Oracle, Broadcom and SpaceX are each seeking debt deals to pay for AI chips, the Wall Street Journal reports, a signal that the cost of accelerator procurement has grown so large that even well-capitalized technology companies now prefer to finance it rather than draw on cash reserves.

The report names three borrowers with distinct positions in the AI supply chain. Oracle operates cloud infrastructure that runs customers' AI workloads and has committed to large-scale GPU deployments. Broadcom designs custom AI accelerators and merchant networking silicon, and counts hyperscalers among its largest customers. SpaceX, through its Starlink unit, is buying AI hardware for compute used across its satellite and telecom operations.

Why are companies borrowing to buy chips?

The move to debt financing reflects the sheer scale of AI infrastructure spending. NVIDIA's flagship AI accelerators carry price tags that run into the tens of thousands of dollars per unit, and cloud operators routinely order them in shipments measured in the hundreds of thousands of chips. Financing those orders through capital markets spreads the outlay over time and preserves balance-sheet liquidity.

For Oracle, the borrowing supports a cloud business that has signed multi-billion-dollar contracts to supply AI compute. For Broadcom, the financing need stems from the working-capital intensity of its custom accelerator business, where the company funds chip design and supply commitments for large customers. For SpaceX, AI compute has become a growing line item as Starlink scales.

What does this signal for the supply chain?

Debt-funded chip purchases mark a shift in how the AI buildout is financed. Earlier waves of data-center expansion were funded largely from operating cash flow. The fact that Oracle, Broadcom and SpaceX are now structuring dedicated debt deals, according to the Journal, indicates that AI capex has outpaced internal cash generation even at companies with substantial revenue.

That shift carries implications for lenders and investors. Banks and debt investors are effectively becoming financiers of the AI hardware cycle, taking exposure to the pace of accelerator shipments and the revenue those chips ultimately generate. If AI demand stays strong, the financing model extends the buying power of chip customers. If demand falters, debt service on hardware that depreciates quickly becomes a harder burden to carry.

The borrowing also concentrates dependence on a narrow supplier base. Most of the AI chips these companies are buying trace back to NVIDIA's accelerator families and to TSMC's leading-edge foundry nodes, meaning the debt is ultimately financing capacity at a handful of suppliers whose output remains constrained.

Who benefits?

Chip suppliers stand to gain most directly. Financed purchasing lets customers place larger orders than their cash positions would otherwise allow, supporting the order books of NVIDIA, Broadcom's custom silicon division, and the foundries and advanced-packaging providers behind them.

For the borrowers, debt keeps equity dilution off the table and lets each company match financing terms to the expected life of the assets. Whether that arithmetic holds depends on how quickly AI accelerators depreciate in value as newer product generations arrive — a depreciation cycle that has historically been shorter for GPUs than for traditional data-center hardware.

The Journal's report did not specify the sizes of the individual debt deals or the lenders involved, and the companies have not disclosed final terms. The scale of the borrowings, once they price in public or private markets, will offer one of the clearest measures yet of how much debt the AI hardware cycle is now absorbing.