Free with a sign-in.
The best 8 min of 22
Market Cracks Under AI Debt Burden | The Weekly Wrap
Steve Eisman says AI-related debt and private credit are the risks to watch, while current bank credit data remain sound.
The Real Eisman Playbook · Steve Eisman
The best minutes
- 12:47Bank credit data does not yet show a broad downturn5 min
- 4:34AI data-center debt is testing confidence in the boomPro pick 3 min
Also worth watching
Skip the war and telecom updates, the France deficit discussion, and the individual bank earnings rundown.
Full analysis · 6 min read
Real Alpha · Nov 7
Steve Eisman says CoreWeave’s $35 billion-plus debt tests AI confidence
- CoreWeave had more than $35 billion in debt against a $43 billion market value, plus $10 billion in leases.
- Eisman’s warning rests on debt pricing, while large-bank credit data remains benign.
- Budget AI infrastructure for refinancing risk, and stress-test banks against a weaker deal cycle.
Free with an account
Read the rest of the analysis
6 sections, the 3 other voices it tests the episode against, and what it means for you. Every episode we cut gets one.
Check your inbox
We sent a link to your address. Click it to finish signing up. Not there in a minute? Look in spam.
or
Signed up before? The same buttons sign you in. The episode, its moments and the transcript stay free.
CoreWeave’s debt repriced before banks showed stress
CoreWeave’s market value was $43 billion. It had more than $35 billion of debt, plus $10 billion in capital lease obligations.
Its debt carried a 9 percent coupon when raised. Later, the debt traded below 90 cents on the dollar, putting its yield around 15 percent.
CoreWeave fell 15 percent that Wednesday, and Nebius fell 10 percent. The broader AI complex also declined as investors focused on debt levels.
But that price move is a signal about financing confidence, not proof that AI demand has failed. Eisman said CoreWeave competes with hyperscalers that have much larger balance sheets.
In November 2026, Steve Eisman revisited CoreWeave on The Real Eisman Playbook. He said 15 percent debt no longer made new data centers pencil out.
Our read
The debt market is pricing a specific funding vulnerability, not declaring the whole AI buildout insolvent. The next test is whether expensive refinancing spreads beyond neoclouds.
Annualized revenue headlines need a cash test
The Financial Times reported OpenAI saying annualized revenue was $50 billion as of September’s end. Eisman said OpenAI had previously put that figure at $70 billion.
He said he had no idea whether annualized meant one day’s, one week’s, or one month’s revenue multiplied across a year. The episode gives no explanation for the $20 billion gap.
That is more than a reporting detail. A run rate can turn a short observation into a yearly headline without showing whether revenue repeats.
In October, Winfield Sickles said on Inside Blackstone that hyperscaler spending would be upwards of $800 billion this year. He said probably over $1 trillion next year, and linked cloud revenue growth to that spending appearing in results.
In August, Ed Zitron said on The Diary Of A CEO that major tech firms finance AI customers. He said those customers’ spending justifies more data centers, contrasting the arrangement with AWS as unusually risky.
- WSWinfield Sickles · Inside Blackstone · Oct 13Cloud revenue is beginning to reflect AI infrastructure spending
Ed Zitron · The Diary Of A CEO · Aug 27AI infrastructure spending is fueled by circular financing
Our read
There is evidence of infrastructure spending showing up in cloud revenue, but that does not verify OpenAI’s annualized figure or CoreWeave’s refinancing capacity. The measurement period and source of customer funding matter alongside the headline run rate.
Big-bank delinquencies remain benign, within their limits
JPMorgan, Citi, Wells Fargo, and Bank of America span consumer and commercial lending. That gives Eisman a broad, though incomplete, view of credit conditions.
He focused on non-accruing loans, usually loans at least 90 days delinquent. Across the four banks, consumer and commercial figures showed no broad year-over-year or sequential deterioration.
JPMorgan’s total non-accruing loans were down 7 percent year over year and flat from the second quarter. Bank of America’s were up 3 percent year over year, but down 4 percent sequentially.
Yet those figures are not a forecast for every borrower or lender. They show that the largest banks’ broad loan books were not yet displaying the deterioration Eisman was watching for.
A calm bank report can restrain warnings of a downturn already underway. It cannot show that every lender or borrower remains healthy.
Our read
The bank data is evidence against a downturn already visible across mainstream consumer and commercial lending. Its reassurance has limits because the figures cover only part of the credit market.
Opaque private credit could carry the next losses
Eisman said private credit is exposed to software, which AI is hitting. He said future weakness could emerge there or in AI-related debt like CoreWeave.
Banks publish detailed, transparent data. Private credit is more opaque, making future weakness harder to spot.
In November 2026, Steve Eisman said on The Real Eisman Playbook that leveraged software buyouts could lose their equity cushion as valuations fall. Those companies face refinancing next year, when lenders may demand fresh equity.
That is a possible path from software repricing to private-credit stress, not evidence those losses have arrived. This episode gives no measure of private-credit exposures or defaults.
So benign bank data and potential weakness elsewhere can coexist. Treating one as a rebuttal to the other would mistake a partial window for the whole market.
Our read
The episode’s point is about where stress might appear, not a prediction that it will. Private-credit exposures deserve separate monitoring because bank disclosures cannot fully reveal them.
Agentic AI threatens margins, but not yet materially
Eisman described another risk alongside debt. Consumers could use agents to move excess cash from low-paying accounts into money-market funds, raising banks’ funding costs.
For example, his Schwab account earns little when cash is swept automatically. He can instead trade into a money-market fund and earn the interest. An agent could remove that friction and pressure net interest margins.
JPMorgan said it would not allow outside agents access to customer accounts. But that signals a possible fight over account control, not evidence that deposit outflows have started.
Take auto insurance. An agent searching by price rather than brand could intensify competition for personal-lines insurers. Eisman doubted agentic AI was yet big enough to make a material difference.
Progressive’s September report showed new premium volume declining for the first time since November 2023. Eisman said softer pricing was the likeliest cause, while investors might still speculate about AI.
Our read
The episode lays out a path from automation to weaker customer relationships, but separates that risk from current operating results. For now, measure actual cash movement and insurance pricing rather than treating the scenario as an established earnings hit.
Bank multiples still require a strong deal cycle
Banks reported strong results, largely from strong investment banking and trading. But fixed-income trading weakened at most companies as rising rates reduced client volume.
Goldman Sachs, JPMorgan, and Morgan Stanley reported returns on tangible common equity above 20 percent. Those returns help explain their richer valuations, but do not guarantee the deal cycle stays strong.
Wells Fargo, Bank of America, and Citi had returns on tangible common equity between 12 and 16 percent. Eisman linked those lower returns to lower price-to-tangible-book valuations.
Investment banking results benefited from AI financing needs and strong M&A. So a weaker deal cycle could pressure fees and the earnings supporting current valuations.
Eisman said current prices leave little room for error, and he saw no great entry point. His conclusion is valuation commentary, not a direct investment recommendation.
The episode’s Indeed and Ramp sponsor reads are separate from that analysis.
Our read
Bank earnings provide a counterpoint to AI’s funding risks because financing demand currently supports fee income. Yet the valuations Eisman described leave little room for that support to fade.
What it means for you
- Map refinancing exposure. List AI infrastructure commitments by borrower, maturity, and financing cost. Flag projects whose economics depend on refinancing near the yields Eisman described.
- Separate revenue measures. Record AI vendors’ revenue over a stated period separately from annualized run rates. Ask what period and customer activity underlie each figure.
- Check credit outside banks. Add private-credit software exposure and company refinancing dates to your quarterly risk review. Track those separately from large-bank delinquency data.
- Stress-test deal fees. Model bank earnings with weaker M&A and AI financing activity. Compare that case with the returns on tangible common equity supporting current valuations.
What to watch
- Whether CoreWeave’s debt repricing spreads to other AI infrastructure borrowers.
- Whether software-buyout companies face tougher refinancing terms next year.
- Whether agents produce measurable cash movement or changes in insurance pricing.
Send this to CFOs, treasury leaders, and teams underwriting AI infrastructure debt.
Watch the moment Eisman gives CoreWeave’s debt load
Written by Real Alpha from the episode's transcript and the other moments it links, then checked against the transcript. Quotes are the speaker's exact words.
Minute by minute
A Pro write-up, open to you here. Try Pro free for 7 days
A Pro write-up. The minute by minute is in Real Alpha Pro; the episode, its moments and the transcript stay free.
Read it with Pro, 7 days free- 2:48The telecom selloff looks overdone
Eisman says Starlink’s spectrum acquisition could improve its U.S. cellular service, but it still needs an expensive terrestrial network or a deal with one of the big three carriers. He considers the selloff in AT&T, Verizon and T-Mobile overblown, while noting that the “telepocalypse” narrative may persist.
- 4:34AI debt is testing confidence
Eisman says CoreWeave has a $43 billion market cap, more than $35 billion in debt and $10 billion in capital-lease obligations; its debt has fallen below 90 cents on the dollar, lifting its yield to around 15%. He says investors’ concerns spread across AI stocks. He also questions OpenAI’s annualized revenue claims, noting its $50 billion figure at September’s end versus an earlier $70 billion claim.
- 6:27France faces rising debt concerns
Eisman says the French-German 10-year bond spread has reached 140 basis points and France’s debt-to-GDP ratio is 120%, with no political consensus to cut spending. He says U.S. debt is also 120% of GDP, but Treasuries’ central role in the global financial system gives the U.S. more time, though not indefinitely.
- 11:57Bank credit remains benign
Eisman says large-bank credit data offer a broad, though incomplete, view of the economy, and he sees no broad deterioration. JPMorgan’s non-accruing loans were $9.4 billion, down 7% year over year and flat quarter over quarter; Bank of America’s were $5.7 billion, up 3% year over year and down 4% sequentially. He says risks may instead emerge in private credit or AI-related debt.
- 17:29Investment banking supports results
Eisman says strong investment banking and trading lifted bank results, with AI financing needs helping bring business to Wall Street. JPMorgan’s EPS rose 10%, though expenses grew 12% against 10% revenue growth; Bank of America was hurt by a 17% drop in fixed-income trading and a $1.2 billion fine. He also notes Wells Fargo’s revenue grew only 2%, while Citi’s EPS rose 53% and Goldman’s rose 39%.
- 20:59Bank valuations leave little room
Eisman says banks with returns on tangible common equity above 20% trade at 2.6 to 3.4 times tangible book, while banks earning 12% to 16% trade at 1.2 to 1.8 times. He says investing at current valuations assumes the investment-banking cycle stays strong, leaving little margin for error and making this a poor entry point in his view.
Published Oct 16, 2026.
Transcript · 16 min read
The Real Eisman Playbook, hosted by Steve Eisman.
Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there. Where the podcast’s audio and the video part ways, some times are a minute or so off.
banks, and other financials. We've had 17 years of amazingly benign credit quality, and investors are wondering if the good times are over. And this brings me to my conclusion about the state of the U.S. economy. Let's get started. Hi, this is Steve Eisman, and this is another edition of the Weekly Wrap. This is for the week ending Friday, October 16, but recorded Thursday night, October 15. I just want to repeat an announcement about a change we made recently. Monday and Friday episodes are available to premium Substack subscribers earlier. The Monday interviews are released 12 p.m. Eastern Standard Time every Sunday, and the Friday wrap is released at 12 p.m. Eastern Standard Time every Friday.
The Monday and Friday episodes will still be available on YouTube and all audio channels at their usual times. at noon, and Friday at 4.15 p.m. For Substack Premium, this Wednesday, we had returning guests Gretchen Morgensen and Tom Gober to discuss life insurance susceptibility to being defrauded. And next Wednesday, we have another returning guest, Dwight Collins, head of sales at Autonomous Research, for his thoughts and insights on the impact of AI on financial companies. On this week's wrap, we will discuss, as usual, number one, the war in the Middle East. Number two, the telepocalypse. Not the sasspocalypse, the telepocalypse. Something of a tech correction. Number four, bad news in France. Five, the impact of agentic AI on financial institutions. Six, and last, earnings season began with, of course, the banks. Let's get started.
The war is still in a stalemate. Oil and energy prices continue to influence the market on a daily basis. On Wednesday, the inflation data was worse than expected. Interestingly, though, interest rates declined on Wednesday. Why? There was negative news about CoreWeave, which I'll discuss in a bit. And that news caused tech stocks and the market to decline. And some investors purchased treasuries as a safety play despite the inflation news. Now, you've heard of the sasspocalypse. We've spoken about it many times. And now there is a new-pocalypse, and I think I've coined this term, and I call it the telepocalypse. This is not an AI story, but an Elon Musk tale. On Friday, last week, October 9, AT&T, Verizon, and T-Mobile were down 9 to 13 percent. Why? SpaceX's most profitable division is called Starlink. It is a company that uses satellites to provide cellular service all over the globe. Sounds great, but the service is not nearly as good as the service provided by telecom providers. Now, Starlink announced that it was acquiring Grain Management's 14 megahertz of 800 megahertz spectrum. Now, I don't want to get too technical, but this will definitely improve Starlink's service in the U.S. However, as best as
Read the whole transcript
I can tell, this is still not nearly good enough. Starlink needs to build a terrestrial network and that can be unbelievably expensive. The most economic way for Starlink to do this is to reach an agreement with one of the big three in the U.S. and use their rails to provide service. And none of the big three, believe me, have any interest in helping Starlink. So the telepocalypse is probably overblown. But as we know, narratives are powerful, and this one is not going away. By the way, between the big three, T-Mobile is down the most this year, down 25%, whereas Verizon is up 5%, and AT&T is down 9% this year.
I think the reason for T-Mobile's poor performance is that it used to have a high multiple. It's been a much better company than Verizon and AT&T, and so it used to have a multiple in the 20s, which was double, double the multiple of its two competitors. But now that the entire landscape has changed, T-Mobile's 2026 PE multiple is only 14 times. Moving on, and this is probably the key news of the week.
“Its market cap is only $43 billion, but it has taken on $35 billion plus in debt, and it has additional capital lease obligations of $10 billion.”
are losing confidence in France's ability to manage its deficit. The spread between the French 10-year and German 10-year bunds is at 140 basis points, a multi-year high. The problem is that France's debt to GDP is at 120%, a dangerous level. There is no consensus or desire by French politicians to cut expenditures. In fact, Marie Le Pen, who may become the next president, is advocating cutting the retirement age from 62 to 60. Of course, someone could immediately point out
that U.S. debt to GDP is also 120%. So isn't that a problem too? Well, of course it's a problem, but it's a lesser problem. Why? The dollar is the reserve currency of the world. More importantly, the entire global financial system functions on treasuries. The repo market, which is a multi-trillion dollar market for banks, only uses treasuries. When large institutions want to park money, they buy treasuries. Until there is a large, safe, and liquid alternative to treasuries, the U.S. has the luxury of time, but not endless time. In other words, no one needs to own French sovereign bonds, but many large institutions need to own U.S. treasuries. And now for the banks.
The banks kicked off earnings season, but before we get to the banks, let me just explain the concern some investors have with respect to the impact of agentic AI on banks and other financials. When you put your money in a bank, generally, you put it in a checking account that pays nothing, or a savings account that pays little. In actuality, for most people, not all people, but for most people, you only need a limited amount of cash in the bank. The extra, you could put it in a money market fund that pays more. Now, hold that thought for a second. Now, my wife and I have our investment accounts at Schwab, and we are very happy with Schwab. I have no complaints.
Now, I have some amount of cash in my account. Now, if I do nothing, Schwab sweeps my cash, pays me very little, and earns... the short-term interest rate for itself. However, if I execute a trade and buy a money market fund through Schwab with my cash, I earn the interest. Now, Schwab does not make this hard, but I have to execute the trade. The important point here is that because of consumer laziness, banks and brokers like Schwab have access to cheap sources of funding. For a typical bank, for example, checking and savings accounts are the most important parts of the bank's funding and therefore the most important sources of its net interest margin. Now let's think about agentic AI.
What investors are worried about is the idea that consumers will use agentic AI to sweep their excess cash out of banks and brokers and into nice paying money market funds. If this were to happen, it would seriously impair the net interest margins of banks. Of course, banks aren't stupid. They're going to try and protect their franchises. This week, JP Morgan announced that it would not allow outside agents access to its customer accounts. The fight begins. Now, that is the long-term risk for banks, but that risk extends to other parts of financials such as personal, auto, and homeowner's insurance. Companies like Progressive and Geico wield powerful brand names. They are very price competitive, and consumers flock to them because of their brands. Agentic AI does not care about brands. It cares about credit ratings and price. Imagine asking an agentic AI to find for you auto insurance on your new car.
Leaving aside credit ratings, agentic AI only cares about price, and this could seriously impair the franchise value of personal lines insurers. Progressive reported some numbers on Wednesday. Progressive does something no other company does, at least not that I know of. Instead of reporting quarterly, it reports monthly. And the September report was disappointing and that new premium volume declined for the first time since November 2023. Now, the most likely culprit is that pricing is soft, so Progressive is pulling back. I doubt Argentic AI has gotten big enough yet to make a material difference, but that will not stop investors from speculating.
And Progressive was down 4% on Wednesday. Now I'm going to share some thoughts on a bunch of the large banks that reported this week and then show you how to think about valuation given the results. This episode is brought to you by Indeed. You're ready to level up your business, but the pressure to find the right person to make that happen is intense. Hence, Indeed Sponsored Jobs can help carry the load. It boosts your job post to help it reach quality talent fast. Cut through the chaos with Indeed Sponsored Jobs. Get a $75 sponsored job credit at indeed.com slash podcast. Terms and conditions apply.
Hiring now? Then this is a job for Indeed Sponsored Jobs. Hi, Steve Eisman here. The most successful businesses in the world share one secret. They know exactly where every dollar goes, what to cut, and where to double down. With Ramp, you can see exactly what's happening across the business in real time. Ramp combines corporate cards, expense management, bill pay, and travel in one platform. You set spending rules up front, and Ramp corporate cards block any out-of-policy spend before it even happens. That's why some of the most valuable businesses in the world, like SpaceX, Shopify, Visa, and Stripe cut their expenses by 5% with Ramp.
A business owner I really respect uses Ramp and he says he loves it. Invest in your growth. Get on Ramp. New customers get $250 off at ramp.com slash eisman. Go to ramp.com slash eisman, ramp.com slash eisman. For large banks, the major variables in terms of revenue and earnings are fixed income and equity trading, investment banking, which is composed of M&A fees and equity and fixed income underwriting, net interest income, which is a function of loan growth and the net interest margin, and of course, credit quality statistics. All of these variables are important in understanding earnings results.
However, credit quality statistics are the most important variable in terms of drawing implications about the overall economy. To summarize, this quarter's results briefly. Credit quality remains benign. As for revenue, the overall results were good, but not as good as prior recorders. The problem was that for most companies, fixed income trading was weaker because rising rates put a crimp on client volume.
“We certainly saw large increases in consumer non-accruals leading up to the GFC. We are not seeing that now at all.”
As for the financial results of the banks this quarter, they once again posted strong results, largely because of extremely strong investment banking and trading. The banks are partial beneficiaries of the AI boom, as companies with AI financing needs are going to Wall Street for help. These results produce record levels of returns, or almost record levels of returns. Here's a brief synopsis. J.P. Morgan, earnings per share was 557 versus 507, 10% growth. Both revenue and earnings per share get this missed, which is a very, very rare event for this company. There was a lack of operating leverage as revenue was up 10%.
but expenses were up 12%. Also, the buyback was a paltry $700 million, and fixed income trading was down 4% versus last year. The return on tangible common equity was a strong 20%. Bank of America reported a poor quarter. It missed on revenue because fixed income trading declined by 17% versus last year. It also missed on earnings because of a government fine. The company paid a $1.2 billion fine because of failures with respect to anti-money laundering, a big no-no. AML is something regulators are very strict about, and it's kind of shocking that a bank of the quality of Bank of America is still having problems in this area. The market was quite shocked by this fine as the stock was down over 4% on Wednesday. Still, Bank of America's return on tangible common equity is a very respectable 15%. So long term, the company's fine. Wells Fargo has a history of occasionally disappointing in the quarter. And it did so once again this quarter. It beat on earnings, but missed on revenue. Now, I really don't understand why Wells Fargo seems to miss every other quarter or so.
I just don't get it. But they just do it. Earnings per share was $1.87 versus $1.66, up 13% versus last year. That was a beat. Revenue missed. Total revenue was up only 2% versus last year. In fact, Wells has not grown revenue in multiple quarters. The company just seems to have trouble growing its revenue. If you were interested in investing in a large bank, this is not the one I would look at. Let's move on. Under the leadership of Jane Frazier, Citi continues to report good results. The company reported earnings per share of $285 versus $1.86, up 53% versus last year, and while higher rates hurt the fixed income results of Goldman, Citi did fine. The return on tangible common equity was at 11.5%. That was lower than the recent 13% achieved in the last two quarters, but still pretty good for Citi. Right now, we are
“To invest in banks at these valuation levels requires you assuming that the investment banking cycle remains very strong.”
We discussed all the controversies surrounding AI, what's real and what's not. Next week on October 17th on our free service, we will release an interview with Danny Moses,
