As US Big Tech companies accelerate their competition to invest in artificial intelligence (AI), "hidden debt" that does not appear on financial statements is surging, according to a new analysis.
An analysis by Nihon Keizai Shimbun of recent financial statements and related documents from Alphabet, Microsoft, Amazon.com, Meta, and Oracle found that the five companies' AI-related contractual obligations totaled 1.65 trillion dollars (about 2,455 trillion won), an eightfold increase in four years, the newspaper reported on the 20th. The figure exceeds the total actual debt of 1.35 trillion dollars (about 1,997 trillion won) recorded on their balance sheets.
Meta had the largest debt burden. Meta's off-balance-sheet debt stood at about 420 billion dollars (about 621 trillion won), roughly 2.8 times the debt recorded on its books.
Such "hidden debt" stems from the race to secure AI data centers and graphics processing units (GPUs). Big Tech companies are signing contracts to pre-purchase GPUs and servers on a multi-year basis or to lease large-scale data centers over the long term to secure AI computing power. However, GPU purchase contracts for units not yet delivered, or data center lease contracts before operation, are not reflected as debt on the balance sheet under accounting standards. Instead, they are treated as "off-balance-sheet debt" recorded only in the footnotes of quarterly reports.
While this is legitimate accounting treatment from the companies' perspective, critics point out that it makes it difficult for investors to grasp the actual financial burden at a glance. Investors cannot easily discern the real risks.
A representative case is Oracle. Oracle is pursuing "Stargate," a mega data center project, together with OpenAI, using a method of leasing data centers operated by external companies over the long term. Oracle's off-balance-sheet debt reached 273.3 billion dollars (about 404 trillion won) as of the end of May this year, more than 30 times higher than four years ago.
Concerns are also growing in the market. Nikkei reported that "Morgan Stanley, in an investor report, analyzed the increase in data center lease contracts as a major risk factor," and that "international credit rating agency Moody's also expressed concern that the expansion of pre-operation lease contracts could increase financial burdens."
The Bank for International Settlements (BIS) defined this method of financing as "Shadow Borrowing," warning that raising external funds without increasing actual debt could heighten market risk.
Big Tech companies are drawing a line against criticism of excessive investment. Combined cloud business order backlog for the three companies—Microsoft, Alphabet, and Amazon—totaled 1.45 trillion dollars (about 2,145 trillion won) as of the end of March this year, and they maintain that future revenue will sufficiently offset the burden of long-term contracts.
Recently, the use of investment fund money has also been increasing. Meta has established a joint venture with US investment firm Blue Owl Capital to build a mega data center in Louisiana. Meta reduced its initial investment burden by choosing a structure in which it invests only a partial stake in the operator and then uses the facility through a long-term lease, but its off-balance-sheet debt increased significantly instead.
Nikkei analyzed that "the AI industry is currently growing into a structure in which funds circulate among Nvidia, Big Tech, data center operators, and AI startups," and that "if supply expansion outpaces actual demand, overinvestment could occur, which is similar to the structure during the early 2000s dot-com bubble when telecom equipment makers supplied funds to internet companies and expanded investment."
It added that "amid the era of generative AI, an astronomical investment race continues," pointing out that "hidden debt not shown on financial statements is emerging as a new risk factor for the global AI industry."
