• Favor diversified AI infrastructure suppliers such as NVDA, MU, ANET, CSCO, and GEV over bets on which AI model will win, while recognizing their shares still depend on continued data-center spending.
• Treat ORCL as a high-risk exposure: its debt has risen to $155 billion, and roughly half its $600 billion-plus backlog is tied to OpenAI.
• Monitor OpenAI’s revenue versus costs and fundraising needs, alongside hyperscalers’ capital spending; a slowdown could pressure MSFT, GOOGL, AMZN, and AI-linked industrials.
• For a more defensive allocation, consider LVHD, SPLV, or KBWP, especially if AI-related spending and valuations weaken.

AI Infrastructure and Market-Wide Exposure

• Eisman describes AI as a genuine, potentially transformative technology, but says the investment outcome depends on how quickly it is adopted, what it costs, and which companies ultimately earn profits.
• His scenario estimates are 30% for the bullish case, 30% for the bearish case, and 40% for an intermediate case in which AI grows but spending, construction, or returns disappoint. He assigns the doomsday scenario roughly 0%.
• The intermediate case could involve excess data-center capacity, slower construction, cheaper Chinese models taking share, and an eventual slowdown in AI capital spending. He cites one analyst’s estimate that U.S. data-center capacity could be double demand by the end of 2027; if so, AI spending could slow in 2028.
• AI exposure is unusually widespread: Eisman estimates that more than 65% of the S&P 500 is directly or indirectly tied to AI. He also says AI-related bonds make up about 15% of outstanding corporate bonds and more than 50% of this year’s issuance.
• He says hyperscalers are expected to spend about $850 billion on AI infrastructure in 2026, and that this spending accounts for roughly half of projected U.S. GDP growth of 2%.

Takeaways

• Consider AI exposure as a market-wide concentration risk, not just a technology-sector issue.
• Eisman’s central caution is that the technology could succeed while many investments still disappoint if spending exceeds demand or profits fall short of expectations.
• The scenario probabilities are the speaker’s judgments, not forecasts or guarantees.

NVIDIA (NVDA)

• Eisman calls NVIDIA the “kingpin” of the AI build-out and identifies it as a core picks-and-shovels supplier.
• He says NVIDIA’s latest-quarter revenue grew by more than 100%, but notes that 70% of accounts receivable came from five customers, indicating significant customer concentration.
• He also criticizes NVIDIA’s investments in, or loans to, other technology companies that then use the funding to buy NVIDIA chips, describing this as circular financing.
• A Blackwell GPU can cost more than $5,000, and a large data center may require at least 100,000 chips, or roughly $500 million in GPUs alone.

Takeaways

• Eisman favors infrastructure suppliers like NVIDIA over AI model providers, since chip demand may benefit from multiple competing models.
• The key concerns he identifies are customer concentration, circular financing, and whether the broader AI build-out earns adequate returns.

Microsoft (MSFT)

• Microsoft is identified as a hyperscaler investing heavily in data centers and AI infrastructure.
• In a downturn involving OpenAI or Anthropic, Eisman says hyperscalers including Microsoft could see slower cloud-revenue growth and pull back on capital spending.

Takeaways

• Microsoft is exposed to both the potential growth of AI cloud demand and the risk that infrastructure spending outpaces profitable demand.
• Eisman’s framework suggests watching customer demand, capital spending, and the returns on AI investment.

Alphabet / Google (GOOGL, GOOG)

• Google is described as a major hyperscaler and AI infrastructure spender. Eisman says Google historically generated more than $80 billion in annual free cash flow, but that cash flow is now being absorbed by AI investment.
• He says Google raised $85 billion in equity capital in 2026 to help fund AI needs.
• He describes Google’s search business as having had a strong moat, while warning that AI competition could challenge business models and that hyperscalers would be vulnerable to a pullback in AI spending.

Takeaways

• Google’s established businesses may provide more resilience than a standalone AI model company, but Eisman emphasizes the scale of its new capital requirements.
• The investment question is whether AI can generate returns sufficient to justify the spending and funding needs.

Amazon (AMZN)

• Amazon is identified as a hyperscaler building AI infrastructure and as a company with a durable cloud and retail business.
• In the bear-case discussion, Eisman says Amazon and Google could retain “rock-solid” businesses even after an AI-related market crash, potentially making their shares available at better prices.
• He also says a failure or setback at OpenAI or Anthropic could slow hyperscaler cloud growth and lead to reduced capital spending.

Takeaways

• Eisman distinguishes Amazon’s underlying business franchise from the risks attached to current AI valuations and spending.
• His bear-case observation is that a correction could create better entry prices, but he gives no specific price target or timing.

Oracle (ORCL)

• Oracle is described as more leveraged than its larger hyperscaler competitors. Eisman says its total debt rose from $94 billion to $155 billion over two years, and S&P downgraded it to BBB−, one level above junk.
• He says Oracle’s backlog, or remaining performance obligations, is over $600 billion, with about 50% tied to OpenAI.
• He warns that if OpenAI failed, Oracle would be in immediate trouble because of its debt and reliance on OpenAI-related business.
• Eisman says Oracle’s stock rose from $230 to $330 after a major backlog report, then later fell to around $130—about 70% below its peak—as investors focused on its OpenAI exposure.

Takeaways

• Oracle is a particularly direct way to express the risks Eisman identifies: leverage, customer concentration, and reliance on OpenAI.
• Investors following his analysis could monitor Oracle’s debt, credit rating, backlog composition, and OpenAI’s financial position.

Meta Platforms (META)

• Eisman says Meta is building its own models and agentic AI applications, and is constructing data centers for its own use.
• He highlights a $27 billion Louisiana data-center financing arrangement involving an off-balance-sheet special-purpose vehicle. Meta owns 20% of the vehicle, while Blue Owl and PIMCO own the remainder; Meta has agreed to rent the facility for 20 years and bears project delay and overrun costs.
• He says Meta’s auditor, Ernst & Young, flagged the arrangement as a critical audit matter in Meta’s 2025 10-K. He views the financing as a warning sign, while noting that the specific transaction is small relative to the overall AI build-out.

Takeaways

• Eisman’s concern is that off-balance-sheet financing can make AI-related obligations harder for investors to assess.
• Watch Meta’s data-center commitments and financing structures, not only the debt shown directly on its balance sheet.

OpenAI (Private Company)

• Eisman says OpenAI is central to demand for AI infrastructure and has a substantial relationship with Oracle.
• He reports that OpenAI’s June-quarter revenue was $6.7 billion, up 18% from the March quarter, while costs reached $12.3 billion, increasing by $3 billion over the same period.
• He says OpenAI may be losing share to Anthropic’s Claude, has lost several senior employees, and initially prioritized a consumer business with low prices before trying to catch up in business-to-business sales.
• He cites a commentator’s estimate that OpenAI may need to raise $100 billion per year for several years to survive, while stressing that he is not yet predicting failure.
• Eisman says an OpenAI failure in the near term could hurt Oracle, other hyperscalers, AI-linked industrial companies, financial firms, and the broader economy. He argues that the timing matters: the impact could be less severe if the industry has diversified by then.

Takeaways

• OpenAI is not a publicly traded stock, but Eisman treats its finances and ability to raise capital as a major system-wide risk indicator.
• Monitor revenue growth relative to costs, customer retention, leadership changes, fundraising needs, and its relationship with Oracle.

Anthropic (Private Company)

• Anthropic is described as a major driver of AI demand and a competitor to OpenAI. Eisman says its Claude model has gained positive attention and that the company chose the business-to-business market earlier than OpenAI.
• He reports that Anthropic’s June-quarter revenue was more than $11 billion, up over 100% from the March quarter.
• Despite that growth, he warns that Anthropic faces low switching costs, cheaper Chinese competition, and the possibility of a price war. He says reports indicate Anthropic and OpenAI are cutting prices to defend market share.

Takeaways

• Anthropic is not publicly traded, but its growth and pricing are important indicators for the AI infrastructure companies that depend on model demand.
• Strong revenue growth does not by itself establish a durable moat or prove that the company can earn attractive returns.

CoreWeave (CRWV)

• Eisman says OpenAI accounts for more than 20% of CoreWeave’s contracted order book, adding to the concentration of AI demand among a small number of companies.

Takeaways

• CoreWeave’s exposure to OpenAI is a concentration risk to monitor, particularly if OpenAI’s growth or funding weakens.
• The transcript does not provide a valuation view or price target for CoreWeave.

Micron Technology (MU)

• Eisman names Micron as a memory supplier that could benefit from the AI infrastructure build-out.
• He favors infrastructure suppliers that do not need to predict which AI model will ultimately win.

Takeaways

• Micron represents the AI infrastructure supply-chain theme in Eisman’s preferred area of the trade.
• The broader risk remains whether infrastructure spending and demand justify current expectations.

Arista Networks (ANET) and Cisco Systems (CSCO)

• Eisman names networking companies as AI infrastructure suppliers. The transcript says “Ariston,” apparently referring to Arista, alongside Cisco.
• His broader argument is that suppliers of the infrastructure required to build AI systems may be less dependent on which individual model provider wins.

Takeaways

• Networking is one of Eisman’s preferred picks-and-shovels areas within AI.
• The investment case still depends on continued data-center construction and spending.

GE Vernova (GEV)

• Eisman says AI data centers require substantial power, while utilities and the electrical grid may struggle to meet demand quickly.
• He identifies GE Vernova as a major gas-turbine builder and says its shares have benefited from the need for new power generation.
• He also warns that power constraints and permitting delays could slow data-center construction.

Takeaways

• GE Vernova is a direct way to express the power-supply theme behind AI infrastructure.
• The opportunity is tied to continued data-center demand, while delays in construction and permitting could affect the pace of investment.

Eaton (ETN)

• Eisman identifies Eaton as a company in the electrification space and a beneficiary of the equipment needed to build AI data centers.
• He also says Eaton could decline in an AI-related downturn as data-center spending slows.

Takeaways

• Eaton represents an infrastructure opportunity tied to AI electrification, but its prospects are exposed to the same capital-spending cycle Eisman says investors should watch.

Quanta Services (PWR)

• Eisman names Quanta as a power-related industrial company that could benefit from AI infrastructure investment.
• He says companies in this group could fall if OpenAI or Anthropic encountered serious trouble and hyperscalers cut capital spending.

Takeaways

• Quanta is exposed to the build-out of power infrastructure supporting data centers.
• The risk is that an AI spending slowdown could affect the broader set of AI-linked industrial companies.

Rockwell Automation (ROK)

• Eisman identifies Rockwell as an automation-related industrial company involved in the AI data-center ecosystem.
• He says AI-linked industrial shares could decline if demand for data-center construction and equipment weakens.

Takeaways

• Rockwell offers exposure to the automation side of the AI infrastructure theme, with the same spending-cycle risk Eisman highlights for related industrials.

Bloom Energy (BE)

• Eisman says hyperscalers are considering alternatives to the electrical grid, including nuclear power and gas-turbine technology such as Bloom Energy’s offerings.
• He presents these alternatives as responses to the difficulty of supplying enough power to AI data centers.

Takeaways

• Bloom Energy is mentioned as a potential supplier to data centers seeking power alternatives, but Eisman does not provide a specific recommendation or valuation view.

ServiceNow (NOW) and Salesforce (CRM)

• Eisman says some software companies are deeply embedded in enterprise operations, making switching costs high. He suggests this may apply to ServiceNow and Salesforce.
• He contrasts these companies with software businesses that have weaker competitive barriers and may face more pressure from AI-generated alternatives.

Takeaways

• In Eisman’s framework, customer switching costs are central to assessing software-company risk from AI.
• The companies he sees as more embedded may be relatively better positioned than software firms whose products are easier to replace, though he does not make a specific stock recommendation.

Adobe (ADBE)

• Eisman describes Adobe as perceived to be particularly vulnerable to AI competition and says its stock fell from an all-time high of $630 in February 2024 to around $245, a decline of about 60%.
• He mentions an open-source, free alternative to Adobe’s Creative Cloud suite, while noting that he does not know how good the alternative is.
• He uses Adobe’s share-price decline as a rough proxy for possible losses in private-equity-owned software companies, while noting that cybersecurity software is a notable exception.

Takeaways

• Adobe illustrates the market’s concern that AI could lower the cost of creating software and intensify competition.
• The relevant question from Eisman’s analysis is whether alternatives are good enough to make customers switch and whether Adobe can maintain pricing and customer loyalty.

Software and Cybersecurity Theme

• Eisman discusses the “SaaSpocalypse” narrative: that AI will make software cheaper to build and could replace existing products. He says this narrative may be too simplistic because some enterprise software has high switching costs.
• He says cybersecurity remains necessary because AI may make hacking easier, and identifies cybersecurity software as a notable exception in his discussion of private-equity-owned software companies.
• He warns that many software companies do not have the same customer lock-in as ServiceNow or Salesforce.

Takeaways

• Assess software businesses individually: Eisman’s key distinction is how difficult it is for customers to leave, not simply whether a company sells software.
• AI may create opportunities for new products while also putting pressure on incumbent software pricing and competitive positions.

Private Equity and Private Credit

• Eisman says private equity bought many SaaS businesses between 2017 and 2022, often using roughly 50% debt and 50% equity, with private credit frequently lending to the buyers.
• Using a 50% decline in software-company values as an illustrative assumption, he says the equity in many leveraged deals could be wiped out, with the issue becoming clearer when loans are refinanced.
• He says this could significantly hurt private-equity investors but does not believe the problem is large enough on its own to seriously damage the economy.
• He also warns that private credit and private equity are financing and chasing parts of the AI build-out.

Takeaways

• The transcript highlights refinancing and asset-value declines as risks for investors exposed to private equity and private credit.
• Eisman does not characterize this exposure as sufficient on its own to cause a major economic crisis.

Defensive Sectors and ETFs

• If the AI trade unwinds, Eisman says investors may reallocate toward perceived safety sectors, particularly healthcare and consumer staples.
• He also identifies property-and-casualty insurance as a defensive subsector within financials.
• He names three ETFs:
  • LVHD — Franklin U.S. Low Volatility High Dividend Index ETF
  • SPLV — Invesco S&P 500 Low Volatility ETF
  • KBWP — Invesco KBW Property & Casualty Insurance ETF

Takeaways

• Eisman presents these as reallocation options for investors seeking lower-volatility or defensive exposure, not as specific stock-picking recommendations.
• His broader point is that diversification may be limited when so much of the market and corporate-bond issuance is tied to AI.

Financials and Investment Banks

• Eisman says investment banks and large banks are exposed to AI financing, and that investment-banking activity has been strong partly because of financing needs related to AI.
• He says these stocks are near peak valuations and could decline if AI financing needs weaken.
• He also identifies alternative asset managers as part of the broader financial sector exposed to AI-related financing.

Takeaways

• The financial-sector exposure is indirect: a slowdown in AI infrastructure spending could reduce financing activity and affect firms involved in providing capital.
• Eisman’s stated concern is that current valuations may be vulnerable if the financing cycle weakens.