Amazon.com (AMZN) is considering moving roughly $8 billion worth of advanced NVIDIA (NVDA) AI chips into a special purpose vehicle and bringing in outside investors to fund these assets, in order to ease pressure on its own balance sheet.

According to reports, Amazon has reached out to potential investors in recent weeks to gauge market interest in this transaction structure. Under the plan currently under discussion, the company would transfer thousands of Grace Blackwell chips into a special purpose vehicle, or SPV, which would hold these chip assets. These chips are currently being used in data centers across the United States.

After completing the transfer, Amazon would not stop using the related equipment. Instead, it would lease the chips back from the SPV and continue using them for its own AI and cloud computing operations. At the same time, the SPV would raise funds from outside investors by issuing debt, thereby providing financial support for the chip assets.

This means Amazon is trying to convert large-scale AI chip investments that would originally need to be directly included in its own capital expenditures into a financing model in which assets are externally held and Amazon uses them through long-term leases. From a financial structure perspective, this model can reduce the one-time asset purchase burden on Amazon and allow part of its AI infrastructure investment to be borne by external capital markets.

According to reports, Amazon also plans to offer external investors up to 10% equity in the vehicle, while it would not hold any equity in the SPV itself. If the plan ultimately moves forward in this form, the SPV would become an independent financing and asset-holding entity, while Amazon would mainly use the computing equipment as a tenant.

The core purpose of this type of structure is to make AI infrastructure construction more asset-light. One of the most expensive assets in AI data centers today is high-end GPUs and related accelerator chips. As purchases of high-performance chips such as Grace Blackwell continue to expand, the chips themselves have become an important part of the capital expenditures of hyperscale cloud providers.

Amazon is expected to spend more than $200 billion in capital expenditures this year, with a considerable portion going to AWS for purchasing more AI chips and building new data centers. With such a massive scale of investment, even Amazon has begun looking for financing methods that reduce balance sheet usage.

This is not simply a chip transaction, but a signal that the financing model for AI infrastructure is changing. In the past, hyperscale cloud providers usually bought servers, GPUs, and data center equipment themselves, then kept these assets on their own balance sheets. As AI infrastructure investment has reached hundreds of billions of dollars, this model has put increasing pressure on cash flow and capital expenditures.

If Amazon's transaction can be completed, external investors would effectively become the funding providers for AI chip assets, while Amazon would obtain long-term usage rights through leasing. This model is similar to financializing AI compute infrastructure: chips are no longer just fixed assets of technology companies, but can also be packaged into investment assets capable of generating rental cash flow.

Therefore, what is truly noteworthy about this potential $8 billion deal is not whether Amazon has reduced its demand for NVIDIA chips. On the contrary, Amazon still needs these chips; it simply wants to change the way it buys and holds them. The core logic behind this is that AI capital expenditures are still expanding, but hyperscale cloud providers are trying to shift more and more infrastructure financing pressure to external capital.

If this model is adopted by more companies, future financing of AI infrastructure may gradually shift from "technology companies buying equipment themselves" to a structure in which "financial investors hold the assets and technology companies lease them for the long term."