Nvidia CEO Jensen Huang introduced Vera Rubin, a next-generation AI data center platform, and Rubin Ultra, a next-generation AI GPU architecture, during a keynote address at the company’s annual GTC developer conference on March 16, 2026 in San Jose, California.
Josh Edelson | AFP | Getty Images
Nvidia CEO Jensen Huang has far exceeded even the most optimistic forecasts for spending on AI.
During an earnings call Wednesday night, Hwang said he believes AI capital spending could reach up to $4 trillion.
“Capex is at $1 trillion and increasing towards $3 trillion to $4 trillion (trillion dollar level),” he said, speaking only of hyperscaler capex, including: alphabet and Amazonother segments of the supercomputing market such as neo-cloud are excluded.
Colette Kress, Nvidia’s chief financial officer, was more specific on the conference call.
“Analysts now predict that hyperscale capital spending will exceed $1 trillion in 2027, and as agent AI becomes more prevalent (in all industries), spending on AI infrastructure will reach $3 trillion to $4 trillion annually by the end of this decade,” he said.
One thing I can tell you is that this is far ahead of the trajectory Wall Street expected.
One analysis by Needham’s Laura Martin shows a consensus estimate that hyperscaler capital spending will reach $1.03 trillion in 2028, which would represent a third to a quarter of what it would be in just two years if Huang’s prediction is correct.
“If Jensen Huang’s predictions are correct, we believe the consensus forecasts included in the chart below will be revised upwards,” she and colleague Dan Medina wrote on Thursday. “(His) vision for hyperscalers is different and more interesting than what hyperscalers are saying on their earnings calls.”
Some on Wall Street predict that capital spending will reach $1 trillion by the end of next year, faster than consensus, but that remains well below Mr. Huang’s prediction that the number would quadruple over the next three years.
No doubt, further infrastructure investment from hyperscalers and others will benefit Nvidia’s business as a leading AI chip maker. But rising cloud revenue and continued advances in frontier algorithms seem to support Huang’s optimism so far.
Quarterly revenue exceeded expectations for all major cloud companies, with Alphabet up 63%, AWS up 28% and Microsoft up 40%.
“There are a billion users in the world, human users. My sense is that there will be billions of agents in the world… and all of those agents will spin off subagents,” Huang said.
Too early to reach consensus on productivity
Despite progress, revenue growth, and frequent historical comparisons to railroads and other capital-intensive industrial development stages, significant questions remain about the long-term impact of AI on profitability, productivity, and ultimate viability.
JPMorgan estimated in November that achieving a 10% return on AI investments through 2030 would require about $650 billion in annual revenue in perpetuity, a number they called “staggeringly large” and “equivalent to 0.58 percentage points of global GDP, or $34.72 per month from every current iPhone user, or $180 per month from every Netflix subscriber.”
For comparison, cloud revenue reached $455 billion in the 12 months starting in April, according to Synergy Research Group.
Cedric Durand, an economist at the University of Geneva, wrote in January: “If efficiency is achieved, there will be no problem. Prosperous companies will have sufficient resources to pay their bills.” “Within a few years, your customer base won’t be able to get out if AI permeates business processes and the cost of exit becomes prohibitive.”
However, productivity gains from AI are not yet a reality. Moreover, there is no consensus among economists.
“Is this the beginning of an AI productivity boom? Maybe!” wrote Martha Gimbel, an economist at the Yale Budget Institute, in February. “Until we get some sort of clear signal, we shouldn’t put all our eggs in the public basket of productivity data.
In March, Federal Reserve economists found “substantial heterogeneity in AI adoption across companies,” explaining the discrepancy between perception and reality about AI’s impact.
“Perceived productivity gains are larger than measured productivity gains and likely reflect delays in revenue realization,” they write.
