
An AI capex spending bubble is a period when investment in artificial intelligence infrastructure, especially data centers and cloud hardware, grows far faster than the real economy and underlying demand, raising the risk that capital will be stranded when expectations meet reality. Right now, that is exactly what alarms some of the sharpest macroeconomic minds. Apollo chief economist Torsten Slok warns that hyperscaler capital expenditures are on track to reach about 3% of gross domestic product annually between 2027 and 2029, up from 0.3% in 2019 and 1.4% in 2025. That trajectory is extraordinarily steep for a single technology stack. The buildout itself is not the danger; the risk is what happens if the promised AI demand fails to show up at the scale investors now assume.
Hyperscaler capital expenditure is climbing faster than past booms
The raw numbers behind hyperscaler capital expenditure look like a late-stage bubble chart. Slok estimates that cloud and AI hyperscaler capex will rise from 0.6% of GDP in 2023 to 3.1% in 2027, a 2.5‑percentage‑point jump. He notes that this buildout will exceed the late‑1990s telecom and fiber peak as a share of GDP, even if it stays below the 6.6% peak reached by residential investment in the housing boom. The key warning is not level but speed. The AI buildout is accelerating by roughly 0.85 percentage points of GDP per year, compared with about 0.5 points at the fastest phase of the housing boom and 0.15 in the telecom cycle. When your AI infrastructure spending curve is almost twice as steep as pre‑crisis housing, you are no longer in a normal investment cycle; you are in a leveraged bet on perfection.
From telecom to housing: the growth rate mirrors pre‑crisis patterns
Economists are not reaching for the housing analogy because it is dramatic; they are doing it because the math lines up in worrying ways. Residential investment climbed 2.2 percentage points of GDP during the housing expansion that preceded the global financial crisis, and then fell from 6.2% of GDP in early 2006 to 3.0% by the end of 2008. That reversal amplified the crisis. Telecom, by contrast, added only 0.4 percentage points at its peak and its collapse helped produce what Slok calls the mildest post‑war recession. Today, hyperscaler AI capex is growing faster than both cycles, with the data‑center buildout “smaller than housing in level but larger in the change in share of GDP, and faster than either previous cycle”. When the slope is this steep, a slowdown does not need to be catastrophic to cascade through suppliers, credit markets, and equity valuations.
Data center investment risk and the question of AI infrastructure sustainability
The most credible critics are not Luddites; they are worried about AI infrastructure sustainability. Slok’s core warning is that “a cycle that builds at 0.85 percentage points a year can unwind at a similar pace, and that, rather than the buildout itself, is the macro risk if AI demand disappoints”. In other words, the danger is an air pocket: demand projections that fail to keep up with the hardware and data centers already ordered. Michael Burry, who built his reputation calling the 2008 housing collapse, reads the same charts and sees an AI capex spending bubble. He has been a vocal skeptic of the current AI supercycle, taking bearish positions against key AI beneficiaries and arguing that today’s AI infrastructure boom could lead to excess capacity. Data center investment risk is not theoretical; it is a live question about whether this wave of hyperscaler capital expenditure can earn an adequate return before the cycle turns.
If the AI capex spending bubble bursts, it could unwind at 2008 speed
Investors like to believe they can ride the AI wave and step off before it crashes. History says otherwise. Slok explicitly connects the pace of today’s AI buildout with the speed of past reversals, noting that the telecom bust contributed to only a mild recession while the sharper housing downturn made the global financial crisis far more severe. The same arithmetic that makes AI stocks soar on capex announcements will work in reverse if budgets are cut, projects delayed, or utilization disappoints. Burry has continued to position against what he sees as an AI bubble, even as semiconductor and infrastructure names rally, on the view that capacity is running ahead of economically sound demand. The choice now is not between AI and no AI; it is between a disciplined, sustainable buildout and another cycle that explodes upward, then unwinds at a pace the financial system cannot comfortably absorb.
