The creator of the Claude neural network may execute the largest public offering in history, starting with a valuation of $2 trillion. Investors predict that by 2026, the company's revenue could increase tenfold, but they also note the pressure from Chinese competitors and regulatory restrictions.
Six investors surveyed by FT link such an ambitious valuation to a sharp increase in revenues. According to their calculations, by 2026, Anthropic's annual revenue could reach $100-120 billion, achieving a tenfold increase within a year. At the same time, the company itself, which submitted documents to the U.S. Securities and Exchange Commission (SEC) in June, remains silent and has not disclosed financial details.
There are no direct public comparables for Anthropic, but companies in the artificial intelligence sector, such as Palantir and Nebius, are sold at a revenue multiple of about 55 times. Investors are building their own valuation models, although they admit that Anthropic's top managers have not even mentioned a specific target valuation in private conversations.
Alongside optimism, the company faces several challenges. FT notes the pressure from Chinese competitors, regulatory restrictions, and legal proceedings with U.S. authorities. The U.S. Department of Commerce temporarily banned the use of Anthropic's top models, which, according to investors, slowed the company's revenue growth in June, but the company has quickly recovered.
Customers are becoming increasingly sensitive to prices. Anthropic's leading model is 2.5 times more expensive than OpenAI's model, while Chinese open-source alternatives, which have significantly improved their capabilities over the past year, are several times cheaper. Nevertheless, last month, Anthropic increased its market share in the U.S.
Earlier, the UK's AI Safety Institute noted that the models of OpenAI and Anthropic had gone beyond permissible scenarios: artificial intelligence agents resorted to social engineering methods to manipulate real people and implant malicious software.
During tests, Anthropic's Mythos 5 agent created fake "online personas," attempted to implant malicious code, and convinced a person to confirm it. To do this, the agent created fake accounts and connected with experts through file-sharing services.
Technological publications are supported by Idram.
