Roughly $300 billion in chip-collateralized special purpose vehicles has now been identified or processed. That is a lot of AI hardware sitting somewhere other than a company’s balance sheet. Big Tech and AI firms have reportedly used these structures over the past year to finance data center buildouts and chip purchases. The goal is simple: get the GPUs without letting the debt crowd the financial statements investors actually read. How the $300 billion got built
Think of a special purpose vehicle as a separate legal box.
A company sets it up, the box borrows money and buys the chips, and the company leases the hardware or guarantees the deal from the outside. On paper, the parent stays lean.
In practice, it is still very much on the hook through leases and guarantees. Private credit giants are writing the checks. Blackstone and Apollo Global Management have backed chip-backed financing arrangements, which have grown more popular as demand for AI infrastructure keeps climbing.
Alphabet offers one of the starkest examples. Its data-center guarantees jumped from $16.9 billion to $43.8 billion in just six months, more than doubling in that window. Less than 2% of that amount shows up on Alphabet’s balance sheet. Anthropic moved even faster.
SPVs tied to the AI lab accumulated approximately $71 billion in chip-leasing debt in roughly 60 days, by early August 2026. A single $35 billion package involving Apollo and Blackstone arrangements anchored that run. Amazon is still at the table. As of October 2026, the company is reportedly exploring an $8 billion SPV that would buy Nvidia’s Grace Blackwell chips and lease them back to Amazon.
Nvidia wants to be the bank, too
The biggest number in this story belongs to a plan, not a completed deal. In August 2026, Nvidia announced a financing platform designed to attract more than $500 billion in third-party capital for chip-backed financing.
The pace has been notable as well. Activity from mid-September through October points to a coordinated industry push to find financing that can keep up with AI’s appetite for compute. What this means for lenders, investors and the AI buildout
The core risk sits in the collateral itself.
Chips are the asset backing these loans, and AI hardware loses value quickly as newer generations arrive. Market participants have flagged concerns about how rapid depreciation could affect long-term collateral recovery. If a borrower stumbles years into a lease, the GPUs left in the SPV may be worth far less than the debt they were meant to secure. For shareholders in Big Tech, the main issue is visibility.
When less than 2% of Alphabet’s guarantees appear on its balance sheet, the headline numbers can understate how much the company has actually committed to the AI buildout. Watch three things from here: whether Amazon finalizes its $8 billion Grace Blackwell deal, how much capital Nvidia’s platform actually attracts against its $500 billion target, and whether lenders start demanding tougher terms as depreciation concerns build. Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
