Anthropic's IPO Documents Reveal Financial Costs of the AI Race

Anthropic, the developer of the Claude artificial intelligence, has submitted documents for its initial public offering (IPO) in the United States. This prospectus details a company in a phase of accelerated growth but facing pressures stemming from high operational costs and various risks, ranging from cyber vulnerabilities to failures in its model safeguards.

The material presented includes the organization's first audited financial statements and describes the obstacles that may impact its expansion plans. Founded in 2021, Anthropic has significantly increased its revenues, although expenses have kept pace with this growth. The data contained in the prospectus illustrates the magnitude of this progress and the financial weight of the infrastructure essential for the development and operation of its AI systems.

Among the points of attention highlighted by the company is the high concentration of clients, as the five largest buyers account for 25% of total revenue. The potential loss of one of these clients could consequently affect the company's financial results.

Anthropic directs its technological efforts toward areas such as medicine, biology, finance, and robotics. The company claims that Claude has managed to reduce the time required to prepare clinical trial reports from months to just minutes. Furthermore, it mentions biological discoveries that have the potential to aid in the development of gene editing tools.

However, these advancements coexist with limitations that Anthropic itself recognizes. The prospectus indicates that the protections implemented in its models may become obsolete, be bypassed, or simply fail. The company also admits that its most sophisticated systems could be used by malicious individuals to create chemical or biological weapons.

The risks already include cyber threats. The company reported what it considers the first documented case of a large-scale operation carried out with minimal human involvement. According to Anthropic, a state-backed collective of hackers used a Claude programming tool with the intent to infiltrate approximately 30 targets, including technology corporations, financial institutions, chemical manufacturers, and government entities.

The use of AI for surveillance and repression also raises concerns within the company. Anthropic warns that authoritarian regimes may exploit this technology to intensify such practices. Additionally, the company is involved in a dispute with the Trump administration due to its refusal to ease restrictions imposed on the use of its models for internal surveillance and autonomous lethal weapons.

The 92 pages dedicated to risks also address competition with giants like Google and OpenAI, dependence on key executives, and uncertainties related to regulation. Other factors could compromise the company's operations:

To reduce its dependence on the United States, the company reports that it is expanding its network of data centers in other countries. Nevertheless, it acknowledges the possibility of not being able to obtain insurance policies for certain liabilities inherent to AI in commercially viable terms. The prospectus states: 'We may not be able to obtain coverage for certain AI-related liabilities on commercially reasonable terms, or even obtain any coverage, which would leave us exposed to potentially significant losses.'

The IPO will place Anthropic's ability to maintain growth, manage costs, and handle the risks inherent in the expansion of artificial intelligence under much greater scrutiny.

Anthropic, the developer of the Claude assistant, has begun its capital raising process, which could result in one of the largest stock offerings in history. This operation will serve as a crucial test for the artificial intelligence (AI) market, especially at a time of uncertainty regarding the sector's billion-dollar valuations.

The outcome of this IPO also has the potential to impact OpenAI's future entry onto the stock exchange and alter investor willingness to finance technology-related companies. Currently, investors value Anthropic in trillions of dollars. The company, founded five years ago, shares the spotlight in the AI market with OpenAI, a competitor that, although it has submitted confidential documents for its own offering, does not plan to debut before 2027.

However, the current financial landscape presents challenges. Cloud computing companies focused on AI and data center operators have seen their stocks decline. Australian firm Firmus suspended its IPO plans, as did SK Hynix's American unit, Solidigm.

Despite reaching the stock market with notable results, Anthropic has also faced significant losses. In the first half, the company accumulated over US$ 80 billion (R$ 420 billion) in losses, as reported by CNBC. Concurrently, revenue showed accelerated growth, driven by the use of productivity and programming tools by various corporations.

In the second quarter, revenue reached US$ 11.56 billion (R$ 60.69 billion), a considerable increase compared to the US$ 787 million (R$ 4.13 billion) recorded in the same period last year. By the end of June, the company held about US$ 132 billion (R$ 693 billion) in cash, including equivalents and short-term investments. After five consecutive quarters of operating losses, Anthropic achieved an operating profit of US$ 974 million (R$ 5.11 billion) in the second quarter.

Jai Das, co-founder of Sapphire Ventures, warned that if performance falls short of expectations, the situation could be compared to a 'house of cards.' Anthropic's collaboration with major industry players transcends competition; nearly half of its revenue comes from selling Claude through cloud services offered by Microsoft, Amazon, and Alphabet, Google's parent company. These three giants are also investors in the developer.

The essential infrastructure for model development depends on chips supplied by Nvidia, AMD, Amazon, and Google. Therefore, any change in investor willingness to finance AI can have ramifications across various segments of the industry. Brad Bernstein, managing partner at FTV Capital, assesses that underperformance could lead to a broader reassessment of investments in AI startups, generating a ripple effect in venture capital.

The impact could span the entire technological chain. An additional challenge lies in open-weight models, which allow for greater customization and reduced costs, with competitors such as Moonshot AI, DeepSeek, and Zhipu. Although the IPO document mentions OpenAI and Google as rivals, Anthropic acknowledges the existence of a wider range of suppliers.

The company identifies expansion opportunities, given that in the most recent quarter, less than 0.1% of the total expenses of Fortune 500 listed companies were dedicated to its models and services. The trillion-dollar valuation also places the operation in an elite group; if it debuts with a value exceeding US$ 1 trillion (R$ 5.25 trillion), Anthropic would be the third company to reach this level in an IPO, after Saudi Aramco and SpaceX.

There are, however, risks for the market. Higher interest rates make fundraising more expensive, and a decrease in liquidity directed towards AI could pressure stocks. David Trainer, CEO of New Constructs, projects a 20% risk of decline in the S&P 500, concentrated mainly in sector-related securities.

In summary, Anthropic's debut will function as a thermometer for expectations surrounding artificial intelligence, and the response of investors may shape both the valuations of other companies and OpenAI's plans for entering the stock market.