Anthropic's $800 Billion Problem Is Hiding In The Fine Print

SUMMARY

Ed Elson hosts a discussion with Paul Kedrosky and Robert Armstrong, analyzing Anthropic's S-1 filing and the broader implications for AI and stock markets. The episode highlights Anthropic's looming $800 billion compute commitments, the fragility of relying on competitors for infrastructure, and growing caution in equity markets as AI-driven optimism faces financial headwinds.

MAIN POINTS

• Anthropic's S-1 reveals significant revenue growth but also large operating losses and a lack of detail on future compute obligations. [2:00]
• Anthropic's filing discloses over $800 billion in future compute commitments, primarily with Broadcom and Google, which are not yet reflected in the income statement. [8:00]
• The company is expected to shift from leasing compute from competitors to building its own data centers, dramatically increasing capital expenditures. [13:00]
• The discussion transitions to broader market concerns, with tech stocks experiencing volatility and investors moving into safer assets like treasuries. [18:00]
• Rising bond yields and declining performance in industrial and small-cap stocks signal a potential turning point in equity markets. [24:00]
• Elon Musk warns that debt markets are closing for AI data center companies without independent revenue, suggesting imminent consolidation and failures among 'neo cloud' firms. [27:00]
• A possible correction in the stock market is discussed, with the expectation that hyperscalers may weather the storm while semiconductor and smaller AI firms face significant risks. [32:00]

DETAILED ANALYSIS

Anthropic’s S-1 filing has drawn significant attention due to both its headline financials and the critical omissions in its disclosures. The company reported $16.3 billion in revenue for the first half of the year, a dramatic increase from $1.2 billion in the same period last year. However, it also posted a net loss exceeding $80 billion, the majority of which—$72 billion—was attributed to non-cash charges related to convertible notes and other fundraising instruments.

The more telling figure is the operating loss of approximately $700 million, which, while narrower than the previous year’s $2.1 billion, is expected to widen as the company transitions its business model.

A key concern raised in the discussion is the lack of transparency regarding Anthropic’s total contractual compute obligations. The S-1 does not specify the full extent of these commitments, nor does it break out training costs or identify major customers. However, deeper analysis of the filing reveals that Anthropic has entered into agreements totaling over $800 billion in future compute spending, primarily with Broadcom and Google.

These are non-cancellable commitments that do not yet appear on the income statement because they represent future obligations rather than current expenditures. This accounting treatment, while not deceptive, significantly understates the company’s future cash flow requirements and risk profile.

Anthropic’s current strategy relies heavily on leasing compute capacity from major competitors such as Microsoft, Google, and Amazon. This arrangement is inherently fragile, as these partners are also direct competitors in the AI space and can exert significant leverage over Anthropic’s operations and margins. The analogy to Spotify’s relationship with music labels highlights the risk: unlike Spotify, which pays a percentage of revenue, Anthropic faces fixed commitments regardless of its own sales performance, and its distributors have the power to exclude it from their platforms at any time.

The company’s own statements in the S-1 indicate a strategic pivot is underway. Anthropic expects to shift from leasing compute to making substantial capital expenditures on its own data centers and infrastructure. This move is driven by the need to reduce reliance on competitors and to capture more value from its AI products.

However, this transition will require significant upfront investment and ongoing maintenance costs, fundamentally altering the company’s financial profile. The current income statement, therefore, is not representative of the future business, which will be far more capital intensive and exposed to operational risks.

The timing of the IPO appears to be influenced by this inflection point. By going public now, Anthropic can present a relatively favorable picture of its operating losses before the full impact of its capital commitments is reflected in its financials. The company’s reported valuation targets, ranging up to $2 trillion, are subject to skepticism given the uncertainties around its future profitability and the broader market environment.

The episode then shifts to a broader discussion of market conditions, with Robert Armstrong noting increased volatility in tech stocks and a significant rally in treasuries as investors seek safety. The NASDAQ and S&P 500 have shown signs of strain, particularly in sectors linked to the AI boom, such as industrials and small caps. The sudden shift in sentiment is attributed to concerns over the sustainability of AI-driven growth and the rising cost of capital.

Elon Musk’s comments underscore the tightening conditions in debt markets for AI data center companies, predicting widespread consolidation and failures among smaller players. The analogy to the railroad boom of the 19th century suggests that while some large, well-capitalized firms will survive and consolidate the industry, many newer entrants will not withstand the financial pressures.

Looking ahead, the consensus is that while the largest technology companies—Alphabet, Amazon, Meta, and Microsoft—are likely to remain resilient due to their strong cash flows and diversified businesses, the semiconductor sector and smaller AI firms face significant risks. A correction in the stock market is possible, but it is expected to be uneven, with the most severe impact on companies heavily exposed to the AI infrastructure buildout. Investors are advised to consider the relative safety of treasuries, which now offer attractive yields and a degree of insulation against further market turbulence.

LINKS

• Prof G Markets YouTube channel
• Prof G Markets newsletter subscription
• Order 'Notes On Being A Man'
• Subscribe to No Mercy / No Malice
• Prof G Markets on Instagram
• Scott Galloway on Instagram
• Ed Elson on Instagram
• Ed Elson on X (Twitter)
• Ed Elson on Substack
• SoFi checking and savings account offer

KEYWORDS