AI Bubble: This is bad for Anthropic on every level | Ed Zitron

Summary

The discussion argues that Anthropic’s IPO filing and the possibility of a below-investment-grade credit rating highlight the financing risks behind AI infrastructure spending. Ed Zitron says Anthropic’s reported financial picture may be unusually favorable: its second-quarter figures reflected discounted computing costs, while its S-1 omitted third-quarter financial statements and figures for some future commitments. The speakers argue that rising compute expenses, large long-term obligations, and uncertain returns could make further borrowing difficult and expensive.

They also question whether AI is generating broad economic benefits. Citing Apollo chief economist Torsten Slok’s analysis, they say the S&P 493—the S&P 500 excluding the Magnificent Seven—has shown no net profit growth, growth in income per employee, or measurable productivity gains attributable to AI. In their view, AI-related revenue and cloud growth may partly reflect companies and venture-capital-backed startups funding one another, rather than durable end-customer demand.

The speakers connect these concerns to recent pressure on large technology stocks. They argue that AI investment is adding costs while some core businesses face challenges, and that AI features such as Google’s AI search responses may generate less revenue than traditional search results. Their overall assessment is that investor patience with AI spending and promised future returns is weakening.

Finance-specific figures and claims

Anthropic commitments and costs

• Anthropic’s S-1 reportedly disclosed $430 billion in non-cancellable commitments, including $161.2 billion in Broadcom equipment leases, excluding interest. Later in the discussion, the speakers refer to commitments of roughly $403–409 billion; the figures are inconsistent in the subtitles.
• The S-1 reportedly lacked figures for commitments across 2026–2029.
• Anthropic’s revenue reportedly rose from 11.5 to 16.8 quarter-on-quarter; the subtitles do not specify the units.
• The speakers estimate roughly $2 of revenue per $1 of compute cost in Q2, declining to about $1.70 per $1 in Q3.
• Anthropic reportedly had $112–120 billion in cash at an earlier point, after raising roughly $30–40 billion in February and another $65 billion in May. Zitron says the filing did not disclose its cash position as of September 30, 2026.
• Anthropic reportedly planned to build its own data centers, which the speakers say could further increase capital requirements.
• The discussion says SpaceX may charge Anthropic up to $3.1 billion per month for computing. Bloomberg was cited as reporting that Anthropic accounted for 34.5% of SpaceX revenue.

Credit and debt

• Blackstone reportedly told investors that Anthropic might not receive an investment-grade rating after its IPO. A possible BB rating was discussed, but the speakers emphasize that S&P and Moody’s had not finalized a rating.
• A below-investment-grade rating would generally mean higher borrowing costs and less favorable terms, and could exclude the bonds from some pension funds and investment-grade funds.
• The speakers describe the risk of “fallen angel” selling if a company with investment-grade debt is downgraded, citing Oracle as an example of a company concerned about this exposure.
• They say a lower rating could also raise costs for counterparties or guarantors, including Nvidia and Broadcom.
• Their central credit concern is that a company with negative EBITDA would need to demonstrate real profitability to improve its rating, while large compute commitments could become liabilities.

IPO valuation and liquidity

• Anthropic’s valuation was said to have reached about $2 trillion in some limited markets, then fallen to roughly $1.5 trillion.
• OpenAI’s latest funding-round valuation was recalled as around $730–765 billion. The speakers say it would need a valuation of about $1.5 trillion for investors to make a “reasonable profit,” while the company reportedly sought a $1 trillion valuation and advisers cautioned against that target.
• Anthropic’s IPO is characterized as a way to provide liquidity before costs and obligations become more visible. The discussion says the lock-up is six months, with most shares becoming available during the first few months.
• OpenAI reportedly postponed a planned IPO. The speakers speculate that its larger existing commitments and less favorable timing may have contributed.

Big Tech spending and results

• Google was described as spending $50 billion per quarter on capital expenditure—a figure presented by a speaker and not independently verified in the subtitles.
• Amazon reportedly cited $77 billion in capital expenditure, with more expected the following quarter. The discussion says annual capex of $220 billion would be lower than the following year’s spending.
• Amazon investments mentioned: $50 billion in OpenAI, $5 billion in Anthropic, and a possible additional $10 billion.
• Google investments mentioned: $10 billion, with an option to invest another $30 billion.
• Alphabet and Meta were said to have reported profit declines in the 10–13% range; the exact measure and period are unclear in the subtitles.
• Cloud growth was reportedly 39%, below a 42% forecast. The speakers argue that AI-compute spending partly supported cloud growth.
• Anthropic reportedly counted 100% of revenue from model sales through Amazon Bedrock and Google Vertex. The discussion says 48% of Anthropic’s first-half 2026 revenue went through Google and Amazon.
• Microsoft reportedly said 11% of its total revenue came from OpenAI, not solely from cloud revenue.
• A cited estimate put AI pricing at about $29 per employee per month; a GPU index was said to show GB300 at $1 per hour.

AI demand and returns

• Slok’s analysis was described as finding no net profit growth, income-per-employee growth, or productivity growth among the S&P 493.
• A chart attributed to Slok reportedly showed that 80% of token spending by the top 1% of customers came from venture capital. Another cited figure said 80% of AI revenue came from the top 1% of customers. The speakers combine these figures into a rough estimate that 64% of AI spending may ultimately be venture-capital-funded, while acknowledging that this arithmetic is approximate and may be an underestimate.
• They argue that venture-capital-funded spending is weaker evidence of sustainable demand than purchases paid for from operating returns.

Framework and investment considerations

• Assess creditworthiness: Check the proposed credit rating, borrowing costs, debt capacity, covenant terms, and whether institutional bond buyers can hold the debt.
• Look beyond headline growth: Compare revenue growth with compute costs, capital expenditure, and operating profitability; examine whether recent results benefited from discounts or prepayments.
• Read filings for omissions and obligations: Review the latest available financial statements, cash position, non-cancellable commitments, lease obligations, and disclosed future spending.
• Test the quality of demand: Distinguish customer spending funded by operating returns from spending supported by venture capital or investment subsidies.
• Evaluate returns on AI investment: Look for measurable productivity, profit, or income-per-employee gains—not only anecdotes, usage growth, or cloud revenue.
• Consider counterparties and concentration: Account for dependence on cloud providers, chip suppliers, financing partners, and a small number of customers.
• Monitor liquidity and timing: Consider whether an IPO is being pursued before less favorable financial periods or larger obligations appear in required disclosures.

Assets, companies, sectors, and instruments mentioned

• Companies and organizations: Anthropic, OpenAI, Alphabet/Google, Amazon, Microsoft, Meta, Oracle, Nvidia, Broadcom, SpaceX, Blackstone, Apollo, Lambda, CoreWeave, and Deutsche Bank.
• Products and platforms: Claude, GPT Astra, Amazon Bedrock, Google Vertex, and Google AI Mode/search.
• Sectors and themes: Artificial intelligence, cloud computing, semiconductors, data centers, venture capital, and technology stocks.
• Instruments and financial measures: IPO shares, corporate bonds, junk/high-yield bonds, investment-grade debt, debt and leases, credit ratings, capital expenditure, EBITDA, and valuation.
• Tickers: No stock tickers are explicitly given in the subtitles.

Disclosures and caveats

No explicit “not financial advice” disclaimer is included. Several figures are presented with uncertainty or inconsistent wording in the subtitles; some are the speakers’ estimates or interpretations rather than independently verified facts. The credit-rating discussion is explicitly described as preliminary, with final S&P and Moody’s decisions still pending.

Presenters and sources

Presenters: Isaac Pound and Ed Zitron, host of the Better Offline podcast and author of Where’s Your Ed At.

Sources cited or discussed: Blackstone investor communications; Anthropic’s S-1; SpaceX’s S-1; Torsten Slok of Apollo; the Financial Times; Bloomberg; The New York Times; The Wall Street Journal and reporter Berber Jin; Ramp; and the Lebeus GPU pricing index.