Blackstone Warns Anthropic Won't Secure Investment-Grade Rating: Cracks in the AI Bubble Are Showing

Blackstone Warns Anthropic Won't Secure Investment-Grade Rating: Cracks in the AI Bubble Are Showing

In the high-flying narrative of the AI industry, a warning from Blackstone to investors lands like a bucket of cold water: Anthropic will most likely fail to secure an investment-grade credit rating after its IPO. According to FT, an investment-grade rating is critical to Anthropic's ambitions—only with it can the company borrow at low interest rates to fund its massive infrastructure projects. If it falls into the junk bond market, hundreds of billions of dollars in debt will become a crushing burden.

This isn't an isolated piece of bad news. Combining Anthropic's S-1 filing, the sharp critique from Apollo chief economist Torsten Slok, and the weak performance in the latest earnings reports from hyperscale cloud providers, a clearer picture is emerging: the AI industry's prosperity—propped up by capital subsidies and circular deals—is facing a reality check.

1. Anthropic's IPO Filing: The Deliberately Hidden Truth

Why There's No Q3 Data in the S-1

Anthropic has filed its S-1 registration statement for an IPO, but the document does not disclose third-quarter revenue, nor does it mention details of future debt on a scale of hundreds of billions. According to podcast host Ed Zitron, information a company deliberately withholds is just as important as what it voluntarily discloses. "When a company is unwilling to publish a certain set of numbers, it means those numbers don't look good."

Anthropic's impressive second-quarter figures were largely thanks to Elon Musk's compute fee discounts in May and June. This information comes from the S-1 filing SpaceX submitted to the U.S. government and can be publicly verified. But the discount period has ended, and in the third quarter Anthropic's compute costs have soared. Some reports note that in Q3, every dollar spent on compute yielded only $1.70 in revenue, compared with slightly above $2 in Q2.

The S-1 is the IPO registration statement required by the U.S. Securities and Exchange Commission (SEC), in which a company must disclose its business model, risk factors, financial condition, and other key information. However, the S-1 allows a company to initially omit the most recent quarter's financial data, as long as it is supplemented via amendment before the filing goes effective. This flexibility is often used by companies to avoid disclosing unfavorable data, especially when quarterly performance has deteriorated significantly. Anthropic's failure to disclose Q3 revenue may be precisely an exploitation of this rule to delay disclosure of the financial pressure caused by soaring costs.

The $430 Billion Non-Cancellable Contract and the Vanished Debt Chart

The S-1 states that Anthropic has $430 billion in non-cancellable contracts, of which $161.2 billion comes from Broadcom's leasing business and excludes interest. The filing also includes a chart titled "Debt 2026–2029," but all the specific figures have been removed.

"These numbers must be disclosed before the IPO, but I'm sure they'll fill them in on the last day before the deadline," Ed Zitron said on the show, using a metaphor: it's like a person saying he's doing fine while holding a thick rope you can't see, with a boulder hanging over his head, insisting everything is completely fine—but the moment you turn away, he lets go.

The Shiny—and Suspicious—Cash Reserve

Anthropic holds $110–120 billion in cash, and the interest it earns exceeds the revenue of many companies. But the S-1 won't disclose how much cash remains at the end of Q3 2026. To industry observers, this kind of maneuver is nothing new in the AI world.

2. Blackstone's Warning: What It Means to Miss Investment-Grade

Soaring Borrowing Costs and Forced Institutional Selling

During the recording of the show, breaking news arrived: Blackstone told investors that Anthropic will most likely fail to secure an investment-grade credit rating after its IPO. This precisely confirms the prediction Ed Zitron made in an article weeks earlier.

The impact of missing investment-grade is multifaceted. Borrowing interest rates will rise sharply, and contract terms will become harsher; pension funds and investment-grade-themed funds will no longer be able to buy its bonds, and once the rating falls to junk, a large number of institutions will be forced to sell. Take a bond with a face value of $1,000: if it can be bought on the market for $950, the actual yield will rise—but the chain reaction from selling is catastrophic for the issuer.

BB Is Already Lenient; OpenAI Will Be Worse

Some reports estimate Anthropic's rating could land at BB, which is speculative grade with relatively high default risk. Ed Zitron believes even this estimate is lenient, and even doubts whether Anthropic can reach BB.

More critically, Blackstone itself holds stakes in Anthropic, OpenAI, and other companies. "When your biggest backer comes out and says the odds of getting a quality rating are slim, that's exactly the risk I've been warning everyone about." For OpenAI, if Anthropic can't get investment-grade, OpenAI will only be worse.

An investment-grade credit rating generally refers to BBB- or above (S&P) or Baa3 or above (Moody's), representing lower default risk. A BB rating is speculative grade (junk), meaning the issuer has a relatively high default risk and must pay a higher risk premium. Institutional investors such as pensions and insurance funds are subject to regulatory restrictions and often can only hold investment-grade bonds. If Anthropic's rating falls to BB, its bonds will be excluded from investment-grade indices, triggering forced selling by passive funds and further pushing up financing costs.

3. The Bubble Slok Punctures: 493 Companies See No AI Dividend

AI Only Circulates Within the Circle

Apollo chief economist Torsten Slok offers a core summary: the future of the AI industry is not determined by the Magnificent Seven, but by the remaining 493 companies in the S&P 500. And at present, these 493 companies show no signs of revenue or productivity gains.

Slok's argument is clear: AI today only creates services sold to companies that make AI products—to a large extent, companies within the circle buying and selling from each other. Venture-backed startups use subsidies to purchase services, the money flows to Anthropic and OpenAI, and these two then pay Microsoft, Google, and Amazon to buy NVIDIA chips—a closed loop.

Data Confirms: 80% of Revenue from 1% of Customers

One set of figures in Slok's post: within total token consumption, the top 1% of customers account for 80% of spending. Combined with Root Figures data, 80% of AI revenue comes from the top 1% of customers, and within that top 1%, 80% of spending originally came from 10% of customers—essentially, venture capital money is footing the bill. A rough calculation: excluding the remaining 99% of customers, 64% of the AI industry's total spending originates from venture capital.

"If most consumption doesn't come from customers who genuinely find the product useful and are willing to pay, but from third-party capital underwriting it, then it isn't a real business." Ed Zitron compares this logic with Spotify: Spotify used venture money to change user consumption habits and then made money from users, whereas the AI situation is more like Spotify's customers spending money that all came from venture capital, while everyone shouts that Spotify is making a fortune.

Round-tripping refers to companies buying products or services from each other to inflate revenue. In the AI ecosystem, venture-backed startups purchase AI services, the money flows to Anthropic and OpenAI, which in turn buy compute from cloud providers, and the cloud providers invest back into AI companies, forming a closed capital loop. While this model can inflate revenue in the short term, it lacks support from genuine external demand—once venture funding tightens, the entire chain could collapse quickly.

4. Cloud Providers' Earnings: AI Is Hurting the Core Business

Google's AI Search Erodes Ad Revenue

After Alphabet and Meta released their latest earnings, their stock prices fell by 10% to 13%. Google management acknowledged in its earnings report that the AI search model is eroding its own revenue—users no longer click through a bunch of web links laden with ads; AI directly generates lengthy answers, hurting ad revenue.

"I said this back in 2023. AI doesn't help companies improve their business; on the contrary, the evidence now shows AI harms business," Ed Zitron commented.

Double-Counted Cloud Growth

There's another detail in Anthropic's S-1: all revenue generated from selling models through Bedrock and Vertex is recorded on Anthropic's own books. In the first half of 2026, 48% of Anthropic's revenue came from Google and Amazon, which provides a source for cloud providers' cloud revenue. But Deutsche Bank originally expected cloud business growth of 42%, while actual growth was only 39%.

Growth comes from two things: Anthropic's compute spending, and the fact that this Anthropic revenue is counted once in Anthropic's statements and again in Amazon's and Google's—double counting. Amazon invested $50 billion in OpenAI this year, $5 billion in Anthropic with plans to add $10 billion, and Google invested $10 billion with a reserved capacity to add $30 billion. The growth is artificially stacked up through large investments.

Bedrock is a managed service offered by Amazon AWS that allows developers to access multiple AI models via API; Vertex AI is a similar platform from Google Cloud. When Anthropic sells models through these two channels, the revenue is fully recorded in Anthropic's statements, while the cloud providers also count the related revenue in their own cloud business revenue. This double counting causes the same economic benefit to be counted multiple times at different links in the industry chain, artificially inflating the scale of cloud market growth.

5. The Urgency to Cash Out

Anthropic is rushing to go public, with the core goal of a quick exit to obtain liquidity. The lock-up arrangement shows that most shares can be unlocked within six months; as long as the IPO doesn't completely collapse, early investors who need to cash out can sell smoothly.

The third-quarter data is simply too worrying; if they delay further, the S-1 will have to include both Q3 and Q4 data—that's why they're rushing to meet the schedule. Ed Zitron's conclusion is blunt: this is very unfavorable for Anthropic, but they have no choice—they must complete the listing at all costs.

As for OpenAI, Sam Altman faces a similar dilemma. According to The New York Times, they originally aimed for a valuation of at least $1 trillion, but advisors directly urged him not to do so. Once publicly listed, a large amount of internal data must be disclosed, and the market will set a real price. The GPT Astra release brought a wave of hype, but OpenAI cannot produce a quarterly report as impressive as Anthropic's Q2—and that report itself was achieved through compute discounts.

The financing environment for the AI industry is tightening, and Blackstone's warning is just one signal. When the biggest backer starts flagging risks, when cloud providers' earnings begin to reveal AI's erosion of their core business, and when nearly two-thirds of industry spending is pointed out as coming from venture capital rather than genuine market demand, the cracks in the bubble can no longer be ignored.