Steve Eisman Warns of AI Bubble Burst

Nvidia's 10-Q filing, specifically Note 7, reveals that 70% of the company's accounts receivable as of the end of July came from only five customers. Nvidia reported 110% revenue growth, though the high concentration of industry capital presents a notable risk profile.

This level of customer dependence raises a warning flag about the sustainability of Nvidia's revenue streams, as a disruption with any of these key clients could have a substantial impact on its financial health.

Microsoft, Google, and Amazon derive 70% of their AI-related revenue from just two companies: Anthropic and OpenAI. This substantial portion of AI revenue constitutes 25% to 30% of their total cloud revenue.

Oracle also exhibits a similar dependence, with its Remaining Performance Obligation (RPO) heavily tied to OpenAI.

A significant portion of the growth seen by hyperscalers is tied to the performance and funding of a select few AI startups.

The entire AI supply chain, from Nvidia to hyperscalers, hinges on the sustained health and growth of OpenAI and Anthropic. These companies are responsible for the large commitments to hyperscalers, forming the base of the current investment structure.

OpenAI, in particular, experienced an 'upside down' financial period where its costs surged by $3 billion while revenue only grew by $1 billion. The financial model faced significant pressure as expenses outpaced income.

Eisman identifies OpenAI as the weaker entity compared to Anthropic, citing its slower revenue growth as a primary concern for the ecosystem's stability.

Investment in AI startups primarily originates from major tech companies such as Amazon, Google, Microsoft, and Nvidia, alongside SoftBank.

This creates a circular funding dynamic where hyperscalers invest in AI companies, which then use that capital to purchase cloud services from those same hyperscalers.

The model risks collapse if these primary startups lose their ability to secure new funding, as the flow of capital and demand for cloud services could diminish.

Despite hundreds of billions of dollars in investment, the large language model (LLM) industry struggles with 'token maxing' and cost consciousness, revealing a fundamental lack of defensible competitive advantages.

Users frequently switch between various LLM models like Gemini, ChatGPT, and Claude, and software developers show little loyalty to specific platforms. This ease of switching prevents the formation of sustainable business moats for LLM developers.

Initial AI adoption saw firms engaging in 'token maxing,' where usage occurred without strict regard for actual need or cost. This led to OpenAI and Anthropic raising prices after undercharging for token costs.

Companies have since become more cost-conscious, leading to a decline in 'token maxing' and an increased preference for open-weight models for non-critical tasks. Some firms rapidly depleted their AI budgets due to early overspending.

Steve Eisman asserts that the narrative portraying AI as an 'end-the-world' threat is a manufactured hysteria, designed to influence government regulation.

He believes that AI companies are pushing for regulation to effectively ban Chinese open-weight models, thereby creating a duopoly that protects current market leaders.

This regulatory framework would serve as a synthetic moat, solidifying the market position of existing dominant players in the AI industry.

Hyperscalers maintain a significant competitive moat due to the massive capital expenditures required to build data centers, a barrier to entry that other companies cannot easily overcome.

Software companies like Salesforce and Service Now also possess moats through deep integration into business processes. Additionally, Nvidia holds a unique moat as a key GPU manufacturer.

In contrast, LLM creators generally lack sticky, long-term competitive advantages that would prevent customer churn and ensure sustained market dominance.

Oracle's stock has experienced high volatility following its earnings report, despite otherwise positive results. This reflects significant investor nervousness.

Concerns center on Oracle's remaining performance obligation (RPO) backlog, where 50% is tied to OpenAI, raising questions about the stability of future revenue.

The company's BBB- debt rating from S&P, placing it close to junk status, further amplifies investor anxiety regarding the sustainability of the hundreds of billions in AI capital expenditure across the industry.

Market demand for chips currently outweighs obsolescence, according to Eisman, who views the debate over extended depreciation schedules as secondary.

However, Eisman argues that the current robust demand for chips significantly outweighs obsolescence concerns, rendering the depreciation schedule a secondary factor.

Eisman views the debate over accounting nuances as secondary to fundamental demand, arguing that a major market correction would stem from an AI failure rather than depreciation methods.

The United States experienced its first increase in electrical production in 15 years, even prior to the boom in AI data centers. This underlying structural shift has been a significant driver.

Combined with the massive power demands from AI data centers, US electricity growth is now projected to be 3-4% per year. GE Vernova has directly benefited from this structural shift in power demand, moving from negative perception to high demand.

Eisman advises against investing in large language model (LLM) companies, citing their lack of competitive moats. He also recommends caution regarding hyperscalers, as their cash flow is currently undergoing heavy deployment.

Instead, he prefers investing in infrastructure firms that provide the 'picks and shovels' for the AI revolution. His favored picks include Nvidia, Micron, GE Vernova, Arista Networks, Cisco, and Eaton.

This strategy focuses on companies that benefit from the foundational energy and hardware needs driven by AI, rather than the software applications themselves.

While the current AI market shows parallels to the dot-com era, Eisman stated that he cannot yet determine if a crash is imminent.

The failure of major players such as OpenAI or Anthropic would trigger a significant market correction, according to Eisman. However, he also posits that such a failure would likely lead to a new generation of companies emerging to pick up the pieces, rather than a permanent collapse of the sector.

Global banking systems rely on US Treasuries to facilitate overnight repo markets, reinforcing their indispensable role in international finance.

While Eisman supports a smaller deficit, he asserts that there is no viable alternative to the dollar's status as the world's reserve currency, mitigating the risk of a debt spiral.

Eisman believes the government is attempting to manipulate long-term Treasury rates to reduce borrowing costs for the United States.

He describes the announced $6 billion buyback plan as 'nothing' in the context of a $40 trillion deficit, indicating its insignificance in affecting broader market rates. He asserts that previous government efforts to ease interest costs through such interventions have been unsuccessful.

Steve Eisman no longer sees a professional advantage in investing overseas, a shift from his past experience with hedge funds.

He expresses complete satisfaction with the ample investment opportunities available within the US market, finding enough to do without venturing internationally.