Anthropic's revenue surges 14-fold...$807 billion AI bet

【Financial News, New York — Correspondent Lee Byung-chul】 Anthropic, a U.S. artificial intelligence (AI) company considered one of the candidates for the largest initial public offering (IPO) ever, has released its first registration statement. Its revenue in the first half of this year surged 14-fold in a year, but its net loss also ballooned to $80.7 billion (about 115 trillion won). Having placed a bet on the growth potential of the AI market and signed infrastructure contracts worth up to $807 billion (about 1,150 trillion won), the company now faces major risks to future growth, including the burden of recouping massive investments, instability in the semiconductor supply chain and challenges securing power.
On the 26th, local time, Anthropic filed a registration statement with the U.S. Securities and Exchange Commission (SEC) for a Nasdaq listing, disclosing detailed financial results, long-term investment plans and business risks. The company is pursuing a listing as early as next month, and its shares are expected to trade on Nasdaq under the ticker symbol “ANT.”

According to the registration statement, Anthropic's revenue in the first half of this year was $16.286 billion, about 14 times the $1.163 billion it reported in the same period last year. The increase came as businesses adopted AI more widely, driving a sharp rise in use of the AI chatbot Claude and the company's enterprise AI models. However, its net loss for the same period widened significantly to $80.748 billion, from $12.969 billion a year earlier.
Most of the net loss, however, was an accounting loss that did not involve an actual cash outflow. As Anthropic's valuation soared, a loss of about $72.4 billion was recorded in the process of revaluing the fair value of financial instruments, including convertible securities. The higher the company's valuation, the greater the recorded value of liabilities that may later convert into shares, resulting in a larger accounting loss.
Its actual operating performance, by contrast, improved. The operating loss in the first half of this year was $693 million, down sharply from about $2.14 billion in the same period last year. The company posted operating income of $974 million in the second quarter, turning a profit.
The challenge is the massive cost of AI infrastructure the company will have to bear going forward. The New York Times (NYT), after totaling the contracts listed in the registration statement, found that the value of Anthropic's contracts for computing and other infrastructure could reach as much as $807 billion. These contracts are for securing cloud computing resources and data center infrastructure needed to train and operate AI models. However, because the contracts vary in duration and termination terms, the total does not represent a fixed amount of investment spending.
For example, a $110 billion contract with Amazon.com, Inc. runs through 2036. By contrast, according to Yahoo Finance, a contract to lease computing resources from SpaceX is scheduled to run through 2029, but can be terminated with 90 days' notice starting in 2027. In particular, non-cancellable contracts require payment even if the computing resources are not actually used. This means that if AI demand grows more slowly than expected, the enormous contract costs could weigh on profitability.
As of the end of June this year, Anthropic held about $132 billion in cash and investments that could be converted into cash in the short term. Analysts say that its pursuit of an IPO to secure additional funding nonetheless highlights the capital-intensive nature of the AI industry's business model.

Anthropic also identified concentration in its customer base and semiconductor supply chain as major risk factors. In the nine months through the end of September this year, about 25% of its total revenue came from just five customers. Anthropic did not disclose their names. A key risk, it noted, is that many customers have not signed long-term contracts and can reduce their use of AI services or terminate their contracts at any time.
The semiconductor supply chain is also a vulnerability. In its registration statement, Anthropic said that most high-bandwidth memory (HBM)—a key component of high-performance AI accelerators—and other high-performance memory products are produced by just three companies: SK hynix, Samsung Electronics Co., Ltd. and U.S.-based Micron Technology, Inc. It noted that although these companies are expanding production capacity, it is uncertain whether they can meet surging AI demand. The company expressed particular concern that it might be unable to secure the semiconductors it needs if major customers or competitors book production capacity first.
Anthropic also cited the concentration of advanced semiconductor production at Taiwan Semiconductor Manufacturing Company Limited (TSMC) as a risk. It warned that disruptions to Taiwan's advanced semiconductor production caused by a military conflict between China and Taiwan, a blockade or natural disasters could affect global supply chains for years. This is because building alternative production facilities and establishing mass-production systems would take considerable time. The company added that tariffs and import and export restrictions on semiconductors produced outside the United States could also drive up costs and delay supply.
Meanwhile, Anthropic devoted 94 of the 335 pages in its registration statement to explaining business risks. These included risks that AI technology itself could pose. Anthropic warned that advanced AI models could pose catastrophic or existential risks to humanity. It said that if AI models behave in unexpected ways or are used maliciously, this could lead to legal liability and reputational damage.

[email protected] Correspondent Lee Byung-chul Reporter