At the latest earnings call, Jensen Huang said that annual spending on AI infrastructure will reach $4 trillion, four times the Wall Street consensus. This money will ultimately reach every ordinary person in the form of electricity bills, subscription fees, or even job opportunities.
Article author and source: AI World
NVIDIA is now worth $5.7 trillion.
This figure exceeds Germany's full-year 2026 GDP forecast of $5.45 trillion.
A company that sells chips is worth more than Europe's largest economy.
On the evening of May 20, NVIDIA released its Q1 fiscal year 2027 financial results, reporting revenue of $81.6 billion, an 85% year-over-year increase, significantly surpassing Wall Street expectations.
The data center business contributed $75.2 billion, a 92% year-over-year surge, accounting for over 90% of total revenue.
Net profit of $58.3 billion, more than tripling year-over-year.
Even more striking is the next quarter's guidance of $91 billion, which exceeds analyst expectations by over $4 billion.
NVIDIA has also added $80 billion to its stock buyback authorization.
This company is making so much money it doesn’t know what to do with it.
Four trillion dollars—whose money is it really?
The financial figures are just the appetizer.
The judgment Huang Renxun made during the subsequent conference call was truly astonishing.
The AI capital expenditures of hyperscale cloud providers have currently reached $1 trillion annually and are expected to grow to $3 to $4 trillion.
What is Wall Street's consensus expectation?
According to an analysis by Needham analyst Laura Martin, industry participants believe that hyperscale cloud providers' capital expenditures will not reach $1.03 trillion until 2028.
The number Huang Renxun mentioned is four times this consensus.
NVIDIA CFO Colette Kress provided a timeline, estimating that annual spending on AI infrastructure could reach $3 to $4 trillion before 2030.
Needham analyst Laura Martin stated in her research report that Huang’s vision differs from—and is more interesting than—the picture described by cloud providers themselves.
The money has already been spent.
In the first quarter, Google's capital expenditures amounted to $35.7 billion, doubling year-over-year; Amazon's reached $44.2 billion, the highest among the four; Microsoft's totaled $30.9 billion, up 84% year-over-year.
Meta was the most aggressive, raising its full-year capital expenditure budget to $125 billion to $145 billion, but the market slapped it back—the stock fell 9.25% the next day.
Combined, the four companies are expected to spend $725 billion in full-year 2026.
Bank of America predicts that cloud providers will issue a total of $175 billion in debt this year, six times the average annual level over the past five years.
What does $4 trillion actually mean?
Approximately equal to Japan's annual GDP.
This money must ultimately be earned back from somewhere.
Your electricity bill is paying for AI.
This high-stakes gamble may sound distant, but it’s already changing everyday life—starting with electricity bills.
John Steinbach, a resident of Virginia, received a $281 electricity bill in January 2026, compared to approximately $100 the previous month.
He lived in this house for nearly 40 years and had never seen such a surge.
Virginia is the region with the highest concentration of data centers in the United States, with data centers alone consuming nearly 40% of the state’s electricity in 2024.
This is not an isolated case.
https://www.consumerreports.org/data-centers/ai-data-centers-impact-on-electric-bills-water-and-more-a1040338678/
According to SemiAnalysis’s research, the PJM grid region, which covers 13 states and 67 million residents in the eastern United States, has seen an average increase of approximately 15% in household electricity bills in 2026 compared to before the AI data center era.
According to data from the International Energy Agency, a typical hyperscale data center consumes as much electricity as 100,000 households.
Meta’s planned Hyperion project in Louisiana requires at least 5 gigawatts of power—three times the total electricity consumption of New Orleans.
By 2028, electricity consumption by data centers in the United States is expected to account for 12% of the nation's total electricity use.
By 2030, U.S. electricity bills are projected to rise by an average of 8%.
This is a simple accounting issue: tech giants want to build AI factories, factories need electricity, so who bears the cost of upgrading the power grid?
At least for now, the answer is everyone.
100 AI employees surround you.
Electricity costs are just the beginning.
Jensen Huang described a broader picture on the earnings call: there are currently one billion human users worldwide, and soon the world will have billions of agents, each of which will spawn sub-agents.
This is not something to be taken lightly.
At the GTC conference this past March, he provided more specific figures, estimating that in ten years, NVIDIA will have 75,000 human employees alongside 7.5 million agents—that is, 100 agents per person.
A McKinsey survey from last November found that 62% of companies are already testing agents.
Andrej Karpathy conducted an experiment in which an agent optimized the training process of a small language model, running 700 experiments over two days and identifying 20 optimization strategies.
https://x.com/karpathy/status/2030371219518931079
However, there is an unavoidable reality here.
The agent's reliability is still far from being ready to be let loose to act on its own.
An agent of a company, after gaining elevated privileges, deleted an entire production database in nine seconds, wiping out customer data, booking records, and backups.
ServiceNow CEO Bill McDermott directly said, "Governance capability is a matter of life and death."
The computational demands of agents have further increased the appetite for computing power.
Jensen Huang revealed that the computing power required for agentic AI has increased by 1,000% compared to generative AI two years ago.
NVIDIA’s next-generation Vera Rubin platform is designed specifically for this, reducing inference token costs to one-tenth of Blackwell’s and cutting the number of GPUs needed to train models of the same scale to one-quarter.
Leading labs such as Anthropic, Meta, OpenAI, and Mistral AI have officially announced they will train their next-generation models based on Rubin.
https://nvidianews.nvidia.com/news/nvidia-vera-rubin-platform
From this perspective, the $4 trillion infrastructure projection is not even aggressive.
Toll booth on the highway to AGI
All numbers, all inputs, ultimately lead to the same destination.
When inference costs drop by a factor of 10, model sizes continue to expand, and billions of agents operate autonomously and collaborate with one another, the end of this technological curve has only one name: AGI. Beyond that lies ASI—artificial superintelligence.
A $4 trillion infrastructure investment is, at its core, building a highway to AGI.
Huang Renxun is betting that the endpoint of this path is valuable enough to make all the investments along the way insignificant.
If AGI truly arrives by the end of this decade, all current discussions about whether AI investments can break even will become irrelevant.
A system capable of autonomously completing nearly all cognitive tasks would redefine the very concept of "return on investment."
At that time, there was only one question: “Who has the right to sit at the table in the AGI era?”
NVIDIA has taken its seat. The four major cloud giants are betting real money.
And every ordinary person will be a stakeholder in this high-stakes gamble, whether you like it or not.
Reference materials:
NVIDIA releases its first-quarter financial report for fiscal year 2027.
https://www.cnbc.com/2026/05/21/ai-spending-expected-to-top-1-trillion-in-2-years-why-that-estimate-may-be-too-low.html