Stock prices indicate that the global AI boom is unstoppable. Trillions of dollars are based on the idea that machines will change work, productivity and profits.
But AI is more than a race in technology. It is one of the largest infrastructure projects ever funded by debt, and strains are beginning to show.
The flow of money into AI-based data centres has changed credit markets and brought private lenders to the forefront of tech finance. It has also tied the fates of many companies together in ways that few investors are able to see.
Some call it growth and others call it leverage.
AI is no longer software
During the majority of the last two decades, large technology companies followed a similar pattern. They invested heavily into research, built software platforms and scaled at low cost.
Cash piles grew faster compared to borrowing. But not anymore.
For training and running advanced AI systems, a large physical infrastructure is required. Data centres need to be built quickly. Power contracts need to be secured. Chips must also be purchased in huge quantities.
UBS estimates that AI data centres and project finance issuance will reach $125 billion by 2025. This is up from only $15 billion just a year ago.
McKinsey estimated recently that total data center spending could reach $7 trillion within the next decade.
Even the largest companies cannot fund this alone. Bank of America reports that Amazon, Google Meta, Microsoft, and Oracle issued $121 billion in new debts this year. This is more than four times the average annual amount they have issued previously.
Another $100 billion is expected in the next year. These companies still generate a strong cash flow but their ability to self fund has grown beyond their current operations.
The result is that the sector looks less like software, and more like utilities or telecommunications. Returns are dependent on timing, utilisation and financing costs.
Credit markets are starting to price this risk differently.
Oracle shows how quickly sentiment can change
Oracle has become a clear test case for the new AI economy. The company’s growth was boosted by optimism about cloud growth and OpenAI.
Oracle shares nearly doubled in value for the year at their peak in September. Since then, things have changed.
Oracle shares dropped more than 11% on a single day after missing revenue expectations. Larry Ellison’s worth has dropped by around $25 billion.
Credit markets’ performance was more important than stock market movements.
Oracle’s investment grade bonds, including the $18 billion issued in September, have fallen sharply. Paper losses have now exceeded $1 billion. Credit default swap spreads are at levels last seen in the financial crisis.
The web of circular finance
Oracle’s stress is part of a much larger network of financial connections. Nvidia and OpenAI are at the heart of this network.
Nvidia has been the clear winner to date. It is a major AI chip supplier and has a strong profit record.
Even Nvidia’s future success depends on the spending of others. Many AI developers don’t have the money to buy chips outright.
Nvidia has taken steps to solve this problem by investing in customers’ equity, extending financing, and backing deals that help fund the data centres. Cash is often withdrawn and then returned as chip purchases.
OpenAI plays an equally central role. It is a major client of Oracle, Amazon and CoreWeave. It is also a major investor in some of these companies.
OpenAI has committed itself to purchasing hundreds of billions in computing power over the course of time, while generating annual revenues of about ten billions and large losses. It is still many years away from profitability.
CoreWeave is another prime example. It currently has no profit, a debt of about $14 billion, and lease obligations worth tens or hundreds of billions. Microsoft accounts for around 70% of the company’s revenue.
Nvidia is a supplier as well as an investor, while OpenAI serves both as a customer and partner.
Money circulates within a small group, amplifying the growth when conditions are favorable and increasing risk when they’re not.
This structure allows for a faster expansion of the system than would be possible with traditional balance sheets. It is also harder to determine where losses will fall if demand slows down or timelines slip.
When debt is written off the books
Companies have sought ways to keep their balance sheets clean as borrowing has increased. Special-purpose vehicles are becoming more common.
Meta, xAI and Google are among the companies that have used these structures in order to finance data centres and chips purchases without recording their full debt load.
The vehicle builds the asset, then leases it back to the tech company. These arrangements preserve flexibility and credit ratings.
They also reduce transparency. Rating agencies and investors perceive less risk at the corporate levels, even though economic exposure remains.
Before 2008, similar structures were used in both banking and companies like Enron.
Private credit has been a major factor in funding this growth. Morgan Stanley estimates that private lenders could provide over half of the $1.5 trillion required for data centres by 2028.
These lenders are not heavily regulated. Disclosure is limited. The number of links to banks and insurance companies is growing, but it’s hard to map them in real-time.
Securitisation is another layer. Cash flows from data centres are being bundled together into asset-backed security.
Bank of America reports that the US ABS market now comprises about $82 billion in digital infrastructure, a nine-fold increase in just five years.
Next year, we can expect more supply. Investors often buy rated products, but don’t always understand the assets within.
Some loans are backed up by GPUs. As newer models of chips arrive, older ones begin to lose value.
If collateral prices drop, lenders may ask for repayment or sell chips on a weaker market, driving prices even lower.
The markets are still absorbing supply. Spreads have increased. Stocks are volatile.
Leverage, timing assumptions, complex financing, and other factors are now a part of the AI economy.
This post Yes, AI boom has a problem with the balance sheet may be modified in light of new developments.
This site is for entertainment only. Click here to read more
