Abstract:
On October 21, the draft IPO prospectus seen by The Wall Street Journal showed that Anthropic’s revenue is growing at an alarming rate, but it has also suffered huge losses, and there is uncertainty about whether it can continue to obtain much-needed computing power. The following are the key points of the Anthropic S1 prospectus:
Anthropic
1. Revenue soared by 1382%, and computing power expenditure exceeded last year
The draft prospectus shows that in the six months ended June 30 this year, Anthropic achieved revenue of approximately US$16.3 billion, an increase of 1382% from US$1.1 billion in the same period last year. During the same period, the company spent nearly $9 billion on computing power and infrastructure, compared with $2.3 billion in infrastructure spending during the same period in 2025.
Revenue and computing power expenditure are growing rapidly
Anthropic’s computing power expenditure in the first half of this year has exceeded the level for the whole of 2025. Last year, Anthropic spent $7.3 billion on computing power and had revenue of $4.6 billion.
2. The total expenditure on computing power in the future will reach US$750 billion, exceeding market supply
The company will have to pay approximately $450 billion over the next seven to 10 years under cloud computing agreements, most of which are non-cancelable.
Anthropic plans to spend $111 billion on Google cloud computing contracts, $110 billion on Amazon cloud services contracts, and $31 billion on Microsoft contracts, which must be paid regardless of actual usage. In addition, the company also agreed to lease Nvidia chips from smaller AI cloud service providers Nscale and Lambda. The relevant contract periods are until 2033 and 2034 respectively, with a maximum expenditure of up to US$86 billion. These expenditures do not include other cloud computing contracts.
Coupled with Anthropic's more than $160 billion in Google chip leasing obligations through its agreement with Broadcom, the company's future infrastructure costs total more than $600 billion. Documents show Anthropic is still negotiating other computing power purchase agreements that could increase its committed spending by up to $150 billion.
Although Anthropic gets most of its computing power from large cloud service providers such as Amazon Cloud Services, Google and Microsoft, the company is facing resource constraints. Anthropic stated in its prospectus that the company seeks to obtain more computing power than "these partners are willing to provide," and that these partners increasingly view Anthropic as a competitor.
Anthropic is building its own data centers through long-term leases and working with third-party service providers to operate the facilities on its behalf. However, Anthropic said some of these operators have "limited operating history" and their facilities may not be as reliable as data centers operated by larger cloud service providers.
3. Huge losses
In the six months ended June 30, Anthropic's net loss reached US$80.7 billion, far exceeding the US$42 billion loss for the full year of 2025, and one of the largest half-year losses ever disclosed.
Net loss exceeds US$80 billion
The vast majority of this loss, $72 billion, was due to accounting adjustment charges. The charges relate to changes in the value of convertible notes and other financial instruments the company issues to raise capital.
4.AI damage is a significant financial risk
About a third of the draft prospectus is devoted to risk factors, including the real-world harm that models that are inconsistent with human goals could cause. Anthropic said that AI models may show "self-protection" tendencies to avoid being shut down; they may also generate "blackmail"-like responses in an attempt to convince humans to allow them to continue running.
Anthropic said that in some cases, companies need to be held legally responsible for certain harmful behaviors of models. This means that if its models cause damage such as data theft or security incidents, companies must bear the corresponding compensation costs.
But the company warned that the scale of "potential damage caused by AI-related catastrophic events" may prevent the company from obtaining insurance sufficient to cover potential losses.
5. Boycott of data centers may become a business problem
Anthropic cited growing political resistance to data center construction as a risk. According to the draft prospectus, most of Anthropic's data center capacity is currently located in the United States, and it is seeking to build more data centers outside the United States and has signed some letters of intent.
The company warned that resistance to data center construction "has become more organized and begun to be coordinated across different jurisdictions." The company said its higher public profile "may bring projects involving us to receive special attention".
6. CEO salary is not high
According to salary data from MyLogIQ, the compensation of Anthropic CEO Dario Amodei and his sister and president Daniela Amodei is roughly in line with the median salary of S&P 500 company executives and slightly lower than that of their technology company peers.
Figure 4: Salaries of the Amodei brothers and sisters
In 2025, the basic annual salary of both men will be approximately US$655,000. Much of their compensation is tied to equity. Dario’s stock and option awards bring his total compensation to approximately $18 million, while Daniela’s total compensation is approximately $16.5 million.
The documents show that the two are also expected to receive additional equity awards in the future after reaching certain milestones.
7. The founder donated 80% of his equity to charity
According to the prospectus, each Anthropic co-founder has pledged to use 80% of his personal Anthropic shares for charity.
Anthropic announced on Sunday that it planned to invest 5% of the company's shares in a newly established "public interest fund." The fund has the flexibility to make charitable donations and can provide funds to "specific groups of people."
Comments