Out of the ashes of the 2008 financial crisis, shadow banking has skyrocketed in the past 18 years. Global shadow banking assets have surpassed $256 trillion, representing 51% of all global financial assets. The sector is growing at double the rate of traditional banks and now represents 10% of total U.S. bank loans.
Shadow banking, a network of non-bank financial institutions (NBFI) that provide similar services to traditional commercial banks, is not beholden to government banking rules. Known also as nondepository financial institutions (NDFI), shadow banks were a driver of the 2008 financial crisis, and they took over ground banks vacated afterward, when newly placed regulations restricted what traditional banks could do.
Initially, the NBFI sector provided capital to mid-sized companies that were unable to access money from traditional banking. While this was the main function of NBFIs in the aftermath of 2008, the sector has grown exponentially and, with that growth, created market distortions and risks that can’t be easily understood due to its opaque nature.
Traditional banks face stricter rules, but skirt them by feeding money into private credit and investment funds. Loans from traditional banks to NDFIs (private credit, hedge funds, private equity vehicle loans), have gone from $56.3 billion in 2010 to nearly $1.5 trillion in 2026, a 2,518% increase.
Banks Found the Side Door
There is growing concern that this arrangement has created a hidden feedback loop that could cause or amplify the next financial crisis. The FDIC’s 2026 Risk Review flags the scale of NDFI exposure and warns that in a downturn, nonbanks may be forced to sell assets, pushing down valuations on collateral pledged to banks.
Per the FDIC’s Risk Review:
“For banks with assets greater than $10 billion, about 57 percent of NDFI lending was to credit intermediaries in the fourth quarter of 2025.”
AI Built a Debt Machine Off the Books
Big Tech, having traditionally relied on its own deep pockets, is now increasingly relying on debt to pay for AI infrastructure. According to UBS, hyperscaler capex in 2026 is expected to use up 100% of cash flows. Morgan Stanley forecasts that AI-derived debt will exceed $570B in 2026.
However, a good portion of that debt appears off the balance sheets of Big Tech. A recent study by Nikkei found that five of the largest U.S. hyperscalers (Alphabet, Meta, Microsoft, Amazon, and Oracle) carry $1.65 trillion in off-balance-sheet obligations.
That means when any of these companies report quarterly earnings, they’re not disclosing this debt as balance-sheet liabilities, even though the off-balance-sheet debt is collectively larger than the $1.35 trillion of debt that does appear on their balance sheets.
Special Purpose Vehicles (SBV) are the mechanism for this trend. SPVs, acting as a regulatory loophole, partner big tech with private credit as a separate, legally independent company. Arranged in this manner, the SPV shields the hyperscalers from liability, isolates financial risk, while also obscuring it on their balance sheets.
Leverage Upon Leverage
Given the opaque nature of shadow banking, there are growing fears concerning the multiple levels of leverage being applied. Operating companies are leveraged, while the private credit funds that finance them are leveraged, while the traditional banks are also leveraged.
If AI revenue takes longer to materialize, the data centers those SPVs own lose value, and the collateral backing their bank loans loses value with them. Interest rates would then rise as liquidity dries up. At that point, it would become an issue for the broader market.
The bet is that AI revenue will grow fast enough to pay the debt. However, there aren’t really good numbers to support this. If the revenue doesn’t come, the massive off-balance debt that’s hidden in the footnotes will have its time in the spotlight.
To assuage the fears of private credit, Big Tech often offers residual value guarantees (RVGs). Essentially, an RVG ensures that if a SPV’s physical asset (i.e., a data center) depreciates below a certain threshold, Big Tech absorbs the cost.
This distinct eventuality is not represented on balance sheets, despite the fact that it’s widely known how quickly the GPUs powering data centers depreciate. Given Big Tech’s rampant spending, credit ratings of some companies are being questioned. S&P has already adjusted Oracle’s credit rating to BBB, two notches above “junk.” Other companies’ ratings could follow if debt patterns continue.
Washington Held the Door Open
Since returning to office, President Trump has pursued a broad deregulatory agenda, easing capital requirements and narrowing banking oversight. Lending to the shadow banking industry has soared since.
Traditional banks have rapidly increased their exposure to NBFIs because they get a higher return. This is due to lower capital requirements, as compared to direct commercial & industrial (C&I) loans.
Regulators require banks to fund a set share of every loan with their own capital, scaled to how risky the loan is. The capital requirements for loans to NBFIs are substantially lower than C&I loans (20-30% as opposed to 100%). This is an enormous difference and hugely profitable for banks.
Financing the Competition
In 2025, Moody’s reported that risk was rising for traditional banks, particularly for smaller banks. NBFIs are eating traditional banks’ lunch. As a result, banks are simultaneously financing and competing with nonbanks.
According to Moody’s, this extremely competitive dynamic could result in the weakening of underwriting standards, while elevating credit risk. Traditional banks’ exposure to private credit tends to be heavily concentrated, which runs a risk when “private credit instruments are illiquid and opaque, and have only internally managed valuations.”
The repeated gripe among analysts and regulators is how difficult it is to judge shadow banking’s effect on the economy due to a lack of data. In the AI sector, there is a certain irony to this. AI thrives on data, and yet no one has good data on exactly who’s ultimately paying for all these data centers, or what happens if they fail.
No Data on the Data Centers
Every layer of this arrangement was built to move risk somewhere the rules do not reach. Banks moved lending into nonbanks because the capital charge is cheaper. Big Tech moved data centers into SPVs because the debt does not show up where investors look. Private credit funds took the exposure because the yields beat anything on offer elsewhere. Each step was rational for the party taking it, and each step made the whole structure harder to see.
That is the part regulators keep circling back to. The FDIC can count the loans. The FSB can total the assets. Neither can tell you what happens when the collateral behind $1.47 trillion in bank exposure turns out to be racks of chips whose useful life nobody agrees on, sitting in buildings leased to companies whose AI revenue has not yet materialized.
The bet is that revenue shows up before the bill does. If it does, this reads as an accounting footnote. If it does not, the losses will surface in exactly the places the structure was designed to keep quiet, and the people holding them will be the same depositors and pensioners who were told in 2010 that the system had been fixed. The industry that promised to make sense of the world’s data cannot produce a straight answer about who is financing it.
Author: Tim Tolka, Senior Reporter
The editorial team at #MRKT3.0 has taken all precautions to ensure that no persons or organizations have been adversely affected or offered any sort of financial advice in this article.
See Also:
Slash’s AI Banker Can Now Move Money Without You. What Could Go Wrong? – MRKT3.0
AI agents raise new fraud challenge for UK banks, BioCatch survey finds – MRKT3.0
