AI Hyperscaler Spending Is Entering Uncharted Territory: 3x the Telecom Boom

The artificial intelligence arms race is reaching a scale that makes the late-1990s telecom boom look surprisingly small.

US hyperscalers could spend about $916 billion on capital expenditures over the next 12 months, according to current consensus estimates. That figure is expected to climb to nearly $1.2 trillion the following year.

The comparison with previous investment booms is even more striking.

Hyperscaler capex is expected to reach roughly 3.1% of US GDP in 2027, according to data compiled by Apollo Global Management Chief Economist Torsten Slok.

Telecom investment peaked at just 1.2% during the dot-com infrastructure boom.

That means today’s AI capex race could become almost three times larger relative to the economy. And unlike the telecom boom, spending is still accelerating.

A Capex Cycle Unlike Anything Before It

In an emailed note shared Thursday, Slok highlighted that the speed of the AI buildout may matter even more than its size.

“What matters is not the level of the share but how much it moves, because that is what adds to or subtracts from GDP,” Slok said.

The numbers put that acceleration into perspective.

Data-center capex is projected to rise by 2.5 percentage points of GDP, from 0.6% in 2023 to 3.1% in 2027.

Telecom investment increased by just 0.4 percentage points during the late 1990s. Even the housing boom added a smaller 2.2 percentage points from the mid-1990s through its 2005 peak.

“On this measure, the data-center buildout is the bigger capex cycle,” Slok said.

The speed of the buildout is even more remarkable.

Data-center capex is expected to jump from 1.4% of GDP in 2025 to 3.1% in 2027. That’s an increase of roughly 0.85 percentage points per year.

Housing’s fastest expansion ran at about 0.5 percentage points annually. Telecom peaked at roughly 0.15.

“The AI cycle is building at close to twice the pace of the housing boom at its fastest,” Slok said.

Six Companies Are Driving the AI Spending Race

Current consensus estimates show six U.S. hyperscalers could deploy roughly $916 billion over the next 12 months, before spending rises toward $1.17 trillion over the following 12 months.

“The AI buildout has turned into its own stimulus program for the economy,” Ed Yardeni said in a note on Thursday.

“And, of course, the government deficit remains very stimulative.”

The striking part isn’t only the size. Companies keep revising the numbers higher.

Amazon recently lifted its 2026 cash capex forecast from roughly $200 billion to $220 billion, citing higher memory costs. AWS still expects capacity constraints despite the additional spending.

Alphabet raised its 2026 capex outlook to $195 billion–$205 billion, up $15 billion from its previous range. Google Cloud revenue surged 82% last quarter.

Microsoft expects roughly $190 billion of calendar 2026 capex. About $25 billion reflects higher component prices. The company still expects compute capacity to remain constrained through 2026.

Then there is SpaceX, which adds another dimension to the race. Capital expenditures reached roughly $18.4 billion last quarter, including $15.8 billion directed toward AI expansion.

Analysts now expect SpaceX to surpass Oracle in capex spending within two years.

What Happens If AI Demand Disappoints?

This is where the telecom comparison becomes uncomfortable. Telecom investment collapsed after companies built more capacity than demand could absorb.

Today’s hyperscalers have vastly stronger balance sheets. But investment cycles still work in both directions.

The AI capex acceleration is providing a powerful tailwind to economic growth today. But it also creates a vulnerability if AI demand eventually falls short of expectations.

“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,” Slok said.

That’s the macro risk hiding inside the AI boom.

For now, investors are watching nearly $1 trillion of annual spending ripple through semiconductors, power and infrastructure.

The danger begins if the revenue needed to justify that infrastructure fails to arrive.

Image: Shutterstock