Oracle’s ability to procure incremental US data center capacity is facing difficulties amid doubts that the company can fund the buildout of OpenAI-linked projects, according to a TD Cowen report.

Oracle has raised around $58 billion in debt to build out data center projects for OpenAI, including $38 billion in financing for its campuses in Texas and Wisconsin and $20bn for a campus in New Mexico.

However, in its Data Center Channel Checks update, TD Cowen said that Oracle will need to procure around three million GPUs and other IT gear to support its existing agreements, which has raised questions from both equity and debt investors who are doubting the company’s ability to finance the buildout.

The investment firm said this was evidenced by widening of Oracle's CDS spreads and pressure on Oracle's stock and bonds.

TD Cowen calculated that OpenAI has an implied $156bn capex requirement, assuming a conservative $30m per MW in IT fit out cost. Compounding this is OpenAI’s outstanding $1.4 trillion in multi-year commitments. This has led a number of US banks to retreat from Oracle-linked data center projects due to its agreements with OpenAI, the investment firm said, and in instances where US banks are willing to lend, borrowing costs have risen.

Banks in Asia that are looking to gain exposure to the AI sector are still willing to lend to operators undertaking Oracle leases, but the significant reduction in US financing has led to a significant slowdown for Oracle and raised questions over its ability to continue growing its revenue.

The report also noted that a number of Oracle data leases under negotiation have been unable to secure financing, hampering Oracle’s efforts to secure capacity via lease.

TD Cowen said that Oracle is now requiring 40 percent upfront deposits to mitigate the incremental capex requirement for revenue growth and is evaluating other paths to address financing questions.

Earlier this month, Oracle was sued by bondholders over accusations that the company concealed how much debt it would need to sell to support the AI buildout for Oracle's $300 billion OpenAI deal.

Late last year, the company denied rumors that its OpenAI data centers were suffering delays.

Highly robust

Looking more broadly, the report paints a positive growth picture for the industry, with overall data center leasing hitting a record 16.4GW in 2025.

Some 5.1GW of capacity was leased in Q4 2025, with 7.4GW in Q3, 2.0GW in 2Q, and 1.9GW in 1Q, making up the record 16.4GW figure. The investment banking company noted that Q4, which is historically the slowest data center leasing quarter, saw the second-highest amount of leasing in 2025.

Hyperscalers, including Oracle, led the leasing tables in 2025 and, going into 2026, TD Cowen sees a “highly robust” US data center leasing pipeline, with around 9.2GW of live US data center leases in progress and “healthy diversity” across hyperscalers.

The report said that this reflects demand from AI labs and hyperscalers, with Meta targeting 15GW of US data center capacity by 2028, OpenAI targeting 10GW by 2029, and Microsoft targeting 10GW by 2030.

However, demand acceleration is resulting in increasing data center build costs and skilled labor costs, which is in turn driving data center spot market rents higher. TD Cowen said that this “bodes positively” for data center lease renewal rates/spreads into 2026, and that companies such as Digital Realty and Equinix will benefit.

The report also identified a “moderate elongation” of equipment lead times for uninterruptible power supplies and power distribution units, which the investment firm said was indicative of broader equipment lead time trends as data center demand exceeds supply. This has also led to an increase in pricing for equipment, which TD Cowen again identifies as positive for suppliers, such as Vertiv.