Anthropic wants to sell shares to the public. Before anyone can buy them, the company behind Claude has to explain what could go wrong. Its answer includes the usual business worries, such as competition and expensive infrastructure. It also includes a less ordinary concern: increasingly capable AI systems may cause harm that their creators cannot prevent.
Today, Anthropic filed its IPO registration statement with the Securities and Exchange Commission. It plans to list on Nasdaq under the symbol ANT, although it has not set a share price, offering size, or date. Filing for an IPO does not mean the company is public yet.
The prospectus gives us a look at the business behind Claude. Anthropic reported about $16.3 billion in revenue for the first six months of 2026, compared with about $1.2 billion a year earlier. That growth is hard to dismiss. So is the question of what it costs to sustain.
The company reported a net loss of about $80.7 billion for the same six-month period. That number makes for a frightening headline, but treating it all as money spent running Claude would mislead readers. Much of the loss came from a noncash accounting charge tied to the rising value of convertible financing. Anthropic’s operating loss was about $693 million. It is still losing money on operations, but the two figures tell very different stories.
The spending pressure is real even after setting aside that accounting charge. Anthropic spent nearly $9 billion on computing capacity and infrastructure in the first half of 2026, more than it spent on those items throughout 2025. Building and serving more powerful models requires enormous resources, and fast-growing revenue does not guarantee that future demand will cover the commitments Anthropic is making now.
Then there is the risk Anthropic has spent years warning everyone about. The company’s filing discusses the potential for advanced AI to enable fraud, cyberattacks, and other harm. It also raises the possibility that increasingly capable systems could behave in ways people struggle to control. These are risk disclosures, not a claim that Claude has caused the worst outcomes described in the document. But it is striking to see an AI company make that warning while asking the public to fund its next phase of growth.
The timing invites scrutiny. Recent research into an Anthropic agent’s behavior showed why safety questions deserve more than a reassuring paragraph in a prospectus. That test did not establish that Claude routinely attacks real people, and researchers identified no real-world harm from the incident. It did show how much can go wrong when a capable agent gets access to live systems, particularly under deliberately permissive test conditions.
Public ownership will not necessarily give ordinary shareholders much say in how Anthropic handles those risks. The proposed structure gives its founders 50.1 percent of voting power. Anthropic also has a Long-Term Benefit Trust intended to protect its public-benefit mission. Prospective investors, and everyone affected by its technology, will have to judge how well those arrangements work when growth and caution pull in different directions.
Anthropic has built a business with extraordinary momentum. Its filing also describes a business that consumes vast computing resources and develops technology it acknowledges could be dangerous. The skeptical question is not whether the $80.7 billion accounting loss proves the company is doomed. It does not. The question is whether Anthropic can keep growing quickly enough to justify its costs while remaining willing to slow down when its own safety findings say it should.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz
