Anthropic's IPO Filing Reads Like a Hazmat Warning Label

Anthropic just did something unusual for a company trying to go public: it filed paperwork that reads less like a sales pitch and more like a confession. The S1, released October 26, targets a valuation between $1.5 and $1.75 trillion under the ticker ANT, with Goldman Sachs and Morgan Stanley steering the deal toward a debut before Thanksgiving. But as Neeta Bidwai lays out on this episode of Good Revenue, the numbers inside that filing tell a story that should give even enthusiastic AI investors pause.

The Revenue Story Looks Great, Until It Does Not

Anthropic's growth is real. Revenue hit $16.3 billion in the first half of 2026, up from just $1.2 billion in the same period a year prior. That is a staggering jump. But net losses came in at $80.7 billion, with $72.4 billion of that chalked up to accounting quirks. Strip that away and the company still burned $9 billion in cash in six months, on top of $7.3 billion burned in all of 2025. Growth and survival, it turns out, are not the same thing.

The deeper problem is where that revenue comes from. The top five customers make up a quarter of total revenue, and Anthropic admits that customers churn aggressively based on which frontier model happens to be winning at any given moment. That is why the company shipped 250 model updates in just nine months. Standing still is not an option, and neither is slowing down the spending required to keep up.

A Staggering Pile of Commitments

The real shock in this filing is the compute math. Anthropic has committed to as much as $800 billion in spending over the next decade, and between $450 and $600 billion of that is non-cancellable. Reporting pieced together from Fortune and the Wall Street Journal points to massive individual obligations: roughly $114 billion to Google Cloud, $110 to $111 billion to Amazon Web Services, $161 billion to Broadcom, up to $120 billion tied to SpaceX's NVIDIA clusters, and about $117 billion through Microsoft and NVIDIA DL Clouds.

Layered on top of that is the now-familiar problem of circular financing across the AI industry. Amazon and Google are simultaneously Anthropic's investors, landlords, and customers. Anthropic acknowledges this arrangement could backfire, potentially leading to its own compute getting deprioritized if those relationships sour. That is not a hypothetical footnote. It is a structural vulnerability baked into the business.

Borrowing Costs and Existential Risk Factors

Anthropic needs to keep raising capital at rates that will not sink it, and that has proven difficult. Kroll Bond Rating Agency gave the company the highest junk credit rating back in June, and Anthropic reportedly offered Blackstone double-digit yields just to finance chips. High borrowing costs make it harder to reach investment grade status, and without investment grade status, borrowing stays expensive. It is a loop with no obvious exit.

The risk factors section of the filing reads like science fiction, citing possibilities including AI systems with blackmail-like capabilities and harms so severe the company may be uninsurable. Add in an ongoing legal fight with the Department of Defense over how the military can use Anthropic's products, and the picture becomes one of a company managing financial, political, and reputational fires simultaneously.

The Industry-Wide Question

Gartner recently estimated that the Neo Cloud industry needs $8.5 trillion in revenue by 2030, and that 80% of these companies are unlikely to survive past that point. Anthropic is widely regarded as the strongest of the frontier AI labs. If its economics look this fragile even in a best-case scenario, the implications for the rest of the sector are not encouraging.

Going forward, the question is not whether Anthropic can grow. It clearly can. The question is whether growth can outrun the compute bills, the customer churn, and the borrowing costs long enough to matter. This S1 does not answer that question so much as it puts the uncertainty on full display for anyone willing to read past the headline valuation.