It's become a common narrative around the U.S. economy and AI firms: AI investment is propping up America. While there is an argument to be made that the stock market is, with the "Magnificent 7" tech firms making up a sizeable portion of the U.S. stock market's gains over the past year, that idea doesn't hold water with general economic growth. Indeed, Goldman Sach's latest insight into the AI industry's hundreds of billions of investment is that the effect on the U.S. economic growth so far is "basically zero," as WSJ reports.
"We don't actually view AI investment as strongly growth positive," said Goldman Sachs chief economist, Jan Hatzius, during a recent interview. "We think there's been a lot of misreporting of the impact that AI investment had on GDP growth in 2025, and it's much smaller than it's often perceived."
How much smaller? "Basically, zero," he said.
Latest Videos From
Lots of investment everywhere else
The problem when it comes to finding benefits to the U.S. economy is that most of the companies investing and receiving investment are spending that money overseas. While Nvidia might be based in the U.S., its manufacturing happens elsewhere. It gets the raw materials from other countries too, and as much as the U.S. is trying to improve American access to critical minerals in cutting-edge semiconductor manufacturing and expanding fabrication facilities on U.S. shores, it's not enough.
TSMC can't bring 40% of its manufacturing base to America, which means that for the foreseeable future, Nvidia and everyone else will be buying the majority of their chips from Taiwan. Considering that's the major component investment for AI data centers, it's clear to see why spending several hundred billion on AI data center build-outs will actually mean spending hundreds of billions on TSMC fabbed hardware.
WSJ's report suggests as much as three-quarters of the investment in building an AI data center goes on the computing components, and the majority of that is spent overseas.
Where's the money going?
The five top U.S. tech companies are collectively expected to spend as much as $700 billion on AI infrastructure in 2026. Although this is bolstering construction industries and leading to stretched electricity grids, it's barely moving the economic needle.
“This is a big deal, but not the be-all and end-all,” said economic analyst Joseph Politano, who calculates that of the U.S. economy's growth of 2.2 percent in 2025, only 0.2% of that is likely to have been down to AI investment, due to the monumental imports required.
This might not be a problem if the AI companies themselves were turning a profit. If they were pulling in money from overseas and developing a successful, profitable global product off the back of their enormous investments. But they're not. OpenAI is the biggest capital-burning company in history, with even the latest (revised down) estimates of capital expenditure on AI infrastructure reaching $600 billion by 2030, and up to $1.4 trillion by 2033.
All while the company's entire revenue for 2025 was less than $20 billion. Nvidia just massively scaled back its OpenAI investments from $100 million to maybe as much as $30 billion.
J.P. Morgan claimed in November that AI needed to be pulling in over $600 billion in revenue each year to even get a 10% return on the enormous expenditures on infrastructure.
Even OpenAI knows it can't afford its promises, which is why it's trying to build its own hardware.
This is a clear example of the main criticism levied at these huge AI tech firms over the past year: They are not profitable. And now it turns out they aren't even making money for America. The stock price of Nvidia might be sky high, Oracle's founder might be worth hundreds of billions, Mark Zuckerberg might have convinced the Meta board to let him run another investment project, but none of that helps U.S. GDP. Indeed, it could even be dragging it down - especially since these companies risk a recession if the bubble pops.
Paying for a possible future
But that's worth the risk, as far as these companies are concerned. They have been promising for years that we're mere months away from AGI being developed. Or maybe years, but it's definitely coming. We also need to be ready for AI hacking, because that's a problem now, but don't worry, Anthropic is here to help.
As much as the AI companies want us all to pay for and use their AI tools right now, a major part of their sales pitch to investors, to consumers, to politicians, and the world at large is that all this AI investment will be worth it. The models will keep scaling in capability, and all these hundreds of billions of investment will boost productivity so much that it will make it all worth it in the end. At least, that's the ideal that AI businesses are trying to sell to investors.
But we're not seeing that yet. The economic impact is basically zero, the productivity gains are poor at best, and possibly actually make us worse workers. The scale of AI expenditure rolls on regardless, and the ongoing buildout and scale of investment will march on, toward a goalpost that's ill-defined, hedged on the promise that it'll change everything. But, for America, the economic impact might as well be nil.
