Big Short Investor Explains How the AI Bubble Could Burst

Anthropic

Across the shows: 105 speakers in other episodes, 136 bullish and 87 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBullish

  Anthropic's revenue is growing far faster than OpenAI's, making it the stronger of the two labs.

  Anthropic reported over $11B of revenue in the June quarter, up over 100% from the March quarter. Over the same period OpenAI's revenue grew only 18%. Eisman concludes that between the two, OpenAI is the weaker entity, implying Anthropic is in a stronger position.

  “Compare that to [Anthropic's] revenue of 11 plus billion in the June quarter, which was up over a 100% versus the March quarter.”at 31:19

• Steve Eisman · The Real Eisman PlaybookBearish

  Enterprise reliance on Claude is not a moat; rivals at lower prices would trigger switching.

  Eisman says he hears strong enthusiasm for Claude, with companies and universities endorsing it, and that some argue enterprise reliance creates a moat. He rejects this. As soon as ChatGPT, Gemini or Chinese models match Claude's quality at a lower price, pressure to switch will be huge, and the moat disappears.

  “The argument is that enterprise reliance is a moat. To this I say, the minute ChatGPT or Gemini or the Chinese models offer the same quality and a lower price, pressure will be huge to switch.”at 39:47

• Steve Eisman · The Real Eisman PlaybookBearish

  Anthropic lacks the balance sheet to withstand a price war and is reportedly already cutting prices.

  Eisman says cheap open-weight models make a price war likely, and there are no moats to protect margins. Anthropic and OpenAI lack the balance sheets to absorb one. Reports say both are cutting prices out of fear of losing market share. If Anthropic gets into trouble in the next year, that would be telling.

  “Anthropic and OpenAI do not have the balance sheets to withstand a price war. And should a price war break out, fears will grow that the entire AI story will just slow.”at 41:37

OpenAI

Across the shows: 41 speakers in other episodes, 21 bullish and 31 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBearish

  OpenAI's financials are weakening and it may fail; costs are outpacing revenue as rival Anthropic grows faster.

  Per the Wall Street Journal, OpenAI's June quarter revenue was $6.7B, up only 18% from the March quarter, while Anthropic's was over $11B, up over 100%. OpenAI's costs rose $3B to $12.3B in the same period, against only a $1B revenue increase. Senior departures include the chief revenue officer after eight months and the COO after eight years, both in August 2026. Eisman says OpenAI chose consumer-first, giving its product away or underpricing it, and is now late to B2B against Anthropic. Citing Ed Zitron, he says OpenAI needs about $100B a year in funding. He is not predicting failure yet, but says the company is potentially in trouble, and that negative press makes raising capital harder.

  “So in three months, revenue increased 1,000,000,000, but costs surged 3,000,000,000.”at 31:35

• Steve Eisman · The Real Eisman PlaybookBearish

  OpenAI has no durable moat; customers can switch models easily, with switching costs near zero.

  Eisman argues AI has no lasting moat, unlike Google's search business. Leadership changes quickly: ChatGPT is on top one day and overtaken the next. Switching costs between models seem perhaps zero, or at least very low. He calls this risk the most important of the five, and says it cannot be overstated.

  “One day ChatGPT is on top, the next day it's clawed, and the switching cost between models seems to be perhaps zero, but certainly very low. I can't overstate the importance of this risk.”at 39:30

• Steve Eisman · The Real Eisman PlaybookBearish

  OpenAI lacks the balance sheet to survive a price war and is already cutting prices.

  Eisman sees a price war as the key risk. Open-weight models are much cheaper, and there are no moats, so the probability of a price war rises sharply. Anthropic and OpenAI do not have the balance sheets to withstand one. Reports say both are already cutting prices out of fear of losing share to cheaper models. If OpenAI runs into trouble in the next year, that would be telling.

  “Anthropic and OpenAI do not have the balance sheets to withstand a price war. And should a price war break out, fears will grow that the entire AI story will just slow.”at 41:37

GE Vernova

Across the shows: 6 speakers in other episodes, 8 bullish and 1 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBullish

  GE Vernova is benefiting from gas turbine shortages as utilities cannot meet AI data center power needs.

  Utilities cannot satisfy all AI data center power demand, and building a gas turbine takes years. GE Vernova is a large builder of gas turbines, which is why Eisman says the stock has done well. He also notes that if the AI story fails, AI-linked industrial power beneficiaries like GE Vernova would decline.

  “That is why a stock like GE Vernova has done well because GEV is a large builder of gas turbines.”at 27:14

• Steve Eisman · The Real Eisman PlaybookBullish

  GE Vernova is a preferred power-side picks-and-shovels play for the AI buildout.

  Eisman favors infrastructure suppliers that sell what the AI buildout needs, regardless of which LLM wins. On the power side, GE Vernova and Eaton are examples. They sell the infrastructure required to build the entire thing.

  “and then on the power side, you've got companies like GE Vernova and Eaton. These businesses don't necessarily need to predict which LLM ultimately wins.”at 42:59

NVIDIA

Across the shows: 79 speakers in other episodes, 66 bullish and 30 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookUncertain

  Nvidia's revenue leans on five customers and circular financing, adding risk despite strong growth.

  Nvidia's revenue grew over 100% in its latest quarter, which supports the bull case. However, its 10-Q shows 70% of accounts receivable, a proxy for revenue, came from five customers, which Eisman assumes are hyperscalers or their close partners. He also notes Nvidia invests in or lends to other tech companies, which then use the money to buy its chips. Eisman does not think concentration or circular financing alone will sink the AI story, but says they make it more dangerous.

  “However, in NVIDIA's 10 Q, it states that 70% of NVIDIA's accounts receivable, which is a proxy for revenue, was from five customers. That's a lot of concentration risk for the largest company in the world.”at 28:27

• Steve Eisman · The Real Eisman PlaybookBullish

  Nvidia sits at the center of the compute buildout and is a preferred picks-and-shovels play.

  Eisman prefers picks-and-shovels businesses to AI model winners. These companies do not need to predict which LLM wins, because they sell the infrastructure needed to build the whole thing. He says NVIDIA obviously sits at the center of the compute buildout.

  “NVIDIA obviously sits at the center of the compute build out.”at 42:28

Adobe

Across the shows: 4 speakers in other episodes, 1 bullish and 3 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBearish

  Adobe is especially vulnerable to AI disruption, lacking a moat; shares already down about 60%.

  The SaaSpocalypse argument is that AI collapses the cost of creating software. Eisman says the narrative is too simple, since embedded enterprise software with prohibitive switching costs holds up. Many software companies lack that kind of moat, and Adobe is the example. Adobe hit a high of $630 in February 2024 and is now down about 60%. An engineer has also released a free open-source alternative to Adobe's Creative Cloud. Eisman says he does not know how good it is, but 'free is free'.

  “However, many software companies do not have that kind of moat. Take Adobe for example. The perception is that Adobe is particularly vulnerable to AI.”at 15:38

Oracle

Across the shows: 39 speakers in other episodes, 10 bullish and 36 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBearish

  Oracle is overleveraged and too reliant on OpenAI; its stock is exposed if OpenAI fails.

  Oracle's total debt has risen sharply over two years, and S&P downgraded it to triple B minus, one notch above junk. About 50% of its roughly $600B backlog comes from OpenAI. When Oracle's backlog jumped in 2025 and analysts flagged OpenAI's share, the stock gave back its gains and is now about 70% below its peak. Eisman argues that if OpenAI failed, Oracle would be in immediate trouble because of its debt load, and the stock would likely fall much further.

  “The ratings agencies are not amused. S and P downgraded Oracle to triple b minus, which is only one level above [junk].”at 22:05

Meta Platforms

Across the shows: 70 speakers in other episodes, 71 bullish and 37 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookUncertain

  Meta's off-balance-sheet Louisiana data center financing looks like accounting engineering; more deals likely.

  Meta financed a $27B Louisiana data center through an SPV, Beignet Investor LLC. Meta owns 20%, with Blue Owl and PIMCO owning the rest. Meta builds and will lease the facility for 20 years and bears delay and overrun costs, yet the debt stays off its balance sheet. Ernst & Young flagged the arrangement as a critical audit matter in Meta's 2025 10-K. Eisman reads that as the auditor being uncomfortable and relying on Meta's description. He calls the structure a fancy footwork reminiscent of past bad times, and expects many more such transactions across the industry.

  “Technically, Meta owns only 20% of Beignet with the rest owned by Blue Owl and PIMCO. ... That's some pretty fancy footwork that's reminiscent of bad times past and doesn't pass the smell test.”at 23:52

Micron Technology

Across the shows: 28 speakers in other episodes, 25 bullish and 16 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBullish

  Micron is a preferred picks-and-shovels play, supplying memory for the AI buildout.

  Eisman favors suppliers of AI infrastructure over bets on which model wins, since they sell into the buildout regardless of the winner. He names Micron as supplying memory for that build-out.

  “You've got companies like Micron supplying memory,”at 42:28

Arista Networks

Across the shows: 2 speakers in other episodes, 2 bullish and 0 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBullish

  Arista is a preferred picks-and-shovels play, supplying networking for the AI buildout.

  Eisman favors infrastructure suppliers that do not need to predict which LLM wins. Arista, alongside Cisco, supplies the networking needed for the AI buildout.

  “Arista and Cisco supplying networking,”at 42:56

Cisco Systems

Across the shows: 3 speakers in other episodes, 3 bullish and 0 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBullish

  Cisco is a preferred picks-and-shovels play, supplying networking for the AI buildout.

  Eisman favors infrastructure suppliers that do not need to predict which LLM wins. Cisco, alongside Arista, supplies the networking needed for the AI buildout.

  “Arista and Cisco supplying networking,”at 42:56

Eaton

Across the shows: 2 speakers in other episodes, 4 bullish and 0 bearish calls, past 30 days

• Steve Eisman · The Real Eisman PlaybookBullish

  Eaton is a preferred power-side picks-and-shovels play for the AI buildout.

  Eisman favors infrastructure suppliers that sell what the AI buildout needs, regardless of which LLM wins. On the power side, Eaton and GE Vernova are examples. They sell the infrastructure required to build the entire thing.

  “and then on the power side, you've got companies like GE Vernova and Eaton. These businesses don't necessarily need to predict which LLM ultimately wins.”at 42:59

AI: overall outlook

• Steve Eisman · The Real Eisman Playbook

  Bull case, where AI keeps lifting markets, has a 30% chance.

  In the bull case, AI changes the world, Anthropic and OpenAI keep driving stock market gains, and NVIDIA sets endless records. This has been the dominant narrative for about two years, and it is not crazy. The largest tech companies believe it, so Eisman assigns it a 30% chance.

  “AI for the win. AI will change the world, and Anthropic and OpenAI will continue to drive the stock market to new and greater heights... I give this narrative a 30% chance.”at 7:05

• Steve Eisman · The Real Eisman Playbook

  Bear case, with AI overhyped and returns failing, has a 30% chance and triggers recession.

  In the bear case, AI is overhyped and adoption is slow. Even with fast adoption, the capital needed for data centers and chips, and the prices paid to Anthropic and OpenAI, will not materialise. Chinese open-weight models trigger a price war, moats vanish, and returns on investment fall short. The result is a full economic recession, a market crash, and an unwinding of private credit and private equity that has chased AI. Eisman gives this scenario a 30% chance, and expects hyperscaler stocks to be available cheaply after the crash.

  “and the returns on investment don't measure up. The result, the bear case leads to a full economic recession and a market crash.”at 7:57

• Steve Eisman · The Real Eisman Playbook

  Intermediate case of slower buildout and price war, market unwinds AI gains, has 40% chance.

  In the intermediate case, Nvidia, the hyperscalers and debt and equity investors keep financing the build-out, but data centers are built more slowly and with more hiccups. Chinese models siphon demand and cause a price war, or data center construction is so large it dwarfs demand. Eisman cites Brad Saflow of PAA Research, who estimates US data center capacity will be double demand by the end of 2027, slowing AI capex in 2028. He says the political backlash against data centers and capacity constraints are partial offsets. His view is that there is simply too much capex chasing demand. The market then unwinds much of its AI gains into a bear market for a time, which he gives a 40% chance.

  “This narrative assumes that AI is a great technology, but there is simply way too much CapEx chasing demand. Result, the market unwinds much of its AI gains and we have a bear market for a time. I give this narrative a 40% chance.”at 9:35

• Steve Eisman · The Real Eisman Playbook

  AI is real, already reshaping the economy, and will generate an enormous amount of money.

  Eisman says caution about parts of the AI trade is not bearishness on AI. He argues AI is real and has already started changing the economy. It is also driving the US economy right now. He expects an enormous amount of money to be made, but the harder question is who captures it.

  “I think AI is real. I think it's already started changing the economy, and I think an enormous amount of money will be made from it.”at 42:28

AI infrastructure & data centers

• Steve Eisman · The Real Eisman Playbook

  Hyperscalers have turned capital-heavy; rising leverage alone won't sink the AI story.

  Before AI, hyperscalers were capital-light cash generators that bought back stock. AI changed that: GPU costs are very high (Blackwell over $5,000 per chip, about $500M of GPUs per large data center), and Google's roughly $80B annual free cash flow is now gone. Google raised $85B of equity in 2026, its first equity raise since 2005. Oracle has taken on leverage and Meta is using off-balance-sheet financing. Eisman does not think the shift in itself is enough to sink AI, because if AI succeeds the leverage will look like good investments. The risk comes in combination with off-balance-sheet and circular financing.

  “Bottom line, AI CapEx requirements have altered the business models of the hyperscalers from capital light to capital heavy. However, I don't think that this change in and of itself is enough to sink the AI story.”at 22:13

• Steve Eisman · The Real Eisman Playbook

  AI data center construction is likely slower than hoped due to power, permits and local opposition.

  A gigawatt-scale AI data center costs around $30B to build and equip. The grid is already stressed, and utility plants, such as gas turbine plants, take years to build. Permitting is being delayed in many parts of the country as communities push back over water and energy needs, and the AI industry has become unpopular. Hyperscalers are therefore looking at nuclear and alternatives like Bloom Energy. Eisman suspects construction is taking longer than hoped, but says he does not know how much impact this will have.

  “My suspicion is that AI data center construction is taking longer than hoped, but how much of an impact this will have is something I just don't know.”at 27:31

• Steve Eisman · The Real Eisman Playbook

  US growth hinges on hyperscaler capex, which hinges on OpenAI and Anthropic; OpenAI failure could trigger recession.

  Hyperscaler AI capex, around $850B in 2026, accounts for about half of projected 2026 US GDP growth of 2%. Reports suggest about 70% of hyperscaler AI revenue comes from Anthropic and OpenAI (Eisman cannot confirm this). OpenAI is about 50% of Oracle's roughly $600B backlog and over 20% of CoreWeave's order book. Eisman concludes US economic health depends on hyperscaler capex, which depends on those two labs. If OpenAI failed, hyperscalers would cut capex, and he thinks that would push the US into an immediate recession.

  “Therefore one must conclude that the health of The US economy is extremely dependent on hyperscaler CapEx, and hyperscaler CapEx is highly dependent on the health of Anthropic and OpenAI.”at 29:40

Open-source AI models

• Steve Eisman · The Real Eisman Playbook

  Chinese open-weight models will pressure AI pricing, sparking a price war and eroding moats.

  The Chinese keep producing a good-enough model that handles much of computing needs. This siphons off demand from Anthropic and OpenAI and leads to a price war. In the bear case, this erodes the moats of closed AI labs and their exorbitant prices for programming, so returns on the AI build-out do not measure up. Eisman raises the same risk in his intermediate scenario.

  “The Chinese open weight models will put pressure on pricing. A price war will ensue. Moats vanish,”at 7:50

• Steve Eisman · The Real Eisman Playbook

  Cheaper Chinese open-weight models are taking market share from closed frontier labs.

  While the token pricing saga played out, Chinese open-weight models began taking market share. Eisman says the flagship models of Anthropic and OpenAI are superior, but open-weight models are much cheaper. Many users do not need the most capable model, so the cheaper alternatives are a real threat to closed labs and drive the price-war risk.

  “Chinese open weight models started to take market share. The important point here is that open weight models are much cheaper than the models of Anthropic and OpenAI.”at 40:58

US stock market outlook

• Steve Eisman · The Real Eisman Playbook

  AI-linked businesses make up well over 65% of the S&P 500, so the index hinges on AI.

  Info tech is 40% of the S&P 500, mostly AI-related. Adding Google and Amazon brings it to about 50%. Utilities, which are about 3% and spending $250B this year on AI power needs, are AI-related. Industrials like Eaton, Rockwell, GE Vernova and Quanta, and financials financing AI, add further exposure. Only staples and health care, about 14% of the index, are not involved with AI. The direction of the market therefore hinges on the future of AI.

  “And add all that up, and you are well north of 65% of the S and P.”at 12:08

• Steve Eisman · The Real Eisman Playbook

  In an AI bust, investors rotate into healthcare, consumer staples and P&C insurance as safety.

  If the AI story fails, Eisman expects capital to reallocate to safety sectors and sub-sectors. Healthcare and consumer staples are the main safety sectors. Within sectors there are also safe sub-sectors, such as property and casualty insurance within financials. He frames this as a reallocation question rather than stock picking.

  “Investors will reallocate to safety sectors and sub sectors. In the safety sector category, think about healthcare and consumer staples.”at 38:29

AI regulation & safety

• Steve Eisman · The Real Eisman Playbook

  AI doomsday talk is nonsense, used by Anthropic and OpenAI leaders to scare government into regulating rivals.

  Eisman dismisses the doomsday scenario, in which AI takes over the earth like in Terminator, as not foreseeable. He says the heads of Anthropic and OpenAI know this. If everyone really believed the world was ending, these companies would not be going public or building data centers. He argues they use fear to scare government into regulating competitors out of existence, and gives the doomsday scenario a 0% chance.

  “They pedal this nonsense and surf the fear curve they've created because they want to scare the government into regulating the competition out of existence.”at 10:39

Credit markets

• Steve Eisman · The Real Eisman Playbook

  AI-linked firms are over half of new corporate bond issuance, so bonds offer little AI diversification.

  About 15% of existing corporate bonds are AI-related, but over 50% of this year's issuance is AI-related. Because AI is so pervasive in the bond market, bonds do not provide the diversification from the AI-driven equity market that investors are looking for. The AI boom is therefore affecting the bond market as well as stocks.

  “About 15% of existing corporate bonds are AI related, but over 50% of this year's issuance is AI related. So bonds don't provide the same degree of diversification that investors are looking for.”at 12:08

Cybersecurity

• Steve Eisman · The Real Eisman Playbook

  Cybersecurity is the exception to the SaaS selloff because AI makes hacking easier.

  In the SaaSpocalypse narrative, AI replaces most existing software except cybersecurity. Eisman says cybersecurity remains necessary because AI makes it easier to hack. He treats it as the exception to the software selloff, while noting that the narrative is overly simplistic overall.

  “with the exception of cybersecurity. Cybersecurity remains necessary because AI makes it easier to hack.”at 14:58

Private equity

• Steve Eisman · The Real Eisman Playbook

  PE-owned software buyouts face equity wipeout as values fall about 50%; loans refinance next year.

  Between 2017 and 2022, private equity bought software companies, mostly SaaS, with roughly 50% debt and 50% equity, with private credit lending. Using Adobe's 60% decline as a proxy, and being more conservative, Eisman assumes PE-held software is down 50%. That wipes out the equity, and the problem becomes clear when these loans are refinanced next year. He does not think the problem is large enough to seriously hurt the economy, but PE investors are likely to be hurt badly.

  “A 50% decline in value means that the equity is wiped out. This will start to become clear when these loans start to be refinanced next year.”at 17:02

Banks

• Steve Eisman · The Real Eisman Playbook

  Big banks trade near peak valuations on the AI financing boom; they would correct if AI needs fade.

  Investment banks and large banks are trading at fairly close to peak valuations. The investment banking cycle is very strong, partly because of the financing needs of AI. If those financing needs lessen, as in an AI failure scenario, the investment banking cycle would weaken and bank stocks would correct from their peak valuations.

  “Investment banks and large banks, these stocks are at fairly close to peak valuations. Also, the investment banking cycle is super strong, partially because of the financing needs of AI.”at 37:50

Enterprise AI adoption & AI software

• Steve Eisman · The Real Eisman Playbook

  Enterprise 'token maxing' has ended as firms became cost-conscious after AI prices rose.

  Last year, subscriptions were priced far below token costs. Enterprises told engineers to use AI around the clock regardless of cost. Earlier this year, prices rose and AI costs soared. Uber, for example, blew through its entire AI budget in a few months. Sometime this summer, token maxing ended as all enterprises became more cost-conscious.

  “Earlier this year, prices were raised and enterprises found their AI costs soaring. Uber, for example, blew through its entire AI budget in just a few months. Sometime this summer, token maxing ended”at 40:37

AI pricing & unit economics

• Steve Eisman · The Real Eisman Playbook

  Open-weight models cost about 20 cents per million tokens versus $2–10 for flagship models.

  AI costs are measured in tokens. Last year, LLM providers charged subscriptions far below the token cost, which encouraged enterprises to use AI around the clock. Prices then rose. Flagship models from Anthropic and OpenAI cost roughly $2 to $10 per million tokens. Open-weight models cost around 20 cents per million. Eisman says the flagships are superior, but most trips do not need a Mercedes, so the price gap drives a price-war risk.

  “The flagship models of Anthropic and OpenAI can cost 2 to $10 per 1,000,000 tokens. Open weight models can cost around 20¢ per 1,000,000 tokens, so much cheaper.”at 41:07

Software & SaaS

• Steve Eisman · The Real Eisman Playbook

  Some software businesses face low switching costs and risk as agentic AI grows more capable.

  Eisman says he would be more careful with businesses where investors assume enormous future profits despite very low switching costs. He names parts of software among them, as agentic AI becomes more capable. The core question is how hard it is for a customer to leave.

  “Where I'd be much more careful is in businesses where investors are assuming enormous future profits despite very low switching costs.”at 43:10

AI bubble & tech valuations

• Steve Eisman · The Real Eisman Playbook

  AI valuations assume an extraordinary future, so mere disappointment, not technology failure, can drive losses.

  Eisman argues that when valuations assume an extraordinary future, the technology does not need to fail, and the stocks do not need to disappoint. Reality only has to fall short of expectations. He sees a plethora of reasons, including Chinese competition, the lack of moats, heavy commitments to OpenAI and Anthropic, and power grid limits, to watch those expectations closely.

  “when valuations assume an extraordinary future, you don't necessarily need the technology to fail or the stocks to disappoint. You just need reality to fall short of expectations.”at 43:53