Tech Giants' $1.65 Trillion Hidden Debt Exposed: Is the 'Shadow Borrowing' Behind the AI Arms Race Collapsing?

As funding commitments for massive artificial intelligence (AI) data center projects continue to swell, coinciding with recent declines in tech stock prices, Wall Street analysts are sounding the alarm: the seemingly endless AI expansion wave among tech giants may be facing a sudden halt. Mitch Feierstein, CEO of Glacier Environmental Fund Ltd. and author of 'Planet Ponzi,' points out that recent second-quarter financial reports and disclosures from Alphabet, Amazon, Meta, Microsoft, and Oracle have begun to reveal concerns about this massive AI infrastructure push. For most investors, accurately gauging the true financial burden behind these projects remains extremely difficult.

According to an analysis by The Nikkei of recent financial reports from major tech firms, the combined off-balance-sheet commitments of Alphabet, Amazon, Meta, Microsoft, and Oracle now total a staggering $1.65 trillion (approximately NT$52.8 trillion). These items excluded from the balance sheet primarily include data center lease agreements, graphics processing unit (GPU) supply contracts, and server procurement agreements.

This $1.65 trillion in hidden liabilities already far exceeds the $1.35 trillion in total book debt these five companies have listed on their balance sheets.

Valuation Distortion: Hidden Exposure Could Soar to $5 Trillion

However, Feierstein believes the figure calculated by The Nikkei still severely underestimates the actual risk in the tech industry. 'The scale of $1.65 trillion in extremely opaque 'circular finance' is definitely a conservative estimate,' Feierstein stated bluntly in a media interview. 'Multiplying the total exposure by 2 to 3 times might be closer to the real situation.'

If Feierstein's estimate is applied and The Nikkei's data is tripled, the total potential exposure would reach $4.95 trillion—nearly $5 trillion (approximately NT$160 trillion)—an astonishing scale.

Under current accounting standards, such financing arrangements are not necessarily illegal or considered concealment. Accounting rules allow companies to temporarily exclude chip and server procurement contracts that have 'not yet been delivered' and data center lease agreements for facilities 'not yet officially operational' from their balance sheets. Companies are only required to disclose these items in the 'notes' section of quarterly and annual financial reports, which are often harder for average investors to notice than traditional book debts.

The market's greatest concern is that as these new data centers and infrastructure come online, the related commitments will gradually 'move onto' the balance sheet, converting into actual liabilities. At that point, if AI demand falls short of expectations or construction plans are canceled, tech companies and their asset financing partners could face severe asset impairments, penalty payments, and massive cash flow pressures.

Morgan Stanley, Rating Agencies, and the BIS Issue Warnings

Investment firm Morgan Stanley has already conducted an in-depth analysis of these hidden financial commitments in its latest report. Coincidentally, credit rating agency Moody's Ratings also issued a warning earlier this year, stating that five major U.S. data center operators have reported about $969 billion in lease commitments, most of which are for leases not yet officially launched. Moody's emphasized that these contracts will significantly increase corporate adjusted debt burdens and lease-related cash outflows in the future.

The Bank for International Settlements (BIS) used even stronger language in its quarterly publication to describe this opaque financing model—'shadow borrowing.'

As the 'central bank of central banks,' the BIS pointed out that obligations economically equivalent to debt are largely existing outside corporate balance sheets, deepening financial interconnections among tech giants, private credit funds, insurance companies, and traditional banks. The BIS warned that due to the enormous scale of AI investments, tech companies can no longer rely solely on internally generated cash flows and have turned to large-scale debt and private credit. The sustainability of this AI boom entirely depends on whether AI-related businesses can meet the market's 'extremely high' profit expectations.

Feierstein calls this high-valuation phenomenon 'AI Mania' and directly states that it is sustained through circular capital flows. For example, chip manufacturers and tech giants invest in AI developers or data center operators, who then recycle part of the funds back to these investors or business partners to purchase chips, cloud services, and computing power. While such transactions may not involve fraud, they greatly blur how much 'real' AI demand actually exists in the market.

The scale of off-balance-sheet commitments among major tech firms is also remarkably high:

· Meta: Off-balance-sheet commitments reach $420 billion, nearly three times its book debt. Meta is also collaborating with a fund under Blue Owl Capital to build a data center in Louisiana, originally estimated at $27 billion. Meta holds a 20% stake in the operating company and leases the facility, but if certain termination conditions occur, Meta must bear investors' losses. In mid-July, Meta revealed that the total investment for the campus is expected to exceed $50 billion.

· Oracle: As of the end of May, its off-balance-sheet commitments have surged to $273.33 billion, over 30 times the level four years ago. The company is aggressively pushing large-scale data center construction, including the 'Stargate' massive computing power project in collaboration with OpenAI.

· Microsoft, Alphabet, Amazon: In response to questions about hidden liabilities, all three point to massive cloud contract backlogs as solid evidence that future revenues will cover expenditures. According to The Nikkei, as of the end of March, the combined announced cloud and related backlogs of these three companies totaled approximately $1.45 trillion. AWS CEO Matt Garman also emphasized that the company's investments are 'definitely not speculative.'

Wall Street Sell-Off: Oracle Downgraded, Korean Market Enters Bear Territory

However, financial markets have already begun questioning the cost of this AI arms race, with Oracle being the most prominent warning sign.

Oracle's stock price hit a record high of $345.72 on September 10, 2025, but has since plunged to around $126, a cumulative drop of over 63%. Global credit rating agency S&P Global Ratings downgraded Oracle's credit rating to 'BBB-,' just one notch above junk status, citing massive capital expenditure needs, uncertain profit prospects, and excessive risk concentration on OpenAI. Feierstein analyzes that Oracle's sharp decline reflects excessive debt and leverage, possibly marking the beginning of the market's reassessment of overall AI valuations.

Beyond individual companies, international markets are also being impacted. South Korea's KOSPI index, heavily reliant on the semiconductor industry, has declined about 28% from its peak on June 19, officially entering bear market territory. Amplified by leveraged products and derivatives, market volatility intensified, and sell-offs expanded further after the U.S.-Iran conflict erupted in late February, shaking global markets.

Feierstein also cites SpaceX as another example of massive capital expenditure, high valuation, and uncertain payback periods. SpaceX conducted its first public offering (IPO) on June 12, pricing shares at $135, issuing about 555.6 million shares, raising approximately $75 billion, pushing its valuation to nearly $1.77 trillion. Although its market cap briefly surged toward $3 trillion, it subsequently plummeted, closing at $115.26 on July 22, with an intraday low of $115.19—the lowest in its history.

Son Masayoshi Rejects Claims of AI Bubble

Feierstein notes that the market value erased from SpaceX's peak has already exceeded $1.6 trillion. 'Among all publicly traded companies in the U.S., only about six have a total market cap exceeding the $1.6 trillion that SpaceX just lost,' he said. The first space company in history to surpass a $1 trillion valuation is expected to release its first quarterly report since going public on August 4, followed by the unlocking of 911.5 million restricted shares on August 6, potentially triggering a new wave of selling pressure.

Additionally, Feierstein raises corporate governance concerns about the high-vote B-class shares held by SpaceX founder Elon Musk, arguing that this voting structure makes it nearly impossible for other shareholders and the board to block major transactions. According to IPO-related public information, SpaceX incurred nearly $5 billion in losses in 2025, and whether it can convert its expensive rocket launches, Starlink satellites, and AI projects into long-term stable profits remains uncertain.