Steve Eisman, known for his prescient identification of market bubbles, including the 2008 housing crisis, has expressed significant concerns regarding the current artificial intelligence (AI) landscape. In a recent discussion, Eisman highlighted two primary areas of risk: extreme concentration and a fundamental lack of sustainable competitive advantages, or "moats," within the AI industry.

Concentration Risk in the AI Ecosystem

Eisman points to Nvidia as a key indicator of this concentration. While Nvidia reported impressive revenue growth of 110%, a deeper dive into their filings reveals a critical dependency: 70% of their accounts receivable, as of July, originated from just five customers.

This dependency extends to the major cloud providers – Google, Meta, Amazon, Microsoft, and Oracle. Eisman states that 70% of these hyperscalers' AI revenue, which constitutes approximately 25% to 30% of their total cloud revenue, is derived solely from Anthropic and OpenAI. He emphasizes that if either of these two AI companies falters, the entire chain, from chip manufacturers like Nvidia to the hyperscalers, could face significant disruption.

Oracle's financial performance further illustrates this point. The company reported a substantial increase in its Orderão (RPO), or backlog, from $150 billion to$400 billion in a single quarter. Subsequent analysis suggested that 50% of this RPO was attributed to OpenAI. Eisman notes that Oracle's current valuation of over $600 billion still relies heavily on OpenAI, a company that is reportedly losing substantial amounts of money.

The Financial Strain of AI Development

The financial models of leading AI companies like OpenAI and Anthropic are a significant concern for Eisman. He describes them as "bleeding money." OpenAI, for instance, reported $6.5 billion in revenue in the second quarter of the year, but incurred costs of approximately$12 billion. During this same three-month period, OpenAI's revenue increased by $1 billion while its costs rose by$3 billion. In contrast, Anthropic's revenue grew by over 100% in the same quarter, indicating it is the stronger entity of the two, though still operating at a loss.

The funding for these AI ventures is not primarily coming from venture capital. Instead, major technology companies such as Amazon, Google, Microsoft, and Nvidia, along with SoftBank, are investing heavily by taking equity stakes. Eisman likens this financial structure to a circular arrangement, where money and commitments flow in one direction, and then commitments flow back. He suggests that this circularity can hold as long as Anthropic and OpenAI continue to grow rapidly and attract further investment. However, a significant misstep by either company could lead to a withdrawal of funds and a collapse of the chain.

The Absence of Moats in AI

Beyond concentration risk, Eisman argues that the AI industry, despite hundreds of billions of dollars invested, lacks sustainable competitive advantages, or "moats." He contrasts this with companies like Google, which historically held a dominant position in search due to its superior product and lack of viable alternatives.

Eisman identifies several factors contributing to this lack of moats:

• Token Maximization Ending: Initially, companies like OpenAI and Anthropic significantly undercharged for their services. As they raised prices to cover their costs, businesses became more cost-conscious. This led to a decline in "token maximization," where employees were mandated to use AI extensively. Companies like Uber reportedly exhausted their AI budgets rapidly, and others spent substantial sums without clear ROI.
• Rise of Open-Weight Models: The increasing availability and capability of open-weight models, such as Kimmy K3, are providing cost-effective alternatives. Businesses are now opting for these models for less critical tasks, reserving premium services like OpenAI and Anthropic for highly important functions.
• Lack of Customer Loyalty: In the LLM (Large Language Model) space, there is little customer loyalty. If a superior LLM emerges, software developers are likely to switch. This contrasts with deeply embedded enterprise software, such as that offered by Salesforce or ServiceNow, where switching costs are prohibitively high.

The "AI is Ending the World" Narrative as a Strategy

Eisman posits that the widespread narrative of AI posing an existential threat to the world may be a manufactured hysteria. He suggests that companies are nervous about the end of token maximization, the lack of moats, and the competitive threat from Chinese open-weight models. By promoting the idea of an AI apocalypse, these companies may be attempting to influence government regulation. Their goal, according to Eisman, is to manipulate regulatory outcomes to create a duopoly, where government intervention, such as barring Chinese AI models, would effectively create the moats they currently lack.

He points to statements from Anthropic CEO Dario Amodei and Elon Musk advocating for a slowdown in AI development. Eisman finds these calls disingenuous, particularly if the companies are simultaneously pursuing IPOs. He argues that genuine concern for safety would lead to postponing such financial events.

Hyperscalers and the Power Problem

While hyperscalers like Microsoft, Amazon, and Google possess moats due to their scale and the immense cost of building data centers, Eisman notes a critical weakness: their current cash flow has been significantly impacted by AI investments. Some are even experiencing negative cash flow. Google, for example, recently raised $85 billion in equity capital, a move Eisman finds remarkable for a company that historically did not need to raise capital.

The immense demand for AI is also creating a significant "power problem" in the United States. Eisman agrees with Elon Musk's assessment that while China faces a chip problem, the U.S. faces a more fundamental challenge with power infrastructure. The demand for electricity to power data centers is substantial, potentially increasing U.S. electricity consumption by 3% to 4% annually, equivalent to adding two large cities' worth of demand. Companies like GE Vernova, which manufactures gas turbines and electrical equipment essential for power generation, are experiencing a surge in orders, with backlogs extending beyond 2030.

Investment Strategy: Picks and Shovels

When considering investments in the AI space, Eisman advises a "picks and shovels" approach, focusing on companies that provide the infrastructure and tools for AI development rather than the AI software providers themselves. He would avoid investing directly in LLM providers due to the lack of moats and is wary of hyperscalers due to their current cash flow issues.

Instead, Eisman suggests looking at companies like:

• Nvidia: For its role in providing AI chips.
• Micron: A memory chip manufacturer.
• GE Vernova: For its power generation equipment.
• Arista Networks and Cisco: For networking infrastructure.
• Eaton: An electrification company.

He also expresses skepticism about the broader software industry, particularly SaaS (Software as a Service), citing the potential disruption from agentic AI that can perform tasks currently handled by various online platforms. However, he believes deeply embedded enterprise software is likely to remain resilient.

Broader Market Concerns

Eisman touches upon other market concerns, including the U.S. national debt, which he believes is less of an immediate threat due to the U.S. dollar's reserve currency status and the role of U.S. Treasuries in the global financial system. Regarding interest rates, he views the 10-year Treasury yield as a critical indicator, suggesting that a sustained level above 5% could trigger a market correction.

He also dismisses the idea of precious metals or Bitcoin as productive investments, likening them to assets that "sit there and do nothing." He finds Bitcoin's inverse correlation to its own thesis particularly perplexing, suggesting it functions primarily as a speculative asset.

Ultimately, Eisman's analysis of the AI landscape is one of caution, emphasizing the significant risks associated with concentration and the absence of durable competitive advantages, while acknowledging the potential for companies providing the underlying infrastructure to benefit.