Steve Eisman — Google's Negative Cash Flow and the AI Capex Reckoning

GE Vernova makes the large gas turbines power companies use to generate electricity, plus the electrical gear utilities need to move that power around. Only three companies on earth build turbines this big — GE Vernova, Mitsubishi and Siemens — which is why the AI data-center build-out has to come through them. Spun out of GE in April 2024 at $143, the stock is now above $1,000.

This quarter the headline earnings number missed, and the raised guidance wasn't as high as the most optimistic private forecasts, so the stock fell. Eisman thinks that's the wrong thing to look at: turbines are sold years ahead of delivery, so orders — what customers have newly committed to buy — are the real gauge. Orders were up 88% to $24.2 billion and the order book (backlog) is $176 billion. He owns it and, "even at these nosebleed elevations," remains confident.

Moody's rates the creditworthiness of bonds. Almost every large borrower needs a rating to sell debt, and effectively only two firms provide them — Moody's and S&P — so it's a duopoly that can raise prices year after year. Eisman has owned it for years for exactly that reason.

Earnings rose 31%, yet the stock is down slightly this year because some investors think AI could take over credit analysis and break the duopoly. He simply doesn't believe that's possible. Meanwhile the share price has been flat since late 2024 while profits kept growing, so you now pay under 30× next year's earnings — "the cheapest the stock has been in quite some time." A good business getting cheaper is his set-up.

ServiceNow sells the software big companies use to run their internal workflows — IT tickets, HR requests, service processes. Eisman calls it "one of the best software companies in the world," and this quarter it grew earnings 10% and revenue 24% with "no weaknesses in these numbers" and no sign AI is stealing its business.

The stock is still down 38% this year and fell another 4% on the print, because investors have decided as a group that AI will eat business software (the "SaaSpocalypse"). His line — "fighting the AI narrative is like fighting a ghost" — is the practical lesson: when a narrative owns a sector, good results stop being rewarded. He treats the company as fine and the tape as the problem.

Lockheed builds fighter jets, missiles and missile-defense systems for the US and allied governments. When wars escalate, orders follow — and this quarter showed it: sales up 11%, earnings well ahead of expectations, and an order book at a record $230 billion, up 24% in just three months.

That backlog is the point. It is revenue already contracted for future years, so a war-driven spike shows up in the backlog long before it shows up in profits. "Clearly, the geopolitical situation is benefiting defense companies."

Northrop is the other defense prime he checked this week — bombers, missiles, space systems. Its quarter was good on paper: earnings beat, revenue beat, order book up $20 billion to a record $105 billion, guidance raised.

The stock still didn't rise, and Eisman explains why: the cost overruns on its missile programs "seem open-ended," and the company absorbs those costs, squeezing today's profit margins. Same war tailwind as Lockheed, but a company-specific problem eating the benefit — hence a neutral read rather than a positive one.

Intel designs and manufactures computer chips. Its quarter was outstanding: a 42-cent profit versus a 10-cent loss a year ago, the fastest revenue growth in 15 years, and data-center chip sales up 59%.

Eisman's caveat is about what it proves, not whether it's good. Intel's customers are the same companies spending record sums on AI data centers, so a strong Intel quarter is the expected consequence of those budgets: "given CapEx budgets, it would be surprising if Intel did not have a good quarter." The market's worry isn't whether the suppliers are getting the money — it's whether the spenders can keep spending it. So the print won't "quell AI nervousness."

In a hard stretch for carmakers, GM "has been executing well": profit per share rose 41% and beat expectations comfortably, and management raised its full-year profit forecast — so the stock went up.

The blemish is what's underneath: US vehicle sales actually fell versus last year, including the big pickups and SUVs that generate most of GM's profit. Growing earnings on shrinking volume works through cost control and pricing, but it isn't the same as growing demand — which is why he flags it rather than celebrating.

Blackstone is the largest private-markets manager: it raises money from institutions and wealthy investors, buys companies, real estate and loans with it, and earns management fees plus a share of profits. The quarter beat expectations, helped by deal activity and by selling assets at better prices than expected, and it raised $68 billion of new money.

Eisman's two unresolved worries are structural, not quarterly. First, the time it takes to sell portfolio companies and hand investors their cash back "keeps lengthening" — money stays locked up longer. Second, private credit (direct lending, much of it to software companies) only really faces its test when those loans must be refinanced next year. He also notes management boasting that Blackstone is the biggest financier of AI data centers: "that may or may not prove to be a great bet."

S&P Global is the other credit-rating giant — the second half of the duopoly Eisman likes. It comes up only as the partner in that structure: some investors fear AI will erode the two firms' grip on ratings, and he says flatly he doesn't think that's possible. Referenced, not rated.

BNY Mellon is a "trust bank" — instead of lending to consumers and companies, it holds and administers other institutions' securities, settles trades and keeps the records. That makes it a fee business tied to asset levels and transaction volumes, not a credit business.

Eisman's honest answer to the mailbag is that he has never studied it: trust banks "operate in a world all their own" and he's "never found that much value to focusing on them." It has done well — up over 30% this year — but so has every large bank, so "you could get the same performance by buying just about any other of the large cap banks." No edge, so no view.

State Street is the other big US trust/custody bank, doing the same safekeeping-and-servicing work as BNY Mellon. Eisman groups the two together as a separate species from ordinary banks and says he doesn't follow either. Mentioned as a category, not as a call.

Citigroup is his cautionary tale about management risk. He owned it before Jane Fraser became CEO and sold after yet another trading scandal: "I had the right thesis, but the wrong management. I should have bought an alternative."

The lesson he draws is transferable — if the reason to own a stock is a turnaround, the analysis has to include whether the people running it can execute; a correct idea attached to the wrong operator still loses money. He owns no banks today.

Morgan Stanley is the stand-in for the whole big-bank group in this episode, and the argument is about diversification rather than quality. Two reasons he isn't buying: the banks have already had a great run and sit "at peak valuations," and — more importantly — the investment-banking cycle driving their profits is now "heavily dependent on AI financing needs" (underwriting the debt and equity that funds data centers).

So a bank stock is a hidden AI position: "owning Morgan Stanley, Goldman, Bank of America, Citigroup, etc. is just one more aspect of the AI trade. Simply put, buying banks to me does not provide diversification from tech." If you buy banks to hedge tech, you haven't hedged anything.

Moonshot is a Chinese AI lab. It released a model called Kimi K3 and claims it is as capable as any leading Western model "but for a fraction of the cost."

Why that matters for a US stock portfolio: the worry about AI used to be that it costs enormous sums to build (capital intensity) and that no company can keep a lasting lead (no moat). A credible, much-cheaper Chinese model adds the third leg — a price war, where everyone's AI revenue gets cut while the spending commitments stay. "Now, the possibility of a price war looms closer."

Two things Eisman does before judging the quarter. First, he strips out Google's gains on its investment stakes, which flow through reported earnings but aren't the operating business — on that adjusted basis earnings missed. Second, he looks past revenue (which was strong: +24%, with cloud +82%) to cash.

And the cash is the story: after all the AI spending, free cash flow turned negative — minus $5.9 billion. Free cash flow is what's left after running the business and paying for equipment and buildings; when it goes negative, growth is being funded from the balance sheet rather than from profits. Google also raised its 2026 AI budget again, from $190 billion to $205 billion. A year ago that raise would have been cheered; this time "the market is beginning to lose patience," and the stock fell 7%.

Tesla sold plenty of cars — revenue beat — but earned much less on them: profit per share fell to 33 cents against 51 cents expected. Part of the margin damage is the near-disappearance of regulatory credits (payments Tesla used to collect from other carmakers for their emissions shortfalls), down 67% after the Trump administration largely eliminated them. That was close to free profit, and it's gone.

The rest is the same capex story as Google: spending rose to $5.8 billion, pushing free cash flow negative (minus $1.1 billion) "for the first time in 2 years." In this new regime that is the unforgivable print — the stock fell 14.5% in a day, and together with Google dragged the Nasdaq down more than 2%.

IBM sells big-iron computers (mainframes) and, mostly, consulting and IT services. It had already warned investors that the quarter would be bad — the stock dropped 25% that day — and the actual results matched the warning, with earnings up only 5% and the revenue outlook cut.

The mechanism is unusual and worth remembering: because the price of tech hardware is soaring, customers are rushing to lock in equipment purchases and cutting the discretionary services budget to pay for it. Mainframe sales fell 42%. So the AI hardware boom is actively draining IBM's services business. "IBM is clearly struggling."

Equifax is one of the three credit bureaus, but its biggest division isn't credit scores — it's Workforce Solutions, which lets lenders, employers and government agencies verify someone's income and employment automatically. That is effectively a software subscription business, which drops it straight into the "SaaSpocalypse" debate: bears think AI-powered verification tools will take its share.

When a stock carries that narrative, "there is no room for error" — and Equifax erred. Government revenue in that division fell mid-single digits when management had guided to flat, and the next two quarters' earnings guidance came in 3% below what analysts expected. Management blamed state budget pressure, which may well be true, but "no one is interested in excuses" while the narrative rules. The stock fell another 4% on top of a 17% year-to-date decline.

Domino's is Eisman's recurring read on the "K-shaped economy" — the idea that higher-income consumers are fine while lower-income consumers are in their own recession. Cheap pizza is what the bottom of that K buys, so its sales are a thermometer.

Reported earnings looked better (up 7%) and revenue beat, so the stock popped 2% — "don't get carried away." The number that matters for a restaurant chain is same-store sales: how much more the existing stores sold versus a year ago, stripping out new openings. That grew just 0.1%, the weakest in five quarters, and the stock gave back the gain and more. Down 20% this year.

SpaceX was the marquee IPO of this cycle and Eisman was never a fan — his objection was that its value increasingly rests on an AI story while its spending on satellites and launch capacity ballooned. It has now fallen below the price at which it was sold to public investors.

He doesn't over-claim: "I'm not sure what this means yet, but it does not bode well for the IPO market." A broken high-profile deal tends to close the window for the ones behind it, since bankers need the last IPO to have worked in order to price the next.

FICO owns the credit score US lenders rely on, and Eisman is short it — he profits if the stock falls. It appears here only as context: Equifax is one of the three bureaus whose data FICO's scores are built on, and he notes he'd previously discussed Equifax only "in the context of my short thesis on FICO." The thesis itself (a mortgage-scoring price war with VantageScore) is unchanged from earlier episodes.