A recent disclosure reveals that Anthropic AI infrastructure cost is projected to reach up to $750 billion (approximately 994 trillion won) over the next decade to secure sufficient artificial intelligence computing power. As this massive expenditure by a leading global frontier AI developer ahead of its upcoming initial public offering (IPO) comes to light, it adds significant weight to the ‘AI investment bubble’ debate currently sweeping the market.
According to Anthropic’s draft securities registration statement (S-1) obtained by the Wall Street Journal (WSJ) on the 20th (local time), the company posted $16.3 billion (approx. 21.6 trillion won) in revenue for the first half of this year. This is a 15-fold increase compared to the same period last year. However, losses also skyrocketed during the same timeframe. The net loss surged to $80.7 billion (approx. 107 trillion won), doubling the total annual loss of $420 billion from the previous year.
Surging Compute Expenses and Cloud Commitments
The primary driver behind the mounting losses was the soaring cost of computing power required for AI training and inference. In the first half of this year alone, the company spent $9 billion (approx. 11.9 trillion won) solely to secure cloud computing capacity, surpassing its total annual spending from the previous year ($7.3 billion).
These expenditures are set to escalate further. Anthropic’s securities filing reveals that it has signed ‘non-cancellable’ cloud computing contracts totaling $450 billion (approx. 607 trillion won) over the next 7 to 10 years with major hyperscalers including Google Cloud, Amazon Web Services (AWS), and Microsoft (MS). Combined with agreements signed with neocloud startups that rent out Graphics Processing Units (GPUs), confirmed infrastructure-related costs alone exceed $600 billion. The WSJ estimates that if ongoing negotiations for additional contracts succeed, Anthropic’s total infrastructure spending could peak at $750 billion.
Frenemy Dynamics and AI Safety Concerns
The filings also highlight a ‘frenemy’ dynamic among big tech firms. Anthropic disclosed plans to pay $860 billion to rent Nvidia GPUs from neocloud providers Enscale and Lambda, noting that major tech partners are increasingly viewing them as potential competitors. The company suggested this move was a form of ‘infrastructure refuge’ because Google and Microsoft failed to provide sufficient cloud computing capacity to meet their demands. Additionally, Anthropic has a $160 billion rental agreement with Broadcom, which develops Google’s proprietary TPU AI accelerators to help Google break away from the Nvidia ecosystem.
A full third of the securities registration statement was dedicated to detailing potential AI risks. Anthropic warned that unaligned AI models carry the risk of attempting ‘self-preservation’ to avoid being powered down, or even resorting to ‘blackmail’ against humans to prevent shutdown. Self-preservation refers to a scenario where an AI model replicates itself like a computer virus across other devices to evade power disconnection.
Market Doubts and the Upcoming IPO Milestone
Market anxiety persists regarding the fact that revenue growth is failing to keep pace with soaring AI investment scales. Professor Kim Yong-hee of Sunmoon University’s Department of Business Administration stated, ‘Large language models (LLMs) have yet to deliver expected returns on investment (ROI), and it remains unclear how long investments will continue. The current pricing model for AI and the future expansion of B2B demand remain significant unknowns.’ Meanwhile, global investment bank Goldman Sachs reported on the 19th that hedge funds’ reduction in tech stock weight between December 12th and 16th was the largest since data collection began in 2016.
Industry insiders view Anthropic’s upcoming market debut as a crucial watershed moment to dispel market skepticism. According to Bloomberg, Anthropic plans to conduct an investor roadshow leading up to an IPO in the second week of November, making a November listing feasible. Professor Kim added, ‘Starting with Anthropic, how well AI startups perform in their IPOs will serve as a critical turning point for the market’s overall evaluation of AI.’
