Oracle is negotiating with Apollo and Goldman Sachs to finance a large purchase of AI chips through a structure that would keep the spending off its own balance sheet, as the company's total debt has reached $169.1 billion against $37.1 billion in cash and short-term investments.

The talks, first reported by the Wall Street Journal, center on a leasing arrangement: investors would fund a separate company that buys Nvidia GPUs and leases them to Oracle over time. No deal amount has been disclosed, and Oracle is still speaking with other potential partners.

The scale of the problem is easy to see. The WSJ noted that Nvidia chips for a single 1-gigawatt data center cost tens of billions of dollars — a price tag that conventional bond issuance struggles to absorb repeatedly.

Why does Oracle want the debt somewhere else?

The company's debt load has grown at a pace that has already drawn legal challenges. Total debt climbed from $99.5 billion in November 2024 to $169.1 billion in August 2026.

Bondholders sued in January, alleging that Oracle understated how much more it would need to borrow when it sold $18 billion of bonds in September 2025. A lease structure lets Oracle add chips without adding bonds — at least on paper.

The pressure comes from a capex surge that has outrun operating cash flow:

• Capital expenditure hit $28.5 billion in the quarter ended August 31, up from $8.5 billion a year earlier.
• Over the last four quarters, Oracle spent $75.7 billion on capex and burned $28.7 billion in free cash flow.
• Operating cash flow reached roughly $23 billion last quarter — rising, but nowhere near enough to fund the buildout on its own.

That gap is the commercial engine behind the leasing talks. Oracle needs GPUs for its AI infrastructure commitments, and it cannot finance them from operations or from incremental bond sales without aggravating both its litigation exposure and its credit profile.

What does the leverage ratio actually show?

There is a counterintuitive detail in the numbers. Oracle's leverage has improved on the standard measure: total debt to EBITDA fell to 3.78x in August from a peak of 4.70x in February.

That improvement does not close the funding gap. EBITDA is growing fast enough to outrun borrowing on a ratio basis, but the company still cannot generate the cash needed for its chip purchases. The leasing structure, in other words, is not a rescue — it is a way to spread payments over time.

Whether it helps the optics depends on accounting. The effect on debt to EBITDA will hinge on the lease classification and on how rating agencies and covenant calculations treat the lease obligations. Off-balance-sheet treatment is not guaranteed.

How does this fit the wider AI financing wave?

The WSJ report described three parallel financing approaches among companies racing to buy AI silicon:

• SpaceX is seeking direct borrowing.
• Broadcom is arranging financing for OpenAI.
• Oracle is exploring chip leasing.

The common thread is that AI infrastructure costs have outgrown what traditional corporate finance can comfortably absorb. Each company is adapting the structure to its own constraints: SpaceX against its private-company position, Broadcom as an enabler for its customer, and Oracle against a public balance sheet already loaded with debt and bondholder scrutiny.

For a company carrying $169 billion of debt and negative free cash flow, renting compute capacity has clear logic. It spreads payments over the useful life of the chips rather than demanding tens of billions upfront.

What should investors watch next?

Two numbers will define the deal's significance. The first is its size, which remains undisclosed. The second is whether the lease terms align with Oracle's reported $300 billion OpenAI contract — described as lasting roughly five years starting in 2027.

If the lease maturity profile matches that revenue stream, Oracle would effectively match its chip financing against contracted income, softening the balance-sheet impact of the largest infrastructure buildout in its history. If the terms are shorter, the obligations would land while the OpenAI revenue is still ramping, and the funding gap would persist in a different form.

Either way, the direction of travel is clear: AI chip purchases have become large enough that the financing structure — not the chip order itself — is now the story for Oracle, its creditors, and the private credit funds lining up behind the GPU boom.