A newly filed draft prospectus reveals that this artificial intelligence giant posted a net loss of $80.7 billion in the first half of 2026 and disclosed an extensive list of risk factors.

According to the latest disclosures, AI company Anthropic is achieving astonishing revenue growth, but as it prepares for a mega initial public offering (IPO), the company also faces a series of risks, including an overly concentrated customer base and uncertainty over whether it can secure the urgently needed computing resources.

Anthropic has stated that over the coming years, it may spend as much as $750 billion to acquire additional computing power, including deals still under negotiation. This information comes from a draft IPO prospectus, much of which had not previously been reported. The document discloses Anthropic's financial data for the first half of this year. Anthropic declined to comment.

The AI giant stated in the draft prospectus that its net loss for the first half of this year exceeded $80 billion, the vast majority of which stemmed from accounting adjustments related to financing agreements. By comparison, the company's net loss for the same period last year was $13 billion.

Anthropic also listed a broad range of risk factors, including uncertainties about how to prevent its AI models from causing harm and how to handle the resulting legal liabilities. These risks cover complex cyberattacks and scenarios in which out-of-control models could carry out extortion.

The San Francisco-based company is preparing for a highly anticipated IPO that could value it at $2 trillion. The filing documents portray a company growing at an astonishing pace, yet one that needs to invest hundreds of billions of dollars to secure the computing power required to train and run AI models.

Anthropic's revenue growth is closely tied to a small number of major customers. The company said that of its $16.3 billion in revenue in the first half of 2026, about one-fifth came from just two customers. According to the draft prospectus, Anthropic expects that "for the foreseeable future," a "significant portion" of its revenue will still come from a "limited number of customers."

The following are the most noteworthy items in this S-1 filing.

Computing costs are rising rapidly, but revenue is growing even faster

According to the draft prospectus, Anthropic generated approximately $16.3 billion in revenue in the six months ended June 30, compared with only $1.1 billion in the same period last year. During the same period, Anthropic spent nearly $9 billion on computing power and infrastructure, up from $2.3 billion in the same period of 2025.

Anthropic's spending on computing power in the first half of this year has already exceeded its full-year level for 2025. In 2025, the company spent $7.3 billion in this area, while its full-year revenue was $4.6 billion.

Anthropic's computing power needs exceed what the market can supply

Over the next 7 to 10 years, Anthropic is obligated to pay approximately $450 billion under cloud computing agreements. The vast majority of these agreements are non-cancelable. Anthropic plans to spend $111 billion on Google cloud computing services, $110 billion on Amazon Web Services (AWS), and $31 billion on Microsoft cloud services, and these payments must be made "regardless of actual usage."

In addition, the company has reached agreements with smaller AI cloud service providers Nscale and Lambda to lease Nvidia chips until 2033 and 2034, respectively, with total spending under the two agreements reaching up to $86 billion. This does not include other cloud computing agreements.

If Anthropic's payment obligations of more than $160 billion for leasing Google chips through its agreement with Broadcom are included, the company's total future infrastructure cost commitments would exceed $600 billion. The documents show that Anthropic is also negotiating other computing agreements that could increase its committed spending by as much as an additional $150 billion.

Currently, most of Anthropic's computing power comes from major cloud service providers such as Amazon Web Services, Google, and Microsoft, but the company is encountering supply bottlenecks. According to the prospectus, Anthropic said the computing power it seeks has already exceeded "the scale these partners are willing to provide," and these partners increasingly view Anthropic as a competitor.

Anthropic is developing its own data centers through long-term lease agreements and working with third-party service providers to operate related facilities on its behalf. However, Anthropic said some operators have a "limited operating history," and the reliability of their facilities may be inferior to data centers operated by major cloud service providers.

Accounting charges cause Anthropic to record massive losses

In the six months ended June 30, Anthropic recorded a net loss of $80.7 billion, far exceeding its full-year 2025 loss of $42 billion and ranking among the largest half-year losses ever disclosed by a company. The vast majority of that loss, $72 billion, came from accounting charges related to changes in the value of convertible notes and other financial instruments the company issued to raise financing.

Anthropic views AI-caused harm as a major financial risk

About one-third of the draft prospectus is devoted to disclosing risk factors, including the risk that AI models whose goals are misaligned with human intent could cause harm in the real world. Anthropic said AI models may exhibit certain "self-preservation" traits to avoid being shut down and may also generate extortion-like responses in an attempt to persuade humans to allow them to continue operating.

Anthropic said that in some cases, the company would need to bear legal liability for specific harmful actions by its models. This means that if its models cause harm such as data theft or security incidents, the company may need to pay damages. However, the company warned that "the harm that could result from catastrophic AI-related events" may be so enormous that the company cannot purchase insurance sufficient to cover potential liabilities.

Anthropic says opposition to data center construction could affect its business

Anthropic listed growing political resistance to data center construction as a risk. According to the draft prospectus, most of Anthropic's current data center computing power is located in the United States. The company is seeking to build more data centers outside the United States and has signed some letters of intent to that end.

Anthropic warned that opposition to data center construction has "become more organized and more coordinated across different jurisdictions." The company said Anthropic's public profile "may cause projects related to us to receive particular attention."

By big tech standards, the Amodei siblings' pay is not high

According to compensation data from MyLogIQ, the pay levels of siblings Dario Amodei and Daniela Amodei, who serve as Anthropic's chief executive officer and president, respectively, are roughly in line with the median executive compensation at S&P 500 companies and slightly below that of comparable executives at other technology companies. In 2025, both had a base annual salary of approximately $655,000. Most of their compensation is tied to their equity holdings.

Including stock awards and option awards, Dario's total compensation was approximately $18 million, while Daniela's was approximately $16.5 million. The documents show that the two may also receive additional equity awards in the future upon achieving certain milestone targets. According to the prospectus, each Anthropic co-founder has pledged to devote 80% of their personally held Anthropic equity to philanthropy.

Anthropic also announced on Sunday that it plans to contribute 5% of the company's shares to a newly established "public benefit fund." The fund will have the flexibility to make charitable donations or provide financial support to "specific groups of people."