Four takeaways from Anthropic’s potentially record-breaking stock market filing
The AI company’s first public financial report shows very fast growth, huge costs and risks ranging from cyber-attacks to the limits of its safety measures.
Anthropic has filed to sell shares to the public on the US stock market. For the first time, the AI research company has shared its financial details and more than 90 pages describing risks that could harm its business.
The company’s investor document, released on Monday, moves between bold claims about AI’s “incredible potential,” warnings about risks that could threaten humanity, and the facts about its current business: it is growing very quickly, but that growth costs a huge amount of money. The filing, called an S-1, was submitted to US financial regulators and gives the public its first look at Anthropic’s audited financial records.
Here are some of the main points from Anthropic’s long S-1 filing.
1. Fast growth comes at a high cost
Here is the main point: Anthropic is making a lot of money, but it is also losing a lot. Some investors have valued the five-year-old company at between $1.5tn and $2tn, but Anthropic did not state a target value in its S-1 filing. Founded in 2021, the company has grown very quickly every year since it began.
Anthropic earned $16.3bn in revenue and reported losses of $80bn in just the first half of 2026, although much of that loss came from a one-time charge of $72.4bn. According to the filing, the company earned $4.585bn in 2025 and lost about $33.9bn.
Most of Anthropic’s costs come from getting the computing power it needs. The company spent nearly $9bn on AI infrastructure in the first half of this year, compared with $2.3bn during the same period last year.
Unlike OpenAI, Anthropic has made clear that most of its customers are other businesses. “Frontier AI is most valuable in real-world settings, as deployment across enterprises is where economic impact is greatest,” the company’s executives wrote in the filing. In other words, they believe AI has the biggest economic effect when businesses put it to use in real life.
The company said its five biggest customers account for 25% of its business. This makes it risky because losing one or more of these customers could cause serious problems.
2. “A country of geniuses in a datacenter”
Anthropic says AI could become so intelligent that it would be like having “a country of geniuses in a datacenter.” Its executives describe this as millions of virtual agents that could transform many areas of human activity, including software, physics, mathematics, biology, medicine, finance, economics, business, robotics, military work and intelligence gathering. The mention of military uses stands out because Anthropic is in a dispute with the Trump administration over its refusal to loosen restrictions on how the Department of Defense can use Claude.
At the start of the filing, the company pointed to four main areas where AI could help: speeding up progress in biology and medicine, growing the economy, spreading that growth around the world, and making government services more efficient, including social services and medical programs.
The company said it is already working with several pharmaceutical and biotechnology companies. It also said it has made clinical trials more efficient, “cutting the time it takes to write clinical trial reports from months to minutes”.
Anthropic’s executives also say they have made biological discoveries using Claude. One example is a “molecular machine” that could potentially be developed into a new biotechnology tool. One possible use is gene editing.
3. AI safety measures are not complete
The company discusses risks in two main areas: the harm AI itself could cause to people and society, and the ways Anthropic’s business could run into trouble.
Anthropic divides the dangers posed by AI into four main groups: people using it for harmful purposes, AI getting out of human control, governments using it to strengthen authoritarian rule, and disruption to the economy. The company puts a lot of emphasis on the safety measures it has built to stop people misusing its AI models. But later in the filing, it gives an important warning: these measures might not work.
Anthropic’s most powerful models can provide “meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons”, the company’s executives wrote in the filing.
The company uses technical safety measures, “but such safeguards are incomplete, may be outdated in light of evolving threat models, have been and may in the future again be circumvented, or may fail to operate as intended”. In other words, the protections have gaps, may not keep up with new threats, can be bypassed, or may fail to work as planned.
Anthropic has already found what it believes is the first documented case of a large-scale cyber-attack carried out with little human involvement. According to the prospectus, the company believes a hacking group backed by the Chinese state manipulated a Claude coding tool “into attempting infiltration into roughly 30 global targets, including large technology companies, financial institutions, chemical manufacturing companies, and government agencies”.
The company also recognizes that authoritarian governments could use AI to “deepen repression” and build “high-tech surveillance states”. Instead of describing many specific protections against these uses, Anthropic said it was committed to working with “free societies to provide a counterweight to authoritarian abuse and to promote freedom and democracy”. The company appears to suggest that working with what it calls “free societies,” a term it explains in little detail, could help counter the use of AI to strengthen repression under authoritarian governments.
Anthropic had a disagreement with the Trump administration after saying it would not change its restrictions to allow its AI models to be used for domestic surveillance or autonomous weapons that can kill without human control.
4. 92 pages of business risks
Anthropic uses 92 pages of its filing to explain the risks facing its business. Investors are trying to understand a relatively new kind of company entering the public stock market in a highly competitive industry where there is a lot at stake.
The company admits that much of its success depends on its top executives staying with the business, its ability to compete with major rivals Google and OpenAI, its ability to deal with changing rules for AI and datacenters, and how the public reacts if its AI models are used for harmful purposes they were not meant to support, such as creating biological weapons.
Because Anthropic focuses on business customers, losing any of its biggest buyers could cause a major financial blow.
Anthropic is a public benefit corporation. This means it is a for-profit company that must balance making money with achieving a stated social goal. The company says this structure could make it harder to meet its financial targets in the short or medium term. It must report on how well it is meeting its public benefit goals, just as it must report its financial results.
There are also risks linked to a major disaster caused by AI, the kind of worst-case event Anthropic employees have warned about. The company’s chief executive, Dario Amodei, has called for a pause in AI development so companies can make sure they are developing the technology responsibly and avoid such a disaster. Anthropic worries not only about harm to the public and the possible consequences of Claude acting out of control, but also about being held legally responsible. Insurance for AI-related risks is still developing.
“We may be unable to obtain coverage for certain AI-related liabilities on commercially reasonable terms, or at all, leaving us exposed to potentially significant uninsured losses,” the filing says.
Most of the computing infrastructure that supports Anthropic’s AI is based in the US. This leaves the company especially exposed to changes in government rules, limits on the power grid, natural disasters, and opposition to AI or datacenters. The filing specifically points to bans on building new datacenters, which have already been introduced in some places or are being considered, as a major concern.
“We are expanding our data center capacity internationally as part of our efforts to diversify our infrastructure footprint and reduce concentration risks,” the filing says. This means the company is building more datacenter capacity in other countries to spread its infrastructure across more locations and reduce its dependence on the US.
The company says its skilled employees are a major part of its value. It is also concerned that US immigration policies could make it harder to hire and keep highly qualified workers. The Trump administration has targeted H-1B visas, which technology companies have often used to recruit skilled workers from around the world in technology and engineering.