The amount of money spent on hardware and building behind the recent AI boom may be enormous by historical standards. How is it financed?
Stijn Van Nieuwerburgh tells the story in “Financing the AI Buildout” (Brookings Papers on Economic ActivityFall 2026, conference draft version). As he explains, new AI technologies are often discussed as new software, and how the software’s capabilities are evolving. The software perspective is correct, but partially so. The new AI technology is also a physical capital boom. Indeed, our public discourse on AI seems to have entered a time where the concern is that AI software will be cheap and available which will lead to mass unemployment, while another concern is that building the physical capital behind the AI boom – including data centers, semiconductor factories, additional energy production, etc. – will be too expensive. I don’t like to make predictions, but I believe that the new AI technology will not be cheap and expensive at the same time.
Van Nieuwenbergh runs through a set of calculations to forecast the size of physical capital AI investment in the next few years, based on current plans and the recognition that not all current plans will actually happen. He produced this historical comparison, which should be taken with a few grains of salt. The AI builder is predictive, while the rest happens. Also, the time period is different: railway investment, the second highest number in the right column, is longer than the forecast for the development of AI. Still, the estimates for the cost of building AI are impressive.
Where did the company get the funds to build this? Remember, stock market wealth is not a direct source of funding—unless you sell the stock to someone else. One of the economic puzzles of the last decade is the large US technology companies holding cash, more than $ 1 trillion (for example, I wrote about this in 2013, here and here). A common question is: What investment opportunities are you waiting for? The answer appears to be AI buildout. Van Nieuwenbergh writes:
Aggregate capital expenditure by Oracle, Microsoft, Amazon, Meta, and Alphabet increased from approximately $97 billion in 2020 to more than $400 billion in 2025 and is expected to exceed $800 billion in 2026. In current estimates, capex will exceed the combined five companies’ combined operating cash flow for the first time in 2026. substantial, but no longer sufficient to finance the projected pace of investment without further reliance on external capital or structure Financing that moves assets and liabilities away from the operating company’s balance sheet.
New external financing is a combination of equity and debt financing. Equity finance comes in the form of “[s]specialized data center developers, REITs [real estate investment trusts]infrastructure funds, and private equity investors. Debt finance includes bank loans and bonds, but private credit funds provide a growing share.
For those with an interest in personal credit in particular, Puriya Abbassi, Iñaki Aldasoro, and Sebastian Doerr provide an overview on “Financing the digital economy: the role of personal credit” (BIS Quarterly ReviewSeptember 2026, pages 17-30). Unlike bank loans, which often require the borrower to have physical collateral to back up the loan, personal loans make loans based on expected cash flow and intangible assets. While the banks have been strictly regulated, private credit funds have increased, and by 2025 they will have an outstanding debt of $2.5 trillion. As the author notes: “Aggregate [private credit] debt by technology companies increased from about $22 billion (22% of total private credit) in 2010 to more than $1 trillion (44%) in 2025. These private credit funds typically operate by taking money from investors who are usually committed for a medium term of five to eight years, and cannot be withdrawn. If the bank has a lot of non-performing loans, the bank has borrowed money from depositors, which can cause problems; if the loan portfolio of a private credit fund becomes sour, investors in the fund will suffer losses, but there is no reason to think that the losses will be in the banking or other financial systems.
One of the strengths of the market economy is that some companies can succeed while others fail, without jeopardizing the health of the economy as a whole. But as the scale of investment grows, and risk is distributed across a network of interconnected companies, it’s worth thinking about which investors will bear the brunt if the AI boom doesn’t turn a profit quickly enough — and whether the losses from those investors might be big enough to affect the broader economy. Some readers will remember the dot-com boom-and-bust when the overall stock market fell by about 50% from March 2000 to October 2002, and many internet companies fell even further. There was a recession for eight months in 2001. However, although the economy collapsed and some companies disappeared immediately, internet technology did not disappear, and the physical capital boom behind the internet was used productively.
In some ways, the physical capital boom around AI reminds me of a rhetorical question I remember hearing about the California gold rush in the mid-1800s: During the California gold rush, which group was richer: the gold miners, or the people who sold picks and shovels to the gold miners? Many gold miners are unsuccessful, so the average profit for gold miners is not all that high. But every gold miner, successful or not, needs to buy these tools. Currently, the stock market price of AI software companies is very high. But some of these companies may turn out to be unsuccessful gold miners—those who have to pay the costs of physical investment in new technology.