Amazon is in exploratory talks to move roughly $8 billion worth of Nvidia Grace Blackwell AI chips into a special-purpose vehicle (SPV), a structure that would shift the hardware off its balance sheet while keeping it operational. The chips are already installed in more than a dozen U.S. data centers across at least five states, including Nevada and Virginia. Outside investors would supply most of the SPV's funding, primarily as debt, and could receive a 10% equity stake in the venture.
The arrangement would let Amazon continue deploying the chips for its cloud services while reducing upfront capital outlay. For a company spending tens of billions on AI infrastructure, the structure eases near-term spending pressure without slowing capacity growth. The talks remain exploratory, no terms are final, and neither Amazon nor Nvidia has publicly commented.
How the SPV structure would work
Under the proposed deal, Amazon would transfer ownership of the Nvidia hardware to the SPV and then lease the chips back. Investors would fund the vehicle mostly through debt, with a 10% equity slice as their upside. This is not a sale in the traditional sense - Amazon retains operational control of the chips and continues using them to power AWS workloads.
SPVs are common in real estate and aircraft leasing, but applying the model to AI chips signals how capital-intensive the infrastructure buildout has become. The chips are already physically deployed, so this is a financing maneuver, not a capacity expansion.
The broader capital picture
The talks surface as Nvidia separately seeks Wall Street partnerships to finance the AI boom, with a reported $500 billion capital target. Nvidia also authorized a $150 billion share buyback, the largest increase in its history. Amazon, meanwhile, continues buying Nvidia chips aggressively - the reported SPV talks do not signal a pullback in procurement.
For finance leaders tracking AI spend, structures like this blur the line between capex and opex. Moving $8 billion in hardware to an SPV funded mainly by debt changes the balance-sheet optics without altering the physical reality of who runs the servers.
Why this matters for finance and operations professionals
The SPV model, if it goes forward, offers a template for financing large-scale AI infrastructure without carrying the full asset weight on the corporate balance sheet. CFOs and VPs of finance evaluating their own AI hardware commitments can watch this deal for signals on how lenders price chip-backed debt and what equity terms investors demand. Professionals building the business case for AI infrastructure may find relevant frameworks in AI Finance Leadership Courses or AI for CFO Training.
Your membership also unlocks:
