WhiteLine Daily｜Where Are the Hidden Mines in the AI Arms Race? Five Tech Giants with $1.65 trillion in Invisible Debt

WhiteLine Daily, bringing together the thinking of the Wu Shuo team to provide readers with the most valuable information and analysis of the day, and to capture the trend shifts in the AI era.

Today’s key takeaways:

Nikkei Asia estimates that Alphabet, Microsoft, Amazon, Meta, and Oracle have about $1.65 trillion in off-balance-sheet commitments, higher than roughly $1.35 trillion in on-balance-sheet debt. This does not mean a crisis is about to happen, nor does it mean all commitments will turn into debt; but as data centers come online and equipment is delivered, related costs will gradually flow into financial statements and cash flow.

1. How the $1.65 trillion in “hidden” debt is formed

Nikkei Asia estimates that the five companies’ off-balance-sheet commitments have grown nearly eightfold over the past four years, based on the companies’ most recent quarterly financial reports and other materials. It needs to be emphasized that these commitments are not all corporate debt in the traditional sense; they are different types of economic obligations, such as not-yet-commenced leases, long-term procurement contracts, SPV financing, and guarantees.

There are three common structures.

First, tech companies sign long-term data center leases in advance, but the facilities have not yet been built or delivered. Before lease commencement, these commitments are typically disclosed only in the footnotes of financial reports.

Second, companies lock in GPUs, servers, and other infrastructure in advance. These procurement commitments will convert into assets, expenses, or cash outflows once the equipment is delivered or services are fulfilled.

Third, data centers are built using loans by joint ventures or SPVs, while tech companies support the projects through long-term leasing, compute-power procurement, or guarantees. Even if the debt stays on the project company’s books, the economic risk borne by the tech company does not disappear completely.

The Bank for International Settlements has referred to some of these arrangements as “shadow borrowing.” Economically, these obligations are similar to debt, but they mainly sit outside tech companies’ balance sheets, strengthening linkages among tech companies, private credit institutions, insurance companies, and banks.

2. What the data truly means

Using Nikkei Asia’s broad statistical definition, the five companies’ off-balance-sheet commitments exceed their on-balance-sheet debt by about $300 billion. This means that if investors look only at balance sheets, they cannot fully judge the long-term funding needs already locked in by AI expansion. Among them, Meta’s estimated off-balance-sheet commitments are about $420 billion, nearly three times its on-balance-sheet debt; Oracle’s are about $273.3 billion, growing more than 30 times over four years.

Meta’s Hyperion data center provides a concrete example. In 2025, Meta and Blue Owl set up a joint venture, with initial development costs of about $27 billion. Blue Owl-managed funds hold 80% equity, and Meta holds 20%, but Meta also commits to lease all facilities once built and provides a capped residual value guarantee. After expansion in 2026, Meta expects its overall local investment to exceed $50 billion.

These structures can reduce the capital tech companies have to put in directly today, and can transfer some construction and financing risks to external investors. However, long-term responsibilities for rent, procurement, and guarantees still need to be supported by future revenues.

3. When the costs surface

$1.65 trillion will not all turn into新增 debt in a single quarter.

Under U.S. lease accounting rules, after data centers are put into use, non-commenced leases typically start to recognize right-of-use assets and lease liabilities. Operating leases mainly affect profit through lease expense; finance leases separately recognize asset amortization and interest. Equipment purchases may first form fixed assets, which then flow into the profit and loss statement gradually through depreciation; service purchases are recognized as costs or expenses as performance is fulfilled. FASB

So what truly matters is not whether off-balance-sheet commitments will mechanically “make it onto the balance sheet,” but whether, after a project is commissioned, revenue and utilization can cover rent, depreciation, interest, and procurement spending.

If AI and cloud business maintain high-speed growth, these investments could become the foundation for future revenue. If commercialization lags behind the pace of construction and payments, free cash flow, profit margins, and credit quality will be affected first, and stock valuation may shift from order scale to realized investment returns.

Conclusion

This iceberg may not hit a ship today, but it won’t stay underwater forever.

The real risk is not the words “off-balance-sheet,” but whether these long-term commitments will gradually become cash outflows faster than AI revenue and profits grow. Key metrics that will need continuous tracking in the future include AI and cloud revenue, data center utilization, not-yet-commenced leases, the pace of depreciation growth, net leverage, and bond yield spreads.