Stance reflects how each name was framed by Eisman in the quoted clips — not by the host, and not a price rating. Rows exist only where Eisman himself names the company, with one flagged exception (NVDA, host-framed, kept for continuity because it is the layer his map sits above). The host's workshop promotion is advertising, not a pick, and is excluded. Research: QT Qualtrim · SA Seeking Alpha · STK StockAnalysis.
| Ticker | Name | Research | View | What he said | At |
|---|---|---|---|---|---|
| MSFT | Microsoft | QT · SA · STK · FA | Positive | The defended layer, and the explicit contrast case: "Google and Microsoft have multiple revenue streams from established businesses which are very unlikely to simply disappear. They also have hyperscaler businesses to balance their vulnerability." Caveat kept in his own words — "their LLM businesses are also questionable." | 10:24 |
| AMZN | Amazon | QT · SA · STK · FA | Positive | Named in the hyperscaler four: "the hyperscalers like Google, Amazon, Microsoft, and Oracle have real businesses here. What the returns will look like, I don't know yet, but they have real businesses." The moat is the barrier to entry itself — "the amount of money it takes to be a hyperscaler is insane and that expenditure itself is a moat." | 9:42 |
| ORCL | Oracle | QT · SA · STK · FA | Neutral | Counted among the hyperscalers with "real businesses," but singled out as the most counterparty-concentrated of them: "of Oracle's 600-plus billion backlog, around half is from OpenAI." That makes it the cleanest public read on the LLM layer's health — and the most exposed if the layer cracks. | 4:53 |
| NVDA | NVIDIA | QT · SA · STK · FA | Neutral | Host-framed, not an Eisman clip — the narrator places Nvidia at the top of the gold-rush stack ("the ultimate pick and shovel merchant… supplying the hardware to the hyperscalers"), one rung above the hyperscalers Eisman defends. Eisman does not name it here; his own NVDA stance is on the Aug 28 page. | 2:39 |
| DeepSeek | DeepSeek (private, China) | — | Neutral | The named face of Eisman's "cheaper open-source Chinese models" — the substitution he says enterprises are already making "in order to control costs." Host adds the price gap: DeepSeek V4 Flash at 14¢/28¢ per million input/output tokens, ~3¢ per test. | 3:02 |
| Moonshot AI | Moonshot AI (private, China) | — | Neutral | Second Chinese lab in the host's cost table — Kimi K3 at ~86¢ per test versus $1.86 for GPT-5.6 and $3.15 for Claude Fable 5. The exhibit behind Eisman's line that "the Chinese models are much cheaper, and this could eventually cause a price war." | 4:04 |
| GOOGL | Alphabet (Google) | QT · SA · STK · FA | Negative | Sold. "I sold my Google. I wanted to reduce my exposure to AI. What scares me is that it's all one trade. So it better succeed." Not a company complaint — elsewhere in the same clips Google is one of the hyperscalers with "multiple revenue streams from established businesses." The sale is an exposure decision, not a verdict on the business. | 0:26 |
| OpenAI | OpenAI (private) | — | Negative | "Much more problematic… there just don't seem to be any moats, or at best, the moats are shallow… the future for these large LLM providers is very questionable." And the transmission channel is its order book: "of Oracle's 600-plus billion backlog, around half is from OpenAI." | 4:53 |
| Anthropic | Anthropic (private) | — | Negative | Same problem, stated as a business-model gap: "Anthropic and Open AI are also problematic because they don't have the breadth of revenue streams of Google and Microsoft." Paired with OpenAI as the one thing to monitor — "a key thing to monitor to determine a catalyst for a real sustained sell-off is the health of Anthropic and OpenAI." | 10:24 |
Google is a hyperscaler: alongside search and ads, it rents out enormous data centres that other companies' AI models run inside. Eisman is not saying the business is broken — in the same set of clips he lists Google among the companies with "multiple revenue streams from established businesses which are very unlikely to simply disappear."
He sold it anyway, and the reason is about his portfolio rather than the company: "I sold my Google. I wanted to reduce my exposure to AI." His worry is that almost everything in a normal portfolio now rises and falls with the same AI story, so owning a great AI-linked business does nothing to spread risk. Selling a name he still respects is the price of cutting a concentration he thinks is invisible to most investors.
Microsoft sits on both sides of the AI business. It runs Azure — a hyperscaler, meaning it owns the data centres AI models live in — and it also has a large stake in the model layer through OpenAI. Eisman treats the first half as durable and the second half as the risky part.
His argument for the durable half is unusual: normally a business that has to spend colossal sums is considered weaker, but here the spending is the defence. "The amount of money it takes to be a hyperscaler is insane and that expenditure itself is a moat" — very few companies on earth can write those cheques, so competition is structurally limited. On top of that, Microsoft earns money from Windows, Office and enterprise software that would not vanish if AI disappointed. He still flags the caveat in his own words: "their LLM businesses are also questionable."
Amazon is in the same defended category. AWS is one of the handful of places an AI company can physically run its models, and retail plus advertising provide revenue that has nothing to do with whether the AI boom continues.
Eisman's summary of the group is deliberately unromantic: "the hyperscalers like Google, Amazon, Microsoft, and Oracle have real businesses here. What the returns will look like, I don't know yet, but they have real businesses." That is the whole claim — not that the returns on all this spending will be good, but that if the AI layer above them stumbles, these companies still have something left. The host's own reading of why Eisman prefers them is worth separating out as commentary rather than Eisman's words: moats, diversified revenue, and the balance-sheet strength to keep spending through an uncertain stretch.
A "backlog" is contracted future revenue — work a customer has committed to buy but has not yet been billed for. Investors treat it as visibility into future earnings, and it is usually a comfort. Eisman turns it into the risk.
"Of Oracle's 600-plus billion backlog, around half is from OpenAI." Half of the promised future business depends on the financial health of one privately held company that is losing money. If that customer cannot pay, or renegotiates, the backlog shrinks — and with it the future revenue already reflected in the share price. He does not call Oracle a bad business; he counts it among the hyperscalers with real businesses. He simply identifies it as the place where the LLM layer's problems would show up first and most visibly in public markets, which makes it as much a monitoring instrument as a stock.
OpenAI builds the AI models behind ChatGPT. It does not own the data centres it runs on — it rents them, in enormous quantity, from hyperscalers like Microsoft and Oracle.
Eisman's problem is that there is nothing stopping a customer from leaving. "There just don't seem to be any moats, or at best, the moats are shallow. Enterprises are switching between models and using cheaper open-source Chinese models in order to control costs." A business with no lock-in and cheaper substitutes appearing is a business that competes on price, and price competition against a rival with a far lower cost base rarely ends well: "the Chinese models are much cheaper, and this could eventually cause a price war."
The reason this matters beyond OpenAI is that its spending commitments are other people's revenue — roughly half of Oracle's $600bn-plus backlog. So its financial health is not a private-company curiosity; it is the load-bearing assumption underneath a large slice of listed technology.
Anthropic is the other pure model company — it builds Claude and, like OpenAI, rents its computing power from the hyperscalers rather than owning it.
Eisman's objection is a business-model one rather than a product one: "Anthropic and Open AI are also problematic because they don't have the breadth of revenue streams of Google and Microsoft." If the AI story cools, Google still sells search ads and Microsoft still sells Office; a pure model company has only the model. That single point of failure is what makes the layer fragile even if the technology is good.
Which is exactly why he ends up naming these two as the thing to watch rather than any listed stock: "a key thing to monitor to determine a catalyst for a real sustained sell-off is the health of Anthropic and OpenAI." They are private today, so the health is hard to see — but both are heading toward IPOs, and an S-1 would make the numbers public for the first time.
Summary & timestamps derived from the public YouTube video (transcript in transcript.txt) for personal study. Steve Eisman appears in quoted clips only; narration is by the New Money channel. Not investment advice. © New Money / Steve Eisman for source material.