Could AI Investment Become the Next Economic Risk? Apollo Warns of 'Subprime Crisis-Style' Reversal Shock

The investment boom in artificial intelligence (AI) infrastructure is sweeping global capital markets at an unprecedented pace. However, behind this frenzy lies a growing macroeconomic risk that is gradually coming to light.

Torsten Slok, Chief Economist at Apollo Management, recently analyzed that capital expenditures by major cloud service providers are expected to rise from 1.4% of U.S. GDP in 2025 to approximately 3% between 2027 and 2029—just two to three years later. This scale would be more than double the peak of the late 1990s telecom and fiber-optic construction boom.

In his report, Slok compares this AI investment surge to two previous major capital expenditure cycles. In terms of absolute size relative to GDP, this data center construction boom still falls short of the peak of the U.S. housing bubble in 2005 (which reached 6.6% of GDP), but it is already more than double the late 1990s telecom boom peak (around 1.2%).

However, what truly alarms Slok is not the scale of investment itself, but the "speed of ascent." He notes that data center capital expenditures are projected to climb from 0.6% of GDP in 2023 to 3.1% by 2027—a mere four-year span—representing a cumulative increase of about 2.5 percentage points. This already exceeds the roughly 2.2-percentage-point rise seen during the housing boom (mid-1990s to 2005) and far surpasses the telecom boom’s mere 0.4-percentage-point gain.

In other words, when measured by "marginal contribution to GDP," AI data center investment has become the largest capital expenditure cycle in history.

Even more noteworthy is the pace of construction itself. The report shows that data center capital expenditures will surge from 1.4% in 2025 to 3.1% in 2027 in just two years—an average annual increase of approximately 0.85 percentage points.

By comparison, the most intense phase of the housing boom (2002 to 2005) saw an average annual increase of only about 0.5 percentage points, while the telecom boom averaged just around 0.15 percentage points.

That means the current AI construction frenzy is advancing at nearly twice the speed of the housing bubble’s peak period.

Slok emphasizes a symmetrical logic in his report: the faster an investment cycle rises, the more forceful the impact may be when it reverses.

He draws on two historical episodes for comparison:

Around 2000, when the telecom bubble burst, the relatively limited scale of investment meant the impact on the U.S. economy was mild, triggering only the shallowest recession in the post-war era.

In contrast, from early 2006 to late 2008, housing investment as a share of GDP contracted sharply by about 3.2 percentage points in just three years—one of the key drivers behind the subsequent financial crisis and severe recession.

By analogy, today’s data center capital expenditures are expanding at an annual rate of approximately 0.85 percentage points. If corporate expectations for AI commercialization fail to be realized on schedule, this wave of capital could contract just as rapidly.

In the conclusion of his report, Slok states plainly that the real macroeconomic risk to watch is not the expansion of AI infrastructure investment itself, but the sharp reversal that could follow if market enthusiasm for AI fades—potentially delivering a significant blow to overall economic growth.