Who Really Pays For Nvidia’s $1.2 Trillion?
Today’s big story: Anthropic IPO: Priced For 2031, Or 2036?
Analysts expect Nvidia to sell $1.2 trillion of chips a year by 2031. That is four times the $303 billion it sold over the four quarters to July, the base we use throughout. Adding Tuesday’s quarter, the four quarters to October come to about $362 billion.
Nvidia reported $115.8 billion of sales for the October quarter on Tuesday. It expects $130.0 billion, plus or minus 3%, for the January quarter. That quarter has an extra week, and the outlook assumes no data center compute sales to China.
So who pays? Someone has to find about $6 trillion of new spending a year. Anthropic’s IPO filing, amended Tuesday, points to a pool that big. It is the more than $40 trillion a year the world pays for knowledge work. AI would have to take over about 15% of it. Until then, investors are paying. If you own Nvidia, or an S&P 500 fund, you own this bet.

Where Does Nvidia’s Money Come From?
Nvidia sells to the companies building AI data centers. They include Amazon, Alphabet, Microsoft, Meta, Oracle and CoreWeave. With a few others, they took in $2.04 trillion over the last four quarters. About $37 of every $100 went into capital spending, mostly data centers. Nvidia’s sales equal about 40% of that spending. So roughly $1 of every $7 the buyers take in reaches Nvidia.
So $900 billion more for Nvidia means about $6 trillion more for its buyers. That is the step from $303 billion to $1.2 trillion. At $1 in $7, the buyers would need about $6 trillion of new revenue every year, on top of today’s $2 trillion. That is almost a fifth of the $33 trillion US economy.
Nvidia’s path to $1.2 trillion runs through a handful of big buyers. If you like finding quality companies like Nvidia, the Trefis High Quality (HQ) Portfolio holds 30 of them in equal weight. It has returned more than 105% since its September 2020 launch. See the HQ Portfolio.
The middle column assumes Nvidia stays at 40% of each buyer’s capital spending. At $1.2 trillion, that spending would be 4.0 times today’s.
| Buyer | Capital spending, last 4 quarters | Needed at $1.2T for Nvidia | Cash from operations |
|---|---|---|---|
| Amazon | $215B | $861B | $168B |
| Alphabet | $160B | $639B | $180B |
| Microsoft | $136B | $543B | $196B |
| Meta | $104B | $418B | $133B |
| Oracle | $76B | $303B | $47B |
| CoreWeave, Nebius | $45B | $180B | $12B |
| Tesla | $15B | $62B | $19B |
| Total | $752B | $3.0T | $755B |
Source: Trefis company data, each company’s last four reported quarters.
The buyers’ total spending doubled in a year. It already equals all the cash their operations bring in.
Who Has $6 Trillion To Spend?
Not the buyers themselves. They already spend every dollar their operations bring in. The gap they would have to borrow is traced in Could Nvidia Stock Fall 50%? Follow The Cash.
Not consumers. A billion people paying $20 a month for AI is $0.24 trillion a year. That is about five times Netflix‘s revenue, but not even 5% of what’s needed.
Not advertisers. Google and Meta took in $713 billion over the last four quarters, mostly from ads. Another 20% would add $0.14 trillion.
Not governments, so far. The federal government spends $7.9 trillion a year. But government agencies are under 2% of Anthropic’s revenue, its filing says. At Anthropic, at least, companies are the ones paying.
Not companies buying AI to help their workers. Even $1,000 a month of AI for 100 million workers is $1.2 trillion a year. That is about a fifth of what’s needed.
These are generous cases, not forecasts. Together they cover about a quarter of the $6 trillion. The other $4.5 trillion would have to come from everyone else, such as governments, AI labs, robots and cars.
What Does Anthropic’s Filing Point To?
The pool is knowledge work. Anthropic’s amended IPO filing puts the work knowledge workers do at more than $40 trillion a year worldwide. It says its opportunity is to make that work more productive. The $6 trillion is 15% of it.
The $40 trillion is a pay bill, not AI sales. The filing adds up what knowledge workers earn. As a rough check, the world produced $118 trillion in 2025. If workers get about 57% of it, as in the US, the world’s pay bill is about $67 trillion. Knowledge work is about 60% of that. The $40 trillion is Anthropic’s own estimate, and the filing cautions readers not to give such estimates undue weight.
So the $1.2 trillion is not a bet that AI replaces the world’s knowledge workers. It is a bet that AI does about 1 in 7 of their working hours, and gets paid for them. If AI charges half what the person costs, it needs about 30% of the hours.
Companies are already paying, and fast. Anthropic took in $17.0 billion in the quarter to September, up from $1.4 billion a year earlier. By the end of October it was running at about $81 billion a year. More than 11,900 customers now pay it $100,000 or more a year, up from more than 950. That one company is already at about 1.3% of the $6 trillion.
Investors are paying the rest, for now. Anthropic has committed $900.9 billion to computing. $451.6 billion of that falls by the end of 2030, and $70.7 billion in the 12 months after September. It has raised $141.1 billion so far. It says its cloud and chip partners funded or supported part of its build-out. So chip and cloud companies help fund the customers who buy their chips. Anthropic runs on Amazon, Google, Nvidia and AMD chips, so not all of its spending reaches Nvidia.
Its IPO is expected to price tonight. It would open this funding to the public. You could become one of the funders.
So the question for Nvidia’s price is not whether AI is useful. It is whether companies move enough of their knowledge-work budgets to AI before investors stop paying in.
Why Does AI Have To Do The Work?
Helping people work is not enough. AI has to replace work that people are paid for today. Take Nike. Over its last four quarters, it spent $16.0 billion on marketing, selling and administration. It earned $3.8 billion of operating profit, its profit before interest and taxes. Suppose AI costs 5% of that budget, or $800 million a year. That is 21% of its profit. Nike can pay in one of three ways:
- AI replaces costs it already has. The bill pays for itself, and shoes can get cheaper.
- Prices go up. You pay, under 2% more.
- Profit goes down. Shareholders pay.
Only the first can grow to trillions of dollars. The table runs the same 5% for two smaller companies, Abercrombie and Under Armour.
| Company | Selling and admin, last 4 quarters | 5% for AI | Operating profit |
|---|---|---|---|
| Nike | $16.0B | $800M | $3.8B |
| Abercrombie | $2.7B | $134M | $0.7B |
| Under Armour | $2.3B | $114M | about $0 |
Source: Trefis company data. The 5% is our assumption, to show the math.
Through the chain, about $119 million of Nike’s $800 million would reach Nvidia. Under Armour has no operating profit, so it cannot pay from earnings at all.
Has Work Ever Moved This Fast?
Getting to 15% by 2031 means AI taking over about 3.2% of knowledge work a year. America has moved work at that pace twice. Farm jobs fell 32% from January 1948 to January 1960, or 3.2% a year, as tractors spread. Factory jobs fell 34% from January 2000 to January 2010, or 4.0% a year, to automation and imports.
But never for long. From 1948 to 2025, farm jobs shrank 1.6% a year. Factory jobs have shrunk 0.9% a year since their 1979 peak. At the farm average, 15% takes about 10 years. The $1.2 trillion would then arrive around 2036, not 2031. At the factory pace, it would arrive around 2044. In our view, the destination looks plausible. The date is the bet.
| Shift | Jobs lost | A year |
|---|---|---|
| Farms, Jan 1948-Jan 1960 | 32% | 3.2% |
| Factories, 2000-2010 | 34% | 4.0% |
| Farms, 1948-2025 | 71% | 1.6% |
| Factories, 1979-2026 | 35% | 0.9% |
| Knowledge work, to 2031 | 15% needed | 3.2% |
Source: FRED, US farm and factory jobs. The last row is what the $1.2 trillion path needs, not a forecast.
What did farm workers do next? They moved into new kinds of work. Total US payrolls grew 3.5 times since 1948. Private health care and education went from 1 job in 22 to 1 in 6. No data can name tomorrow’s jobs for knowledge workers. The record says they come over decades, not quarters.
Is AI Pushing Up Prices?
Possibly, at the edges. Over the five years to October, electricity rose 34%, against 22% for prices overall. Building data centers adds demand for power, chips and construction now. The savings come later, and only if AI replaces costs.
What Does It Mean For Your Money?
You may already own this bet. In State Street’s S&P 500 fund on Nov 16, Nvidia was 5.9% of the holdings. The 10 biggest holdings, most of them in this chain, were 37%. If the money slows before companies move their budgets, those holdings feel it first.
Our read: in the end, Microsoft and Amazon do not pay for Nvidia’s $1.2 trillion. Companies do, by moving part of the $40 trillion they spend on knowledge work to AI. Until then, investors pay. The farm burst says this can happen in about five years. The full farm record says closer to ten.
A later date means a longer wait for Nvidia holders. Yesterday’s Does Nvidia Have Another 10x Run In It? shows how deep the stock fell during its past runs. If Nvidia is a large share of what you own, plan now how much of it to keep.
Too Much Riding On One Stock?
If Nvidia is a big part of your wealth, so are its risk and its tax bill. What’s your exit plan? Our partner advisory firm offers a free 30-minute strategy review covering a floor under losses and a lower-tax way out.
For education, not a recommendation to buy or sell.
For questions about this article or underlying research, email help@trefis.com.