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

Ai Bubble: This Is Bad For Anthropic On Every Level | Ed Zitron

https://www.youtube.com/watch?v=884dTvxSk4c

TLDR AI companies’ soaring valuations and infrastructure commitments are hard to justify when financial disclosures are incomplete, credit may be weak, and much of their revenue appears tied to investor-funded spending rather than proven customer value. Anthropic’s rising compute costs and huge contracts, alongside OpenAI’s ambitious commitments, could make borrowing and future growth difficult; meanwhile, broad productivity gains remain elusive, major tech firms are spending heavily, and investors are losing patience with AI’s uncertain returns.

Key Insights

Start by separating reported results from the full financial picture

When evaluating an AI company, check whether its latest filing includes recent quarterly results, cash balances, and the timing of major expenses. The discussion highlights that Anthropic’s filing omitted Q3 results and its cash position, making it harder to judge its financial health. A strong earlier quarter may not reflect ongoing performance if it benefited from temporary discounts or other favorable arrangements. Treat incomplete disclosures as uncertainty, not as evidence that the missing figures are harmless.

Map compute commitments and their costs over time

Identify the company’s major cloud, chip, and data-center agreements, then distinguish cancellable commitments from obligations it cannot readily avoid. Anthropic was reported to have about $430 billion in uncancellable contracts, including large Broadcom-related leases, while its compute costs were expected to rise. Focus on when those payments come due and whether current revenue can cover them. Large commitments are especially risky when they are disclosed without clear cost schedules or complete recent results.

Test whether revenue reflects customer value or temporary funding

Ask who ultimately pays for AI products and whether customers continue paying because the products deliver measurable returns. The speakers argue that some AI demand may be supported by venture capital and investor funding rather than durable customer economics. They also note that AI companies can sell services to one another, making revenue appear stronger without proving broad, independent demand. Track customer renewals, usage limits, pricing, and return on investment to assess whether demand can persist if funding slows.

Look for circular revenue and dependence between partners

Trace the flow of money among AI companies, cloud providers, and investors instead of treating every reported sale as independent demand. The discussion notes that AI firms may buy compute from cloud providers that also invest in them, while model sales through those same platforms contribute to reported cloud revenue. This can make growth figures difficult to interpret. Compare partner-driven sales with revenue from customers that are not also investors, suppliers, or financing sources.

Assess credit quality and financing access before relying on debt

A below-investment-grade credit rating can increase borrowing costs, impose stricter lending terms, and limit access to investors such as some pension funds. The discussion reports concern that Anthropic could receive a speculative-grade rating after an IPO. For any heavily indebted or capital-intensive AI company, examine whether it can service debt through operating cash flow rather than future fundraising or optimistic accounting assumptions. Also consider whether lenders may receive financial details that public investors do not yet have.

Compare AI investment with productivity beyond the largest tech firms

Do not judge AI’s economic impact solely by the performance of a handful of major technology companies. The discussion cites an assessment that the broader S&P 493 has yet to show clear gains in revenue per worker, profits, or productivity attributable to AI. Look for measurable improvements across ordinary businesses, such as lower costs, increased output, or sustained revenue growth. If those gains remain unclear after substantial investment, expectations may be running ahead of demonstrated results.

Balance cloud growth against capital spending and business cannibalization

Cloud revenue can grow while the infrastructure required to support it consumes substantial capital, so evaluate both sides of the equation. The speakers point to rising spending by major technology companies and note that AI features may weaken existing revenue streams—for example, AI search may generate less revenue than traditional search. Compare incremental revenue and profit with capital expenditures, depreciation, and the possibility of displaced legacy sales. Growth is more convincing when it produces durable returns rather than merely requiring ever-larger investment.

Stress-test the business for a slowdown in AI funding and demand

Model what happens if venture funding, customer budgets, or investor enthusiasm decline, rather than assuming AI spending will continue to expand indefinitely. The discussion warns that funding-dependent demand could fall sharply if those sources retreat, putting revenue and infrastructure commitments under pressure. Consider scenarios involving lower usage, cheaper models, reduced cloud spending, and tighter credit. Companies with flexible costs, diversified customers, and demonstrated profitability are better positioned to withstand a downturn.

Questions & Answers

Why does the speaker argue that Anthropic’s S-1 may be misleading or incomplete for investors?

It omits crucial recent financial information, including Q3 results and the timing and amounts of major compute obligations. The speaker says this prevents investors from assessing rising compute costs, falling revenue per dollar of compute, and the risk that margins will compress.

What major compute commitments has Anthropic disclosed, and what additional reported payment raises concern?

Anthropic disclosed about $430 billion in uncancellable contracts, including $161.2 billion in Broadcom leases. It is also reportedly paying SpaceX as much as $3.1 billion a month for compute, which the speaker describes as part of a circular arrangement.

Why are the speakers concerned about Anthropic’s financial position and cash disclosures?

Anthropic has raised large sums and still holds substantial cash, but it has not disclosed its Q3 2026 cash position. The speakers suspect it may need customer prepayments and say discounted compute helped make its reported quarter look stronger than its underlying economics.

Why might Anthropic be rushing to go public?

The speakers suspect its Q3 results were concerning and that waiting could force it to disclose weaker Q3 and Q4 figures. They also note that the IPO lockup schedule may allow shareholders to sell relatively quickly.

What would a below-investment-grade credit rating mean for Anthropic?

It could face higher borrowing costs, tougher debt terms, and less access to pension funds and other buyers restricted to investment-grade debt. A reported double-B rating would be speculative and signal elevated default risk; improving it would require genuine profitability.

Why do the speakers think OpenAI may have delayed its IPO?

They suggest public markets may not support the trillion-dollar valuation reportedly sought by its advisers. OpenAI’s accumulated compute commitments also make it harder to present a convincing financial picture and raise questions about whether it can afford its obligations.

What does Torsten Slok’s analysis suggest about AI’s broader economic impact?

Slok argues that AI’s success cannot be judged by the Magnificent 7 alone. The broader S&P 493 has shown no clear gains in revenue per worker, profits, or productivity, and the promised benefits have yet to show up in wider economic data.

Why do the speakers say the economics of AI and GPU compute remain unclear?

Many companies are absorbing AI costs or limiting usage, and token-based billing remains uncommon. As a result, it is difficult to tell whether customers are receiving enough value to support the cost of compute.

How might AI weaken existing businesses even as it requires heavy investment?

Google executives acknowledged that AI Mode search generates less revenue than traditional search. This suggests AI could reduce revenue from an established business while also requiring substantial new investment.

How much do Anthropic’s sales through cloud platforms contribute to reported cloud growth?

Anthropic’s compute spending and model sales through Google Cloud and Amazon Bedrock contribute to both companies’ reported revenue. Its filings indicate that nearly half its revenue flowed through those platforms in the first half of 2026.

Why are investors questioning whether AI investment is paying off?

Recent results and share-price declines have heightened concern, while cloud growth depends on enormous capital spending. The speakers also argue that reported growth can be supported by AI companies buying from one another rather than by independent customer demand.

Why might AI demand be unusually dependent on venture capital?

Figures cited in the discussion suggest that most spending by the biggest AI customers is venture-funded. That means a substantial portion of AI demand may ultimately rely on investment money rather than customers paying from durable business revenues.

What is the risk if investor-funded AI spending declines?

The speaker estimates that perhaps around two-thirds or more of reported AI revenue may be funded by investors rather than by customers paying for demonstrated value. If even half of that funding disappears, revenue could fall by more than a quarter, creating potentially catastrophic losses.

What does the speaker say AI companies need to demonstrate for their businesses to be durable?

They need to show that customers will pay for real, demonstrable value without depending on venture capital, debt, or other outside investment. Those funding sources are finite, so assumptions of unlimited investment cannot sustain the industry indefinitely.

Summary of Timestamps

The discussion opens with concerns that Blackstone warned investors Anthropic might not achieve an investment-grade credit rating after its IPO. The speakers question how the company could rely on an already strained junk-bond market to fund its plans.

Anthropic’s S-1 allegedly omits important recent financial details, including Q3 results and the timing and size of its substantial compute obligations. The speakers argue that investors need this information to assess the company’s true costs and financial condition.

The speakers say Anthropic’s Q2 costs benefited from discounted compute, while compute expenses are rising and revenue generated per dollar of compute is falling. They point to the company’s own warning that margins could compress and that past growth may not continue.

Anthropic reportedly disclosed about $430 billion in uncancellable contracts, including $161.2 billion in Broadcom leases, and may be paying SpaceX as much as $3.1 billion per month for compute. The speaker calls this arrangement circular and says Anthropic may be going public before the full impact of its costs is reflected in its financial statements.

Although Anthropic may be complying with disclosure rules, the speakers argue that investors still lack essential information. They compare the omission to claiming everything is fine while hiding an anvil overhead, emphasizing the perceived gap between reassuring statements and financial risks.

Anthropic’s finances are difficult to evaluate because it has raised enormous sums and still holds substantial cash, but has not disclosed its Q3 2026 cash position. The speakers suspect it may need customer prepayments and criticize the limited transparency, while noting that discounted compute made the reported quarter look stronger.

The speakers suggest OpenAI delayed its IPO because public-market investors might not support the trillion-dollar valuation reportedly sought by its advisers. OpenAI’s accumulated compute commitments also make its financial promises more difficult to justify and its results harder to present convincingly.

A Financial Times report says Anthropic may not receive an investment-grade credit rating after its IPO. A below-investment-grade rating could mean higher borrowing costs, stricter debt terms, and less access to pension funds and other investors that are restricted to investment-grade securities.

The speakers describe a reported double-B rating as speculative and associated with elevated default risk. They argue that Anthropic would need genuine profitability—not accounting adjustments they characterize as “magic math”—to improve its credit standing, and say the implications could be even worse for OpenAI.

Anthropic’s expected weak credit rating could make borrowing more difficult just as the company appears to be moving quickly toward an IPO. The speakers speculate that concerning Q3 results may make waiting costly, since delaying could require disclosing weaker Q3 and Q4 figures.

The debt process could require Anthropic to share financial information with lenders that the public has not seen. That creates additional uncertainty for the company and for investors such as Blackstone and other firms connected to AI financing.

The discussion turns to Apollo chief economist Torsten Slok’s argument that AI’s economic impact cannot be judged by the Magnificent 7 alone. The broader S&P 493 has shown no clear improvement in revenue per worker, profits, or productivity, raising questions about whether AI’s promised benefits are reaching the wider economy.

The speakers suggest that AI companies may be generating revenue by selling to one another, while broader productivity gains remain difficult to see. After years of attention and heavy investment, financial analysts and businesses are becoming less patient with uncertain returns.

AI spending remains in a “messy middle”: many companies are pulling back, token-based billing is uncommon, and organizations often absorb the costs or restrict usage. Because the economics of AI and GPU compute remain unclear, workers may feel the slowdown when employers limit AI use or switch models.

Recent earnings have added to investors’ doubts about AI returns. Alphabet and Meta shares fell after results, while parts of Amazon’s and Google’s legacy businesses missed expectations; cloud remained a growth area, but it requires enormous capital spending.

Google executives acknowledged that AI Mode search earns less revenue than traditional search. This suggests AI may weaken an existing source of income even as companies spend heavily to develop and operate new AI products.

Some reported cloud growth may be less independent than it appears: Anthropic’s compute purchases and model sales through Google Cloud and Amazon Bedrock contribute to those platforms’ revenue. Anthropic’s filings reportedly indicate that nearly half its revenue flowed through the two platforms in the first half of 2026.

Amazon and Google are investing heavily in OpenAI and Anthropic while also providing infrastructure for their businesses, and Amazon has signaled that its already enormous capital expenditures will rise further. The speaker says investors are losing patience with claims that AI is broadly driving growth and points to Nvidia’s next earnings as an important test.

The discussion raises the possibility that AI demand depends heavily on venture capital. Figures cited by the speakers suggest that much of the largest AI customers’ spending is venture-funded, meaning a substantial share of AI demand could ultimately rely on investment money rather than established customer revenue.

The speaker distinguishes this situation from venture capital funding a service that later changes consumer behavior. In the AI example, customers themselves may be using venture funding to buy AI, leaving demand vulnerable if investors reduce their support.

The speaker estimates that perhaps two-thirds or more of AI revenue may be funded by investors rather than by customers paying for clearly demonstrated value. If investment and hype decline, the industry may lack a sustainable plan for replacing that demand.

Even if only half of the investor-funded demand disappeared, the speaker argues, AI revenues could fall by more than a quarter—a potentially severe shock. Appeals to business spending or high-profile firms such as Goldman Sachs and JPMorgan, they say, do not resolve the underlying question of whether customers receive enough value to keep paying.

The speaker concludes that businesses dependent on venture capital, debt, or other investors’ money may be closer to investment operations than durable businesses. These funding sources are finite, so AI companies’ assumptions of unlimited resources will eventually face financial reality.