Amazon is in discussions to shift roughly $8 billion worth of Nvidia's high-end Grace Blackwell AI chips into a special-purpose vehicle (SPV) and lease them back, according to a report from the Financial Times on Friday. The arrangement, if completed, would let Amazon tap into the computing power it needs for its US data centers without carrying the full cost of the hardware on its own books.
Special-purpose vehicles are separate legal entities created for a specific financial purpose. In this case, the SPV would buy the chips and raise money from outside investors, mainly by issuing debt. Amazon would then use the equipment through lease contracts, paying for it over time rather than upfront.
The FT said Amazon may offer outside investors up to a 10% equity stake in the SPV, a structure that could help attract capital while Amazon keeps control of the rollout. The move is part of a broader trend among big tech companies looking for creative ways to fund the massive spending required for artificial intelligence infrastructure.
Why Amazon would do this
AI chips like Nvidia's Grace Blackwell are among the most expensive components in modern data centers. A single server rack can cost hundreds of thousands of dollars, and building out new facilities to house them runs into the billions. For a company like Amazon, which is already spending heavily on cloud and AI capacity, adding $8 billion in new assets to its balance sheet would increase reported assets and could weigh on key financial ratios.
By using an SPV and leasing the chips back, Amazon can keep some of that spending off its balance sheet. Reported assets would rise more slowly, and profitability ratios that compare earnings to the asset base—like return on assets—could look stronger than they would under outright ownership. That can matter to investors who watch how efficiently a company uses its capital.
But the economic commitment doesn't disappear. Lease obligations are still real claims on future cash flow, and credit analysts treat them as debt-like liabilities when assessing a company's financial health. So while the structure may improve certain accounting metrics, it doesn't change the underlying cost of building and running AI infrastructure.
A new corner of the debt market
For investors, the deal could also create a new type of asset-backed security. Bonds backed by Amazon's lease payments and collateralized by high-end AI chips would tie their performance to two things: Amazon's creditworthiness and the resale value of the hardware over time. If the chips retain their value and Amazon keeps paying rent, the bonds could be relatively safe. But if AI technology advances quickly and older chips lose value, the collateral could weaken.
This isn't the first time Nvidia's chips have been floated as collateral. Wall Street has shown skepticism about using GPUs as loan collateral, partly because of how fast the technology evolves. Still, the idea is gaining traction as AI spending balloons and companies look for ways to finance it without stretching their balance sheets too thin.
The broader context is a surge in AI-related capital spending across the tech sector. Anthropic's IPO filing revealed a $42 billion loan from Broadcom for AI compute, and other firms are raising large sums to build data centers. Data center companies are going public at multi-billion-dollar valuations, reflecting the scale of demand.
What it means for investors
For everyday investors, the key takeaway is that Amazon's move is a financing tactic, not a change in its business strategy. The company still needs the chips to power its cloud services and AI products. The question is how it pays for them and how that shows up in its financial statements.
If the deal goes through, investors may see Amazon's reported capital expenditures look smaller than they otherwise would be. That could make its free cash flow look healthier in the short term. But lease payments will still eat into cash flow over time, and analysts will adjust their models to account for the off-balance-sheet obligations.
There's also a potential ripple effect. If Amazon successfully uses an SPV to finance AI hardware, other companies may follow suit. That could create a new asset class for institutional investors and change how the market values AI infrastructure spending. Retail investors have already shown strong interest in Nvidia and AI-related stocks, and this kind of financial engineering could add another layer of complexity to how those companies are valued.
For now, the deal is still in discussion, and there's no guarantee it will close. But it highlights a growing tension in the AI boom: the technology is incredibly expensive, and even the biggest companies are looking for ways to manage the cost. Whether through leases, SPVs, or other structures, the way AI infrastructure is financed will be a key story for markets in the coming years.
