To The Moon

Investors closely track Big Tech’s quarterly capital expenditures, but reported CapEx only captures part of the AI buildout. Nine major technology companies reportedly have about $3 trillion in future lease, chip-purchase and infrastructure commitments that are not yet fully reflected on their balance sheets.

1. Where Did the $3 Trillion Come From?

According to a Wall Street Journal analysis of financial-statement footnotes, nine major technology companies reported roughly $600 billion in combined CapEx over their latest 12-month periods.

However, their broader future commitments approach $3 trillion, including approximately:

• $1.2 trillion in data-center leases that have not yet commenced;

• $1.9 trillion in long-term purchase agreements covering chips, memory, power and other infrastructure.

This does not mean the companies have already spent $3 trillion. It also should not be treated as $3 trillion of conventional debt.

It represents a future bill that has already been partially locked in through contracts. These obligations may appear in financial-statement footnotes without being fully recognized as current liabilities.

Source: Wall Street Journal

2. Why Isn’t It Included in Reported CapEx?

Consider two different ways to build a data center.

If Microsoft spends $10 billion today to construct a facility, that amount is recorded as capital expenditure.

If Microsoft signs a 20-year lease for a data center that will begin operating in 2028, the full future commitment may not yet appear as a lease liability. Accounting recognition generally begins when the lease commences.

The same logic applies to chip purchases.

Cloud companies are signing multiyear contracts to secure GPUs, networking equipment, storage and electricity. Until products are delivered or payments become due, the full value may remain in the footnotes instead of appearing as current CapEx or debt.

Reported CapEx shows what is being spent now. Contractual commitments reveal how much future spending has already become difficult to cancel.

3. How Off-Balance-Sheet AI Financing Works

The Bank for International Settlements says hyperscalers are increasingly using joint ventures and special-purpose entities to finance data-center expansion.

A common structure looks like this:

1. A technology company partners with infrastructure investors or private-credit funds.

2. A separate entity raises debt and builds the data center.

3. The technology company owns only a minority stake.

4. It signs a long-term lease or commits to buying power and computing capacity.

5. Rental cash flow is then used to service the project’s debt.

This converts a large upfront investment into years of operating expenses.

Most of the project debt remains outside the hyperscaler’s balance sheet, although the company still has a long-term payment commitment. The BIS describes these arrangements as a form of “shadow borrowing.”

Traditional borrowing is also increasing. Bond issuance from Amazon, Alphabet, Microsoft, Meta and Oracle is expected to double to about $250 billion in 2026 and nearly double again by 2027.

BIS research on AI infrastructure financing
Big Tech financing data: Reuters

4. Nvidia’s $105 Billion Guarantee Shows How the Model Works

The newly announced Nvidia-OpenAI data-center agreement provides a real-world example.

OpenAI has agreed to lease an Ohio data center developed by SoftBank-owned SB Energy for 20 years. The site could eventually reach eight gigawatts of capacity, with the first 800 megawatts expected to come online in 2028.

Nvidia will:

• Invest $1.5 billion in SB Energy;

• Serve as the facility’s exclusive chip supplier;

• Provide a guarantee of up to $105 billion.

The guarantee does not cover the project’s entire cost or all of OpenAI’s obligations. It covers portions of lease and power payments, along with a commitment supporting the facility’s minimum residual value.

If OpenAI defaults, Nvidia could be responsible for part of the shortfall remaining after the owner attempts to re-lease or sell the facility.

This arrangement helps the project attract equity and debt financing. OpenAI secures long-term computing capacity without paying the full construction cost immediately, while Nvidia locks in future demand for its chips.

Nvidia CEO Jensen Huang said the deal is not circular financing. Still, the structure shows how closely connected AI suppliers, customers and financiers have become.

Source: Reuters

5. Why Are Companies Making Such Large Commitments?

Three critical resources remain scarce.

Chips

Advanced GPUs, high-bandwidth memory, networking equipment and enterprise storage are still supply-constrained. Long-term contracts help companies secure future deliveries.

Land and Power

AI data centers require reliable electricity, cooling systems, substations and high-speed network connections. Suitable locations are limited.

Time

A data center can take 12–18 months to move from construction to revenue generation, while larger projects may take longer. Companies that wait for demand to become completely certain risk losing customers and market share.

This helps explain why AI hardware orders remain strong. A significant portion of future demand has already been locked in and will not disappear immediately because of a short-term market correction.

6. What Does It Mean for AI Stocks?

Hardware suppliers currently have the easier business model.

They can recognize revenue when equipment is delivered. Hyperscalers must wait for customers to use AI services before earning back their investment.

That helps explain why semiconductor stocks have outperformed some of the largest AI spenders. The suppliers are already seeing sales, while the buyers are still waiting for returns.

7. The Real Question Is the Speed of Monetization

Reuters estimates that hyperscalers could generate about $340 billion more in annual operating cash flow in 2027 than in 2025. Their CapEx, however, is expected to increase by roughly $534 billion over the same period.

Existing business growth may therefore be insufficient to cover the entire expansion.

Corporate disclosure around AI returns also remains limited. A Goldman Sachs review of S&P 500 earnings reports and conference calls found that:

• 11% quantified specific AI use cases;

• 2% quantified AI’s contribution to earnings;

• 7% discussed AI-related expenses.

Investors should watch four indicators:

• AI revenue growth versus incremental CapEx;

• Operating cash flow after CapEx, leases and interest;

• Data-center utilization rates;

• Cloud pricing as new capacity comes online.

Source: Reuters

Tiger Comments’ Take

The $3 trillion in commitments provides strong visibility for the AI infrastructure supply chain.

Data-center projects will not stop immediately during a market correction, and multiyear chip, storage, networking and power contracts are difficult to cancel. That continues to support companies such as NVDA, MU, SNDK, WDC and VRT.

The larger risk comes later.

If AI-service revenue fails to grow fast enough to cover depreciation, rent, interest and electricity costs, hyperscalers will face pressure on free cash flow. Slower future CapEx would then spread the pressure back to chip and equipment suppliers.

The AI trade now has two different questions:

• Who benefits from infrastructure spending already under contract?

• Who can eventually prove that the investment generates sufficient revenue and profit?

The first question favors hardware suppliers. Amazon, Microsoft, Alphabet, Meta and Oracle will ultimately have to answer the second.

Today’s Poll

What is the biggest risk to the AI investment boom?

A. AI revenue takes too long to materialize
B. Data-center capacity eventually exceeds demand
C. Debt, lease and interest costs rise too quickly
D. Hardware becomes obsolete faster than expected
E. Demand remains strong enough to justify the spending

Disclaimer: This post is for informational and discussion purposes only and does not constitute investment advice. Off-balance-sheet commitments include future contractual obligations and financing arrangements; they should not be treated as equivalent to current debt or cash spending.