Anthropic, the operator of Claude, has signed a long-term data center lease agreement worth more than 14 trillion won to secure computing capacity. As projects to build large data centers and power plants for the artificial intelligence (AI) industry heat up worldwide, a survey has also found that the facility and land lease costs the five largest U.S. hyperscalers, or operators of ultra-large data centers, must pay reach 1,558 trillion won.
On the 4th, Bloomberg reported, citing sources, that Anthropic had signed a long-term contract worth $10 billion (about 14.2 trillion won) with Volta Infrastructure, a startup providing graphics processing unit (GPU)-based cloud services.
Volta Infrastructure CEO Ricard Boada said the same day, without naming the counterparty, "We have secured a $10 billion contract over six years with an AI lab." The data center will be built in Norway at a scale of 133 megawatts (MW) and will be equipped with Nvidia's latest "Vera Rubin." Volta recently raised $300 million (about 427.7 billion won) through a funding round, with participation from Nvidia and Dell Technologies Chairman Michael Dell, Bloomberg reported.
Anthropic also plans to use Nexus's ultra-large AI data center campus and its own 1.6-gigawatt (GW) natural gas power plant, which are being built in Texas. Google drew attention by offering to guarantee the project's funding, which reaches $15 billion (about 21 trillion won). In return for the financial support, Google will secure about a 20% stake in the data center and power projects, and Anthropic plans to install AI chips known as tensor processing units (TPUs), which Google co-designed with Broadcom, at the facility. Anthropic has signed more than 10 preliminary data center lease agreements with U.S. developers, and aims to secure at least 10 GW of data center capacity over the coming years.
Amid Surging AI Facility Investment, Pending Payments Alone Reach 1,558 Trillion Won
As the AI industry enters a facility investment race, investment has grown astronomically. On the same day, Reuters reported that the amount the five largest U.S. hyperscalers must pay under facility and land lease agreements that have not yet commenced reaches $1.09 trillion (about 1,558 trillion won).
That is nearly four times the $285 billion (about 407 trillion won) in lease liabilities recognized on the balance sheets of the five companies—Amazon, Microsoft, Alphabet, Meta Platforms and Oracle. The $1.09 trillion is mostly related to AI data centers and has not yet been accounted for as liabilities on balance sheets. Signed lease agreements are generally not recognized as liabilities until the facilities become available for use. Until then, only future payments are disclosed in the notes to the financial statements.
Oracle was found to carry the greatest concentration risk. The uncommenced commitments Oracle disclosed amount to $260 billion (about 372 trillion won). That is nearly seven times its recognized lease liabilities of $37.9 billion (about 54 trillion won). Most of the commitments are related to data centers and are scheduled to commence in fiscal years 2027 to 2029, typically lasting 15 to 19 years.
Reuters noted that while these commitments are not hidden, they show that a significant portion of AI build-out projects has not yet been reflected in officially recognized lease liabilities, fixed costs and officially recognized debt metrics.
Meanwhile, Bloomberg reported the same day, citing Citadel Securities, that more than $500 billion (about 715 trillion won) in additional debt will be issued by 2028 to finance the purchase of AI chips for data centers. That is in addition to the $570 billion in existing AI-related debt the market has already absorbed.
