Anthropic projects up to $750 billion in compute ahead of its IPO: frontier AI is now an infrastructure business

By Anadolu Ajansı (based on a Wall Street Journal report) · October 20, 2026.

Anthropic could spend up to $750 billion on additional computing capacity over the coming years, according to the Wall Street Journal, which cites a draft prospectus for its initial public offering (IPO). The figure includes deals still under negotiation. According to the newspaper, the company will file the documents publicly in the coming weeks.

The prospectus, again according to the WSJ, shows a net loss of $80.7 billion in the first half of 2026, compared with $13 billion in the same period of 2025. About $72 billion of that loss comes from accounting charges tied to changes in the value of convertible bonds and other financial instruments issued by the company. In other words, most of it is not cash burned but an accounting revaluation of those instruments.

Operating spending on compute and infrastructure, however, is real and growing fast: nearly $9 billion in the six months to June, compared with $2.3 billion a year earlier. That half-year alone exceeds the $7.3 billion spent in all of 2025.

Firm commitments total around $450 billion over 7 to 10 years, and most of them cannot be cancelled. They include about $111 billion with Google Cloud, $110 billion with Amazon Web Services and $31 billion with Microsoft. The remainder up to $750 billion would be capacity not yet contracted.

The prospectus also lists risks, according to the report: model behaviors aimed at avoiding shutdown or responses resembling blackmail, and political opposition to building data centers in the U.S., where most of its capacity is located.

**Our take.** The headline loss is a distraction. What matters is the structure: Anthropic is locking itself into non-cancellable obligations for a decade, betting that demand for its models will grow enough to absorb them. It is an infrastructure bet closer to a utility or a telecom than to a software startup. If it pays off, the secured compute will be an enormous competitive advantage. If demand slows or prices fall, those clauses go from leverage to dead weight.

There is another consequence: how the money is distributed. Google, Amazon and Microsoft are at once suppliers, investors or rivals of Anthropic. Overall, frontier AI consolidates power in whoever controls chips, energy and data centers, concentrating risk and power in a few hands. It is a short-term problem worth watching, along with local resistance to data centers, which could turn energy and permits into the real bottleneck.

That is why the IPO will be an opportunity: for the first time, the market will see audited, consistent figures, something currently missing for comparisons with OpenAI. It is worth waiting for the final prospectus before drawing conclusions, since everything above comes from a leaked draft.

In the long run, this outsized investment is only justified if compute translates into real advances, such as drug discovery, accelerated science and cheap services for everyone. It is the road to abundance, but with a huge and uneven transition bill paid by investors, power grids and communities.