Oracle may slash up to 30,000 jobs to fund AI data-center expansion as US banks retreat

Oracle is considering cutting 20,000 to 30,000 jobs and selling some of its activities as US banks pull back from financing the company’s AI data-center expansion, according to investment bank TD Cowen.

The job cuts would free up $8 billion to $10 billion in cash flow, TD Cowen said in a research report seen by CIO. Oracle is also weighing a sale of its health-care software unit, Cerner, which it acquired for $28.3 billion in 2022.

The measures come as multiple US banks have pulled back from Oracle-linked data-center project lending. “Both equity and debt investors have raised questions regarding Oracle’s ability to finance this buildout,” the report said.

The financing challenge stems from the scale of Oracle’s infrastructure commitments, amounting to $156 billion in required capital expenditure, TD Cowen estimated.

Oracle did not immediately respond to a request for comment.

Borrowing costs are up

The banking retreat has driven up Oracle’s borrowing costs sharply. Lenders have roughly doubled the interest rate premiums they charge Oracle for data-center project financing since September, TD Cowen said, pushing borrowing costs to levels typically reserved for non-investment grade companies.

The higher costs have stalled deals. “Multiple Oracle data-center leases that were under negotiation with private operators struggled to secure financing, in turn preventing Oracle from securing the data-center capacity via a lease,” the report said. Without financing, private data-center operators can’t build the facilities Oracle needs, creating a bottleneck in the company’s infrastructure rollout.

Oracle has already tapped debt markets heavily, raising approximately $58 billion in just two months: $38 billion for facilities in Texas and Wisconsin, and $20 billion for New Mexico. But that represents only a fraction of what the company ultimately needs, and US banks are increasingly reluctant to provide more.

Asian banks have stepped in where US lenders are retreating, remaining willing to lend at premium rates as they seek exposure to AI infrastructure growth. That provides Oracle an alternative path for international expansion but doesn’t solve the company’s US capacity challenges. TD Cowen added that the US financing constraints raise fundamental questions about whether Oracle can grow revenue if it cannot secure the data-center capacity its customers are expecting.

Scrambling for solutions

Faced with these constraints, Oracle is pursuing multiple strategies to reduce its capital needs. The company has begun requiring 40% upfront deposits from new customers, TD Cowen said, effectively asking clients to help fund the infrastructure buildout. It’s also exploring “bring your own chip” (BYOC) arrangements where customers would supply their own hardware, shifting capital requirements off Oracle’s books.

TD Cowen said some combination of BYOC and workforce reductions represent the most likely path forward, since BYOC would directly address the capital expenditure challenge while job cuts would improve cash flow. But both options carry risks. BYOC could require renegotiating existing contracts that assume Oracle provides the hardware, while major layoffs could affect the company’s ability to execute its infrastructure plans.

The potential workforce reduction would be Oracle’s largest in recent years. The company cut an estimated 10,000 jobs in late 2025 as part of a $1.6 billion restructuring plan. Oracle has also repeatedly reduced headcount at Cerner since acquiring the healthcare technology company, including layoffs in 2023 following problems with a Veterans Affairs contract.

The financing pressures are already reshaping Oracle’s customer relationships. OpenAI has shifted its near-term capacity needs to Microsoft and Amazon, the report said, a significant change from just months earlier when Oracle leased approximately 5.2GW of US data-center capacity across Texas, Wisconsin, Michigan, and New Mexico specifically for OpenAI workloads.

More broadly, Oracle’s data-center procurement has slowed dramatically. TD Cowen said Oracle was “notably absent” from the list of companies with major long-term US data-center roadmaps, noting that “financing struggles have led to a deceleration in incremental data-center procurement.” Private operators who would normally be signing large deals with Oracle are instead holding back “as the market digests the current Oracle financing requirements,” the bank said.

A shared infrastructure risk

Industry analysts differ sharply on how seriously enterprises should view Oracle’s situation.

Sanchit Vir Gogia, chief analyst at Greyhound Research, sees the banking divergence as a critical warning sign. “The difference in sentiment between US and Asian banks isn’t just a minor detail; it’s the first serious sign of financial friction in Oracle’s hyperscale ambitions,” he said. The $300 billion OpenAI deal may look impressive, he added, but “when you look closer, it’s built on backlog with no guaranteed revenue and massive capex requirements.”

Gogia argued that enterprises need to fundamentally rethink how they view Oracle cloud contracts. “CIOs need to treat Oracle’s cloud buildout not as a service agreement, but as a shared infrastructure risk,” he said. “If they can’t fund it, they can’t build it. And if they can’t build it, you can’t run your workloads.”

Franco Chiam, VP for cloud and data-center research at IDC Asia/Pacific, takes a more measured view. He suggested the potential Cerner sale “could indicate a consolidation of its core services (AI-driven infrastructure) rather than a sell-off to fund/stop bleeding.” Oracle’s underlying business remains strong, he noted, with cloud infrastructure revenue growing 66% year over year in the three months ended November 30, and GPU-related infrastructure up 177%, according to the company’s latest earnings report.

Still, even Chiam acknowledged that enterprises need to protect themselves. Both analysts agreed that companies should implement multi-cloud, multi-vendor platforms to reduce dependency on any single provider, though they differed on how urgently customers need to act.