Huang Just Reset the Ceiling — Again

Wall Street thought $1 trillion in AI capex by 2027 was the aggressive call.

Jensen Huang thinks that's the starting line.

On Nvidia's Wednesday earnings call, Huang said AI capital expenditures are already at $1 trillion and headed toward $3 to $4 trillion annually. He was talking specifically about hyperscalers — Alphabet, Amazon, Microsoft — not the broader AI infrastructure market.

Nvidia CFO Colette Kress put a timeline on it: "AI infrastructure spending is on track to reach $3 to $4 trillion annually by the end of this decade," according to CNBC.

The Analyst Gap Is Staggering

Laura Martin at Needham ran the numbers. The Wall Street consensus has hyperscaler capex hitting $1.03 trillion in 2028. That's one-third to one-quarter of what Huang is projecting just two years later.

Martin and her colleague Dan Medina wrote Thursday: "If Jensen Huang's prediction is correct… then the consensus estimates included in the chart below will be revised upwards."

The gap between these numbers is substantial. Analysts are working from a fundamentally different model of how fast AI deployment scales.

The Revenue Numbers Back Him Up

Huang isn't guessing. He's watching his customers' revenue.

AWS grew 28% year-over-year. Microsoft Azure climbed 40%. Google Cloud also posted strong growth. According to CNBC, quarterly revenues came in above expectations for every major cloud provider.

When customers are printing money at those rates, they spend more on infrastructure.

Gartner's Numbers Tell the Same Story — Differently

For broader context: Gartner estimates total global AI spending — not just hyperscaler capex — hits $2.59 trillion in 2026, up 47% year-over-year, according to The Hindu BusinessLine. By 2027, Gartner projects that number reaches $3.49 trillion.

Breaking it down: AI infrastructure alone goes from $975.6 billion in 2025 to $1.89 trillion by 2027. That's more than half of all AI spending. Software spending nearly doubles to $638 billion. AI cybersecurity — one of the fastest-growing segments — jumps from $25.9 billion to $86 billion by 2027.

Huang's projections and Gartner's projections aren't apples to apples — Huang is talking about hyperscaler capex specifically, Gartner is measuring total global AI spend. But directionally, both are pointing at numbers that make current analyst consensus look like a lowball.

The Federal Government Is Also Going All-In

Private sector money isn't the only source.

A May 18, 2026 analysis by Brookings Institution researchers James Denford, Gregory Dawson, and Kevin Desouza found federal AI spending on a "sharp, upward trajectory" with intense focus on the Department of Defense.

The Trump administration's AI Action Plan, released in July 2025, explicitly targets "global AI dominance" through deregulation and infrastructure investment — a hard pivot from the Biden-era approach. Brookings notes the federal AI contract landscape has shifted from short-term experimental pilots to multi-year implementation contracts, with larger vendors now entering government work.

Washington is moving beyond the pilot stage.

What Mainstream Coverage Is Getting Wrong

Most financial media is covering Huang's $4 trillion projection as a bold outlier — a CEO talking his own book to boost Nvidia's stock.

Huang isn't projecting demand for Nvidia chips in isolation. He's describing the aggregate infrastructure commitment already being made by the companies writing the actual checks. Alphabet, Amazon, and Microsoft don't announce capex projections to please Nvidia — they announce them because their boards approved the spending.

Every major buyer of AI infrastructure is spending faster than analysts modeled, and analysts are still catching up quarter by quarter instead of resetting their frameworks.

Federal spending also deserves more attention. The private-sector AI race gets all the headlines, but the DOD is quietly becoming one of the biggest AI customers on the planet, with significant implications for national security and technology policy.

What This Means for Regular People

If Huang's projections hold, the scale of capital being deployed into AI infrastructure over the next four years will dwarf anything seen in the history of technology investment. That money builds data centers, buys chips, runs cables, and consumes electricity — in real places, employing real people.

Companies that aren't integrating AI into their operations are falling further behind every single quarter. The window to dismiss this as hype closed a long time ago.

Four trillion dollars annually isn't a projection anymore. It's a budget.