W.090The Fed’s Bowman says AI-related debt is on track to exceed $800B this year, with potential lease and guarantee payments estimated at more than three times the debt. Three times $800B is $2.4T; five hyperscalers report $1.30T of leases that have not started
Federal Reserve Vice Chair for Supervision Michelle Bowman said on November 10 that AI-related debt is on track to exceed $800B this year, and that potential payments on leases, contracts and residual value guarantees are estimated at more than three times the debt. Three times $800B is $2.4T; she gave no dollar figure and no source. Five hyperscalers report $1.30T of leases signed but not yet started. She also attributed roughly a quarter of above-target core inflation, about 0.3 points, to AI-related price pressures.
What she said
A Fed governor put a number on AI debt and a multiple on what sits beside it. Speaking at the Council on Foreign Relations on November 10, Bowman said AI revenues are rising quickly but do not yet cover the cost of the models and data centers, and that hyperscalers and others are increasingly raising money from credit and equity markets. AI-related debt, she said, has risen rapidly in corporate bonds, private credit and project finance and is on track to exceed $800B this year. She added that other obligations, in the form of long-term lease and contract commitments and residual value guarantees, have also surged, and that their “potential payments are estimated at more than three times the amount of debt.” Separately, she said hyperscalers and large technology firms have assumed off-balance-sheet liabilities that smaller, riskier companies use as collateral to finance data centers and computing equipment.
The speech does not say what the $800B counts or where it comes from. The sentence does not say whether it is debt issued in 2026 or debt outstanding, and neither figure carries a footnote. Three times $800B is $2.4T; that multiplication is ours, and it assumes “the amount of debt” means the $800B. Bowman is the Fed’s vice chair for supervision. She said the largest bank holding companies have some exposure to AI-related financing, which the annual stress tests scrutinise, and that life insurers, pension funds, private credit and mutual funds are exposed too. Most are well capitalised, she said, but the structures make counterparty risk hard to assess.
Bowman’s $800B is above the bank counts we found
Counts of AI debt in 2026, $ billions. Solid bars are amounts issued so far in 2026 as of the report date shown; dashed bars are full-year forecasts; Bowman’s (pink) is “on track to exceed” for the year. The counts differ in scope: Bowman’s covers corporate bonds, private credit and project finance; Goldman’s and Morgan Stanley’s are AI-linked or AI-related issuance as those banks define it; the last two rows cover only the largest companies. The bank figures are as reported by the outlets listed in the sources.
How it compares
The nearest count we found is Goldman’s, and it is a count of issuance. Goldman Sachs strategists had tallied about $500B of AI-linked debt issued in 2026, as reported on September 30, and expected more than $750B by the end of the year, according to a summary of the Financial Times’s report. Bowman’s figure is 7% above that and 40% above the roughly $570B Morgan Stanley forecast for 2026, as Forbes reported it in July. The closeness to Goldman’s number may suggest Bowman’s is also issuance during 2026, though the speech does not say. Hyperscaler bonds, the part W.060 covered, were $296B by early November on Goldman’s count, 37% of $800B.
Leases that have not started are the one item of the kind she described that all five hyperscalers disclose the same way. Microsoft, Meta, Oracle, Amazon and Alphabet report $1.30T of leases signed but not yet commenced in their latest quarterly filings, 2.4 times the $547B of debt on their balance sheets (W.007). That is 54% of $2.4T. Bowman’s multiple also covers contract commitments and residual value guarantees. Meta’s maximum exposure under residual value guarantees for data center ventures was $41.5B at September 30, up from $28.1B at the end of 2025, and Alphabet’s backstops for partners’ lease and energy payments had a notional maximum of $48B, up from $16.9B. Nikkei Asia counted about $3.2T of leases and purchase commitments at the five on November 3.
Three times $800B is $2.4T; five hyperscalers disclose $1.30T of leases not yet started
$ billions. The dashed bar multiplies Bowman’s two figures and is a lower bound on them, since she said the debt is on track to exceed $800B and the payments are more than three times it. The second bar stacks leases signed but not yet commenced at Microsoft, Meta, Oracle, Amazon and Alphabet, in that order, from their latest 10-Q (September 30, 2026; Oracle August 31); the third is the debt on the same five balance sheets. Nikkei Asia’s count adds purchase commitments to leases. Anthropic’s committed spend is the customer’s side of contracts that are revenue backlog for the clouds, and some of the leases above serve it, so the rows are not additive.
Leases not yet commenced and debt, by company
| Company | Leases not yet commenced ($B) | Debt on the balance sheet ($B) | Leases ÷ debt |
|---|---|---|---|
| Microsoft | 364.6 | 35.5 | 10.3× |
| Meta | 355.7 | 80.9 | 4.4× |
| Oracle | 288.0 | 125.3 | 2.3× |
| Amazon | 164.9 | 180.0 | 0.9× |
| Alphabet | 130.5 | 125.1 | 1.0× |
| Five companies | 1303.8 | 546.8 | 2.4× |
Inflation and rates
Bowman attributed about a quarter of excess core inflation to AI. Over the 12 months to September, she said, tariffs, energy and AI-related price pressures each accounted for roughly one-fourth of above-target core PCE inflation, with the rest likely reflecting earlier shocks. Core PCE rose 3.1% over that period, according to the Bureau of Economic Analysis, 1.1 points above 2%, so a fourth is about 0.3 points. She described the AI pressure as demand for specific equipment: prices of memory chips, semiconductors, computers and other electronics have risen sharply this year, in a few categories by double digits. She estimated core PCE at 3.3% in October and said monetary policy has limited influence on AI-related price pressures.
She did not call for a rate rise, and said higher rates appear to be holding back investment at some newer and medium-sized firms in AI-intensive sectors. Bowman said she supported holding the federal funds target range at 3.75% to 4.00% in October and that the benefit of further tightening has to be weighed against its costs. She said the cost of financing AI investment has been rising notably. She added that some reassessment of expected returns may already have played out in recent weeks, as corporate bond spreads widened and credit default swap prices rose for some companies in the buildout. If AI revenues fall short of sharply rising costs, she said, the result could be a sharp correction in AI firms’ equity prices, higher financing costs and reduced access to credit, which would slow the buildout. The speech came the day the October consumer price report showed prices up 0.4% on the month and 0.3% excluding food and energy. Futures put the chance of a quarter-point rise in December at about 85% to 90.5% that day, with the rest on a half point, according to AFP and TheStreet.
A fourth of the 1.1 points above target is about 0.3 points
Core PCE inflation over the 12 months to September 2026, percent, split as Bowman described it: the 2% target, then roughly one-fourth of the excess each for tariffs, energy, AI-related price pressures (pink) and the persisting effects of earlier shocks. The 3.1% is the Bureau of Economic Analysis’s figure; the equal steps of 0.275 points are our division of the 1.1-point excess by four, and Bowman said “roughly.”
What this changes on 06.ai
- W.060 and the hyperscaler funding pack count hyperscaler bonds. Bowman’s $800B is a wider count that adds private credit and project finance; hyperscaler bonds on Goldman’s count are 37% of it. Lead: a series for AI debt across all three markets, with Goldman’s AI-linked tally as a starting point.
- W.007 put leases not yet commenced at 2.4 times the five hyperscalers’ debt. Bowman’s “more than three times” is a different ratio, of lease, contract and guarantee payments to AI-related debt. Lead: track residual value guarantees and backstops alongside leases; Meta’s and Alphabet’s both rose sharply this year.
Sources and method
| Item | Figure | Source |
|---|---|---|
| Bowman | AI-related debt, “with notable gains in corporate bonds, private credit, and project finance”, on track to exceed $800 billion this year; potential payments on long-term lease and contract commitments and residual value guarantees estimated at more than three times the amount of debt; AI revenues not yet sufficient to cover costs; off-balance-sheet liabilities; the cost of financing rising; risks concentrated in a few large firms | Federal Reserve, speech at the Council on Foreign Relations, Nov 10 |
| Bowman | Tariffs, energy and AI-related price pressures each roughly one-fourth of above-target core PCE inflation over the 12 months ending in September; core PCE likely 3.3% in October; about 25% of businesses and roughly 10% of workers report using AI daily; supported holding the target range at 3-3/4 to 4 percent in October | Federal Reserve, same speech |
| Core PCE | Core PCE price index up 3.1% over the 12 months to September 2026 | Bureau of Economic Analysis, Oct 29 |
| Fed | Target range held at 3.75% to 4.00% on Oct 28, by 10 votes to 2; both dissenters preferred a quarter-point rise | Federal Reserve, FOMC statement, Oct 28 |
| CPI | October CPI up 0.4% on the month and core up 0.3% | Bureau of Labor Statistics, Nov 10 |
| Rates | Futures put about 85% on a quarter-point rise in December and the rest on a half point; economists had expected 0.3% for the headline index | AFP, via Yahoo Finance, Nov 10 |
| Rates | CME FedWatch: 90.5% on a quarter-point rise in December, 9.5% on a half point; Waller and Logan remarks | TheStreet, via Yahoo Finance, Nov 10 |
| Goldman | About $500B of AI-linked debt issued so far in 2026; expected to exceed $750B by the end of 2026 and to approach $1.2T in 2027 | AI Weekly, on the FT’s report of Goldman’s tally, Sep 30 |
| Morgan Stanley | Forecast of about $570B of global AI-related debt issuance in 2026 | Forbes, Jul 17 |
| JPMorgan | Michael Cembalest’s estimate: roughly $320B of debt issued so far in 2026 by five hyperscalers and Nvidia, including vehicles they stand behind | 24/7 Wall St., Sep 10 |
| Goldman | $296B of hyperscaler bonds this year to early November, against an estimate of $250B for the year | W.060 |
| Leases | Leases not yet commenced: Microsoft $364.6B, Meta $355.7B, Oracle $288.0B, Amazon $164.9B, Alphabet $130.5B; balance-sheet debt $547B in all; Nikkei Asia’s count of about $3.2T | W.007, from each company’s 10-Q |
| Meta | Leases not yet commenced $355.73B; maximum residual value guarantees for data center ventures $41.493B, from $28.091B at Dec 31, 2025 | Meta 10-Q, Sep 30 |
| Alphabet | Backstops for partners’ data center lease and energy payments, notional maximum $48B, from $16.9B at end-2025 | The Next Web, on the 10-Q |
| Anthropic | $900.9B aggregate committed spend | W.084, from the S-1/A of Nov 10 |
Bowman’s figures are from the text of her speech; they are projections or estimates and carry no source. Three times: 3 × 800 = 2,400, a lower bound on her two figures. Leases: 364.6 + 355.73 + 288 + 164.922 + 130.5 = 1303.8; debt 35.481 + 80.922 + 125.337 + 179.952 + 125.1 = 546.8; 1303.8 ÷ 546.8 = 2.38; 1303.8 ÷ 2,400 = 54.3%. Amazon’s debt is long-term debt of $173.125B plus the $6.827B current portion. Comparisons: 800 ÷ 750 = 1.067; 800 ÷ 570 = 1.404; 296 ÷ 800 = 37.0%. The bank tallies are taken from secondary reports and differ in scope and date from Bowman’s figure and from each other. Inflation: 3.1 − 2.0 = 1.1 points; 1.1 ÷ 4 = 0.275. The Fed’s 2% target is defined on headline PCE inflation; Bowman applied it to core. Anthropic’s commitments and the hyperscalers’ leases overlap and are not summed.
Wire reports are written within hours of the news, on a six-hour schedule. Each lists its sources and shows the arithmetic behind its figures. Not investment advice.