Oracle’s Commercial Real Estate Problem

Oracle [is] tempting fate by putting itself in a position where not only does the AI boom need to continue and pure play AI companies need to become profitable, but the latter needs to happen...soon.

As of November 30, 2025, Oracle announced almost $250 billion in planned new lease commitments, substantially all of which are related to data centers and “cloud capacity arrangements.”

The filing in which this announcement was made notes that the leases commence over the next two to three years and generally contain lease terms of 15 to 19 years. The leases are not reflected on Oracle’s consolidated balance sheets and Oracle likely hasn’t assumed any directly related expense thus far aside from legal, commissions, and other typical initial costs that are minor compared to overall lease expenses.

In simple terms, what this means is that Oracle has contractually committed to incurring and paying $250 billion in rent expense over the next 20 years in support of its push into the artificial intelligence space.

When a company makes such a commitment, they must intend to earn revenue from whatever activity takes place in those leased buildings. In order to make the venture feasible, that revenue must exceed the related expenses.

But the related expenses don’t stop at simple rent payments. Oracle leases the data centers, but they also incur significant expenditure for the hardware inside the data centers. Hardware that, by all accounts, is highly capital intensive and requires refreshed expenditure as existing chips obsolesce over the course of three to five years, maybe quicker.

Utility costs for data centers are notoriously high, a reflection of massive electricity and water consumption therein. These costs are generally paid by tenants and certain to be paid – directly or indirectly – by Oracle in the specific case of their data centers.

Buildout costs – the expenditures required to populate the empty buildings with the equipment and fixtures necessary for their chosen operation – may be subsidized partially up-front by Oracle’s landlords, but represent a potentially large expenditure for Oracle on top of the $250 billion in occupancy costs.

General operating costs such as staffing and maintenance of the data centers will also add ongoing expense to the operation.

Financing the entire operation, presumably with debt – Oracle is not profitable from a free cash flow perspective and therefore can’t self-finance such a project like mega-tech companies Google and Microsoft – will be additional cost to Oracle, particularly any statutory debt payments associated with the borrowing.

All in, a better estimate of Oracle’s expected cost over the 15- or 20-year lease period is $500 billion to $1 trillion.

Cash Flow Negative

Today, Oracle is free cash flow negative on roughly $50 billion in annual revenue. Based on the above, a reasonable estimate of incremental leasing and other operational costs related to their AI investment over the next 20 years is $800 billion. They would need to earn incremental annual revenue of $40 billion over the coming two decades – an 80% increase over current revenues – just to break even on the margin.

Such growth may seem feasible for the tech industry in a period of forever bubbles, but consider who the customers are for the products and services Oracle will be offering through these leases. Oracle is building out “computing power” for players in the AI industry. OpenAI, with whom Oracle has a well-publicized relationship, is expected to be one of their primary customers.

But OpenAI loses cash and is essentially reliant on the venture capital industry for its future survival through cash infusions.

Oracle itself is on shaky financial ground, even now. Currently losing money on a free cash flow basis, the investment community has viewed Oracle’s planned expansion as anywhere from heroic to suicidal, resulting in massive stock price swings and a significant increase in premiums on its credit default swaps (“CDS”). These increases in CDS premiums suggest the market believes Oracle faces a higher risk of future default on their credit obligations.

High Risk for Meager Return

Oracle appears to be the poster child for how the AI bubble could burst. Highly profitable mega techs like Google and Microsoft are far less susceptible, given their higher return on invested capital (“ROIC”) and healthy cash positions.

Oracle’s ROIC on AI projects is in the single digits by at least one credible measure, thus they have put themselves in a position where they’re committing to massive near-future spending ($800 billion over the next twenty years) with a highly questionable source of revenues (AI companies that currently burn cash and rely heavily on VC infusions) to earn only paltry returns.

Oracle appears to be tempting fate by putting itself in a position where not only does the AI boom need to continue and pure play AI companies need to become profitable, but the latter needs to happen fairly soon. If not, a pullback in credit resulting in less VC funding for AI companies will mean serious trouble for Oracle.