Markets & FinanceDigital InfrastructureRegulatory Policy
Is the AI Boom Becoming a Credit Bubble?
AI-infrastructure debt on pace toward $570B in 2026 is reframing the AI-bubble debate as a credit-and-supervision question, with Senate Democrats pressing FSOC to probe the risk just as regulators move to deregulate the private credit that carries much of it.
- Morgan Stanley projects AI-related debt issuance more than doubling to nearly $570B in 2026, with about $236B issued by end of May.
- Senators Warren, Blumenthal, Smith, and Van Hollen asked FSOC to probe the risk in January; Warren and Blumenthal introduced the AI Bubble Transparency Act in June.
- The risk has migrated from tech valuation to bank-supervised credit via bonds, private credit, and off-balance-sheet lease vehicles.
- Moody's has flagged ~$662B in off-balance-sheet hyperscaler lease commitments and moved Oracle’s outlook to negative on its AI-contract commitments, and S&P has since cut Oracle a notch to BBB-.
- The seam is the contradiction: more supervision demanded as the OCC, FDIC, and Treasury move to deregulate private credit.
The Buildout Moved From the Equity Page to the Supervision Page
Over the first half of 2026 the money funding the AI buildout has shifted from tech equity toward debt, and the debt has grown large enough that a bank regulator, not just a stock analyst, would notice. Senate Democrats have asked the Financial Stability Oversight Council to look at it, and rating agencies have started flagging the credit strain on named borrowers. Read together, those threads move the AI-bubble debate off the equity page and onto the supervision page. That migration, from valuation risk to bank-supervised credit risk, is an analytical judgment, not a finding any regulator has issued.
- The Scale Figure. Morgan Stanley projects global AI-related debt issuance will more than double to nearly $570B in 2026, with roughly $236B issued as of May 31, about four times the same period a year earlier, per an International Finance summary of the forecast.
- The Off-Balance-Sheet Layer. Moody's has flagged that the five largest US hyperscalers hold about $662B in future data-center lease commitments that have not yet begun and sit off their current balance sheets, alongside rising leverage for 2026-28 buildouts.
- The Senate Ask. On January 22, 2026, Senators Warren, Blumenthal, Smith, and Van Hollen wrote to Treasury Secretary and FSOC Chair Scott Bessent, pressing FSOC to probe the financial-stability risk of more than $1 trillion in projected AI-infrastructure debt, and requested a written response by February 13.
- The Legislative Step. On June 11, 2026, Warren and Blumenthal introduced the AI Bubble Transparency Act, which would direct the A Treasury Department office created after the 2008 financial crisis to collect and analyze data on emerging risks across the financial system. to collect data from banks, insurers, and private-credit funds on their AI-related exposures.
- The Deregulation Backdrop. In December 2025 the OCC and FDIC withdrew Supervisory expectations bank regulators issued in 2013 that discouraged banks from lending to companies already carrying heavy debt. they called overly restrictive, a step they noted had pushed such lending toward nonbanks and outside the The boundary separating financial firms and activities that regulators supervise from those they do not., while Treasury framed the private-credit market's growth as a sign banks were overburdened.
Debt Turned an Equity Story Into a Supervision Question
When the buildout was funded with tech equity, a disappointment repriced a handful of stocks. Financed with bonds, private credit and special-purpose vehicles, the same disappointment lands on bondholders, private-credit funds, insurers and the banks connected to them. The composition of the debt stack, project finance, high yield, private credit and broadly syndicated loans, now matters as much as its size, which is why Moody's watches leverage at the largest builders. Morgan Stanley projects AI-related debt issuance more than doubling toward $570 billion in 2026, with about $236 billion issued by the end of May, roughly four times the same period a year earlier. That is no longer a handful of marquee bond deals.
Credit is the domain prudential regulators supervise, which is what gives the January 22, 2026 Senate letter its point of entry. It asks the Financial Stability Oversight Council to examine the structures that make the exposure hardest to see: off-balance-sheet lease vehicles and private-credit funds. Moody's has identified about $662 billion in future data-center lease commitments at the five largest hyperscalers in the United States, commitments that have not yet begun and sit off current balance sheets. The AI Bubble Transparency Act, introduced June 11, 2026, would compel banks, insurers and private-credit funds to report their AI-related exposures to the Office of Financial Research.
The judgment here is that the ask arrives pointed against the direction prudential policy is moving. In December 2025 the OCC and FDIC withdrew leveraged-lending guidance they called overly restrictive, noting it had pushed such lending toward nonbanks and outside the regulatory perimeter, while Treasury framed the private-credit market's growth as a sign banks were overburdened. The exposure is concentrating in the channels the supervisors are stepping back from. No formal FSOC review has opened, so the systemic characterization is a forward view rather than a finding, and the scale figures are Morgan Stanley's and Moody's rather than any regulator's.
The AI buildout is now a credit position, not just an equity bet
Markets & Finance. Morgan Stanley projects AI-related debt issuance more than doubling toward $570B in 2026, and private-credit funds have gone from near zero to well over $200B in outstanding AI-related loans, so a growing share of investment-grade issuance and private credit now sits on the buildout.
Strategy & Operations. Moody’s affirmed Oracle at Baa2 but moved its outlook to negative, and on July 9, 2026 S&P Global cut Oracle a full notch to BBB-, one step above speculative grade, citing the AI buildout and its OpenAI concentration, citing counterparty risk and the spending and commitments behind its roughly $300B in AI contracts, one named illustration of how the capex is reshaping issuer credit profiles.
Full-year 2026 AI-related issuance lands near the 2025 level rather than the projected $570B, with the largest builders funding the buildout from operating cash flow.
On the credit side: S&P cut Oracle on its OpenAI concentration, so which other issuer's rating rests on one AI counterparty, and where would that show first?
A push for more supervision meets a push for less
Regulatory Policy. Senate Democrats asked FSOC to examine opaque data-center financing, off-balance-sheet vehicles, and private-credit structures, and introduced a bill to compel AI-exposure data from financial institutions.
Markets & Finance. At the same moment, the OCC and FDIC withdrew leveraged-lending guidance and Treasury framed private credit as a sign of over-regulation, so the demand for supervision of AI debt runs directly against the direction prudential policy is moving.
FSOC opens a formal review of AI-infrastructure or private-credit risk, or the OCC and FDIC replace the withdrawn leveraged-lending expectations. Either puts both pushes in one direction.
On the supervision side: most of the exposure sits in private credit and nonbank channels, so which agency can see it at all if the transparency bill never moves?
The collateral is capital-intensive infrastructure whose revenue lags its spend
Digital Infrastructure. The debt is secured against data-center capacity whose buildout cost runs years ahead of the tenant revenue meant to service it, which is the operational fact under the credit concern, and litigation has already begun over how some of it was financed and disclosed.
Markets & Finance. Conservative estimates from Morgan Stanley and Moody's put multiyear AI data-center capital spending at $3T or more, a scale at which refinancing and rate sensitivity become macro-relevant rather than firm-specific.
A large data-center borrower shows contracted take-or-pay revenue covering site debt service from the first year of operation. The spend would not be running years ahead of the revenue.
On the operations side: servicing turns on tenant durability, utilization and power at each site, so where would a bondholder find any of those disclosed?
The Weave maps a single development across domains and across time. Each row follows one domain from where things stand now through the next eighteen months, and expands for the reasoning behind that trajectory.
- The exposure sits in a private-credit market too young to have run through a full credit cycle.
- Concentration on one asset class means many lenders end up holding the same risk under other names.
- A lightly regulated channel leaves supervisors little record to read when the cycle finally turns.
- Financing a fast-moving compute cycle on infrastructure terms locks it into slow-moving debt.
- Once the buildout is funded like infrastructure, credit conditions gate expansion as much as demand.
- Credit cycles now sit beside chip and power supply as a limit on how fast this layer can grow.
- The gap between where the risk sits and who watches it is becoming a standing governance choice.
- Declining to look is itself a decision, and it fixes the supervisory perimeter for the next cycle.
- A letter can be ignored, but a reporting mandate would outlast whoever runs the agencies next.
v1 · July 6, 2026The Rating Action Landed. The Supervision Question Has Not
The seam resolves into observables over the next several weeks and months. Whether FSOC acts, whether the transparency bill moves, and whether the credit strain shows up in a rating action or a refinancing will show up in a handful of checkable signals.
- The FSOC Posture. Whether FSOC opens any formal review of AI-infrastructure or private-credit financial-stability risk, or declines to, after the Senate ask and Treasury's non-commitment to the January letter. On July 6, 2026, NOTUS reported a leaked internal Treasury draft warning that AI-sector risk could ripple into banks and private credit, which, if the reporting holds, would suggest the supervision concern this piece tracks is surfacing inside Treasury itself, though Treasury leadership calls the draft unvetted and it is not stated policy.
- The Transparency Bill. Whether the AI Bubble Transparency Act advances, gets a hearing, or draws Republican co-sponsors, the concrete test of whether the disclosure demand has legs.
- A Marquee Rating Action. Resolved, and it tightened. On July 9, 2026 S&P cut Oracle a full notch to BBB-, one step above speculative grade, the first hard downgrade rather than an outlook change on a large AI-infrastructure borrower.
- The Private-Credit Channel. Whether the projected wave of private-credit data-center lending materializes, and whether any early stress surfaces in a fund or a deal.
- The Deregulation Path. Whether the OCC, FDIC, and Fed proceed with steps that narrow private-credit and nonbank oversight, widening the gap between the exposure and its supervision.
Every issue
- 01Intersection of the week
- 02Impact of the week
- 03The week in review
Wiiver Weekly
One free email, Saturdays at 7:00 AM ET.
Unsubscribe anytime.
Primary sources2
- U.S. Senate Banking Committee (Minority)Warren, Colleagues Press FSOC to Launch Probe into Financial Stability Risks of AI Debt BubbleJan 22Primary · A push for more supervision meets a push for less · Why is this a supervision story, not just a market one? · Looking Forward · Where Things Stand
- U.S. Senate Banking Committee (Minority)Warren and Blumenthal Introduce the AI Bubble Transparency ActJun 11Primary · A push for more supervision meets a push for less · Why is this a supervision story, not just a market one? · Looking Forward · Where Things Stand
Secondary sources, by sector10
- NOTUSTreasury Internal Report Warning of the Dangers of an AI BubbleJul 6Secondary · Looking Forward
- International FinanceAI-related debt issuance up for massive surge in 2026, says Morgan StanleyJun 12Secondary · The AI buildout is now a credit position, not just an equity bet · Why is this a supervision story, not just a market one? · Where Things Stand
- Moody's RatingsData centers: credit risk in the AI buildoutJun 4Secondary · The collateral is capital-intensive infrastructure whose revenue lags its spend · Why is this a supervision story, not just a market one? · Where Things Stand
- Investing.comOracle's credit ratings affirmed amid AI infrastructure expansionJan 15Secondary · The AI buildout is now a credit position, not just an equity bet · Why is this a supervision story, not just a market one? · Looking Forward
- Banking DiveOCC, FDIC scrap Obama-era leveraged lending guidanceDec 8Secondary · A push for more supervision meets a push for less · Why is this a supervision story, not just a market one? · Looking Forward · Where Things Stand
- Quinn EmanuelClient Alert: Emerging Litigation Risks in Financing AI Data Centers BoomJun 1Secondary · The collateral is capital-intensive infrastructure whose revenue lags its spend · Why is this a supervision story, not just a market one?
- Advisor PerspectivesAI's Trillion-Dollar Debt Binge Fuels Century-Old Private MarketJul 1Secondary · The AI buildout is now a credit position, not just an equity bet · Why is this a supervision story, not just a market one? · Looking Forward
- Insurance JournalThe $3 Trillion AI Data Center Build-Out Becomes All-Consuming for Debt MarketsFeb 3Secondary · The collateral is capital-intensive infrastructure whose revenue lags its spend
- Seeking AlphaUS Treasury draft warns of economic mayhem if AI mimics dotcom bubble: reportJul 6Secondary · Looking Forward
- The Next WebThe Next WebSecondary ·