Anthropic's $80B Loss: $72B of It Is an Accounting Charge

This week’s “the AI bubble is bursting” headlines rest on three stories: the loss reported from Anthropic’s draft IPO filing, the IMF’s downside scenario for AI investment, and a sharp jump in CoreWeave’s bond yields. All three are real, and all three got bigger on their way into the headlines. For a brand that works with AI tools, the takeaway is practical: avoid depending on a single provider, and be ready for price and plan changes.

What happened?

Anthropic’s draft IPO filing

On October 20 (US time), The Wall Street Journal reported on Anthropic’s non-public draft IPO filing. According to Al Jazeera and investingLive, which relayed the figures, revenue for the first half of 2026 was about $16.3 billion, up from about $1.1 billion a year earlier. The net loss was more than $80 billion (investingLive puts it at about $81 billion), compared with about $13 billion the year before. Roughly $72 billion of that loss is an accounting charge tied to changes in the value of convertible notes and other instruments the company used to raise money.

On the operating side, compute and infrastructure spending came to nearly $9 billion in the half, against about $2.3 billion a year earlier. Total compute commitments exceed $600 billion, about $450 billion of which are cloud agreements that largely can’t be cancelled. Two customers accounted for roughly a fifth of first-half revenue. The filing is a draft, the figures haven’t been independently verified, and Anthropic did not comment. Press reports put the IPO in November.

The IMF’s downside scenario

The IMF published its World Economic Outlook on October 13. The baseline forecast has global growth at 3.1% in 2026 and 3.4% in 2027. Box 1.2 in Chapter 1 tests the risks around that baseline with scenarios. In “Scenario C: Market correction”, the productivity gains expected from AI fail to arrive and AI investment unwinds. Global output then ends up 1.1% below the baseline by 2028, with larger losses in the United States (1.8%) and Asia (1.7%).

The Global Financial Stability team’s blog post reviewed 14 “circular” AI deals since 2023 and found about $300 billion in equity and $650 billion in debt, in exchange for about $1.4 trillion in future revenue commitments. On October 15, Managing Director Kristalina Georgieva wrote that hyperscalers’ share of investment-grade bond issuance has risen from under 1% to 10% over the past two years.

CoreWeave’s bonds

Shares of CoreWeave, which supplies compute capacity to AI companies, fell 15.5% on October 14. According to CNBC, the yield on its 9.625% bond due 2032 rose to 14.8%, and the yield on its 2031 bond to around 14.9%. CEO Mike Intrator wrote in a memo obtained by CNBC: “We are confident in our backlog which more than supports all of our debt obligations.” CNBC reported on October 21 that the stock was down 36% for the month. The company is now worth about $31 billion, against a post-IPO peak of about $90 billion.

What the headlines overstate

What’s circulating	What the source says
“Anthropic burned $80 billion in six months”	About $72 billion of the loss is an accounting charge tied to the value of financing instruments. Cash flow wasn’t reported. By our own rough math, taking that charge out leaves a loss of close to $9 billion. That’s our calculation; the filing doesn’t contain such a figure.
“The IMF warned of an AI bubble”	The word “bubble” doesn’t appear in the IMF text. The 1.1% loss comes from one of the downside scenarios around the baseline, and the baseline itself has the world economy growing 3.4% in 2027.
“CoreWeave’s borrowing cost jumped from 9% to 15%”	The figure above 9% is the bonds’ coupon, the rate the company pays on them. The figure near 15% is the yield a buyer of those same bonds would earn on the secondary market today. The gap shows the market pricing in more risk. The coupon on the existing bonds stays the same.
“The AI bubble is popping”	The three stories show investors scrutinizing AI spending more closely. The same filing shows Anthropic’s revenue growing nearly fifteenfold in a year.

The figures we’d watch more closely than the loss are the compute commitments and the customer concentration. A commitment above $600 billion, most of it non-cancellable, requires revenue to keep growing fast for years. Two customers making up a fifth of revenue means part of that growth rests on a small number of large deals.

What it means for your brand

A brand using Claude, ChatGPT, Gemini or a video model will feel these stories in the price list well before it feels them in the stock market. Companies with huge commitments and investors to convince may change prices, plans and free offers more often. There’s a recent example. On October 9, Anthropic paused the free Claude Team year and $1,000 API credits in its Claude Startups program, saying it had underestimated demand. The offers had been announced on October 6; we covered the details in our Claude Startups post. That’s why we’d advise against building your workflow around one provider’s current prices and quotas.

What doesn’t change

None of these stories says a tool you use today is going away. A campaign film, image or website made with AI keeps working when its provider’s share price falls.

What you can do

• List which parts of your work depend on which AI tool (copy, images, video, voice, code), and name at least one alternative for each.
• Keep prompts, style references, character designs and voice samples in your own archive, outside the tool.
• Don’t lock your annual budget to today’s prices. Leave room for changes in pricing and quotas.
• Don’t plan around free credits and trials as a lasting cost advantage. In the Claude Startups case, the offers were paused three days after launch.

Our recommendation at Klyne

On AI animated film projects, we recommend building the work so it doesn’t depend on any one model. What stays with the brand is the set of decisions about character, style, color, sound and story; the model is one of the tools used to produce them. We made two separate AI animated films for Gala Fresh (MBA Tarım), one for Fruit Attraction Madrid and one for Fruit Logistica Asia. Keeping the same visual world across several films for one brand gets easier when those decisions are documented independently of the tool, and if the tool changes, those assets are still yours.

This post is not investment advice.

FAQ

Is Anthropic going under?

The reporting on the draft filing doesn’t say so. Revenue grew nearly fifteenfold in a year, and most of the loss is an accounting charge. The compute commitments are very large, though, and the filing isn’t public yet; the picture will be clearer once the prospectus is published.

Is the IMF predicting that an AI bubble will burst?

No. The IMF calculates the impact of a downside scenario in which the expected productivity gains from AI don’t materialize, and it doesn’t use the word “bubble”. Its baseline forecast is 3.1% global growth in 2026 and 3.4% in 2027.

Should I cancel my AI subscriptions?

These stories don’t give you a reason to. A more useful step is to name an alternative for each tool and keep your files and prompts outside it.

If you’d like to set up your AI animated film or video project so it holds up when tools change, get in touch and we’ll plan it together.