The ~$900 billion in commitments in Anthropic's IPO: purchase obligations, not hidden debt, but the sector's biggest capital bet

By Bloomberg Tax · October 30, 2026.

In its IPO filing, Anthropic has committed to spending nearly $900 billion on future obligations for servers, buildings and energy. That is according to Bloomberg Tax, which adds that the figure far exceeds the liabilities on the company's balance sheet. The company filed its registration statement on October 26, ahead of its expected stock market listing.

A note on method: the original article is behind a paywall and we were only able to read the opening. The details (timelines, counterparties, breakdown) fall outside this analysis, and the "hidden debt" label is Bloomberg's framing, not our conclusion.

Other coverage of the same event adds two pieces. According to a Financial Times report cited by TechCrunch, Anthropic would receive a high "junk" rating from credit agencies after the IPO. TechTarget, for its part, argues that the IPO puts the spotlight on the real return of enterprise AI. Both are attributions to third parties; we have not verified the FT report.

**Our take.** Calling this debt "hidden" is imprecise. Capacity and energy purchase commitments are common practice and appear in the prospectus; in other words, they are disclosed. What is true is that they do not show up on the balance sheet, so an investor looking only at accounting liabilities will underestimate the company's operating leverage. The useful conversation is not the scandal, but how much of that spending is firmly committed and how much is flexible if revenue falls short. That answer is to be found in the prospectus, not in the headline.

In general, frontier labs have turned compute into their main cost item, and those who cannot finance that race are left behind. A commitment of this scale turns Claude's revenue into a condition for survival: if enterprise demand grows at the expected pace, the infrastructure pays for itself; if it slows, those contracts become a burden. That is why TechTarget's question about the return on enterprise AI is the underlying question.

For our thesis, this is the uncomfortable side of long-term optimism. The abundance AI promises requires massive investment in energy and data centers, and in the short term that investment creates financial fragility, concentration of power in those who can pay for it, and pressure on power grids. Anthropic going to the public market has a positive reading: more transparency, more scrutiny and more capital to build the capacity the sector needs. But the test will not be the size of the commitment, but whether revenue covers it. Until the full prospectus is seen, it is wise to be wary of both the alarmist headline and the IPO euphoria.