Wall Street’s biggest banks spent fifteen years learning to hold more capital, stress-test their books and stay clear of the riskiest corners of lending. They are now moving back toward those corners, and the doorway is a record wave of credit aimed at artificial intelligence. Morgan Stanley expects investment-grade corporate bond issuance to reach a record near $2 trillion in 2026, much of it earmarked for AI, while the Financial Stability Board (FSB, the global body that monitors the financial system for the G20) warned on May 6 that the links between banks and private credit are deepening fast.
Here is the uncomfortable part of that arithmetic. The rules written after the last crisis pushed the riskiest lending out of banks and into private credit, the non-bank lenders that fill the gap. Now demand for credit is so large that the two sides are blending back together, and the people running the banks concede they cannot fully see what they are buying.
The Rule That Pushed Risk Off Bank Balance Sheets
The centerpiece of the post-2008 overhaul was the Volcker Rule, part of the 2010 Dodd-Frank Act. It curbed proprietary trading and forced lenders to carry thicker capital cushions, run regular stress tests and accept tighter supervision. For the most part it did the job. There has been no repeat of 2008, and the banking industry has stayed strikingly profitable through the whole stretch.
What the rules did not do was kill the appetite for risk. They relocated it. The demand for risky lending never went away, so the supply migrated to creditors who sit outside the bank rulebook. That migration is the entire origin story of private credit, a market built to hold the loans regulators no longer wanted on insured balance sheets.
For more than a decade that division of labor held. Banks stayed inside their guardrails, and the riskier money lived somewhere a depositor never had to think about. You can read the Federal Reserve’s summary of the Volcker Rule restrictions and see the intent in plain language. The intent and the outcome have started to drift apart.
A Record $2 Trillion Credit Wave Aimed at AI
The supply of credit right now is enormous. Morgan Stanley’s forecast of close to $2 trillion in investment-grade issuance for 2026 would be an all-time high, and the bank estimates AI-linked debt issuance alone could more than double to roughly $400 billion this year. JPMorganChase has separately flagged that about $1 trillion of corporate debt is due to be refinanced over the same span.
Stack those figures against the longer arc of AI spending and the scale gets harder to wave away. Bloomberg Intelligence projects combined investment in AI, cloud and data-center infrastructure could reach $3 trillion by 2029, with debt increasingly the funding tool of choice. Even the hyperscalers, the handful of cloud giants that generate cash like utilities, are tapping bond markets to cover capital spending they once paid for out of pocket.
Some of that demand lands with the banks. A growing share lands with private credit. The reason matters: this is not a normal corporate borrowing cycle spread across thousands of issuers, it is a concentrated bet on one industry’s future cash flows.
- $2 trillion in projected 2026 investment-grade bond issuance, a record, per Morgan Stanley.
- $400 billion in potential AI-linked debt issuance this year, more than double the prior pace.
- $1 trillion of corporate debt set to be refinanced in 2026, per JPMorganChase.
- $3 trillion in cumulative AI and data-center investment projected by 2029.
Trade-body data tracks how heavy this cycle already is; the U.S. corporate bond issuance statistics published by SIFMA show issuance running near the top of its historical range well before the AI build-out peaks.
Where Banks and Private Credit Now Touch
The May 6 FSB study put numbers on the blending. Private credit has swelled to an estimated $1.5 trillion to $2 trillion in assets, heavily concentrated in a few jurisdictions, and the report describes an ecosystem where banks, asset managers, insurers and private equity firms are increasingly stitched together through financing lines and strategic partnerships.
The $220 Billion Banks Can See
Across FSB member jurisdictions, the available data captures around $220 billion of drawn and undrawn bank credit lines extended to private credit funds. That is the measured exposure, the part regulators can actually total up. Commercial estimates run far higher, between $270 billion and $500 billion, which tells you how wide the error bars are even on the visible piece.
The Exposure They Cannot
Beyond the credit lines sit the indirect channels: leverage stacked inside the funds, liquidity mismatches, and cross-border ties that no single supervisor sees end to end. The FSB warned that this layout poses problems for the banks themselves.
| Measure | Figure | Source / note |
|---|---|---|
| Private credit assets | $1.5T to $2T | FSB, end-2024 estimate |
| Measured bank credit lines to funds | ~$220B | FSB, drawn and undrawn |
| Commercial estimate of bank exposure | $270B to $500B | Industry estimates cited by FSB |
| Projected AI-linked debt issuance, 2026 | ~$400B | Morgan Stanley |
The FSB’s blunt phrasing is that fragmented oversight makes the danger harder to spot. You can read the warning in the FSB’s report on vulnerabilities in private credit, which notes that complexity, leverage and interconnectedness could amplify stress in an adverse scenario. The same anxieties about loan recovery and provisioning are surfacing in bank earnings elsewhere, as Canada’s lenders showed when Scotiabank leaned on wealth management while credit risks loomed.
The Transparency Gap Bankers Will Admit To
Bankers are not pretending otherwise. Asked in March how well lenders understand the risk they are absorbing in private credit, M&T Bank chief executive Rene Jones was candid about the limits of his own visibility.
One of the concerns is that we don’t have that full transparency, as much as we would like. And so we have to be cautious as we move in that direction.
That is a striking admission from someone whose job is to price risk for a living. It raises a question the FSB report circles without resolving: how cautious can a bank really be about an exposure it cannot fully measure? The bargain after 2008 assumed the riskiest credit would sit where regulators could watch it. The honest answer from inside the industry is that a chunk of it now sits where almost nobody can.
The AI Question Sitting Under the Debt
All of this would be a manageable worry if the borrower at the center were a sure thing. The borrower is the AI build-out, and its economics are still unproven. PIMCO’s analysts have laid out just how stretched the cash math has become for the companies driving the spending.
Capex Is Swallowing the Cash Flow
By PIMCO’s estimate, capital spending will absorb about 94% of the hyperscalers’ operating cash flow in 2026, up from less than half just two years earlier. Consensus capex for the five largest hyperscalers has climbed toward $690 billion for this year, and their index-eligible new debt issuance has already topped roughly $136 billion, surpassing the full prior-year tally. When internal cash no longer covers the build, the gap gets bridged with borrowed money.
The Questions Nobody Has Answered
The open issues are not technical footnotes. They are the whole investment case, and they remain unsettled:
- How much real, paying demand for AI services will actually materialize.
- How large the ongoing cost of running these data centers turns out to be.
- How much revenue any of these companies can generate, and whether it is enough to service trillions in loans.
PIMCO’s framing of the risk, set out in its analysis of AI credit expansion risks, lands on a single sentence that should give every lender pause: whether these liabilities will ultimately be justified by future earnings growth remains the central question. The money is being committed now. The proof of repayment comes later, if it comes.
Why 2010’s Bargain Gets Tested Now
The post-crisis rules have not been repealed so much as worn down. Capital requirements have been eased at the margins, supervisory pressure has softened in places, and demand for credit has grown large enough that the wall between regulated banks and unregulated private lenders has become porous. The model asked banks to stay risk-minimized. The market is asking them to take on a great deal of risk. Both cannot be fully true at once.
When the Volcker Rule was being written, American Banker noted that a new systemic-risk oversight body was a welcome idea whose success or failure would be hard to judge until the next crisis. Sixteen years on, the framework has never faced its real test. A historic surge of lending, concentrated in an industry full of unanswered questions, is exactly the kind of setup that produces one.
If AI revenue arrives on schedule and the loans get repaid, the blending of banks and private credit looks like efficient capital doing its job. If the earnings lag the liabilities, the opacity the FSB flagged stops being a monitoring problem and becomes a transmission line, and the answer to whether banks can be both safe and adventurous arrives the hard way.
