Amazon pursues $8bn Nvidia chip offload as AI infrastructure costs soar

Amazon Nvidia Chips Move to Special Purpose Vehicle

Amazon is pursuing a deal to offload $8bn of Nvidia chips to external investors through a special-purpose vehicle, marking a strategic shift toward an asset-light strategy as AI infrastructure financing demands escalate1

1

. The cloud giant has held talks with investors in recent weeks to gauge interest in transferring thousands of Nvidia Grace Blackwell AI chips already deployed across more than a dozen data centers in five U.S. states, including Nevada and Virginia1

1

5

5

. Under the proposed sale-leaseback arrangement, Amazon would lease the advanced semiconductors back from the SPV, which would raise capital by issuing debt to outside investors1

1

. Amazon plans to offer an equity stake of up to 10% in the vehicle while retaining no ownership interest itself1

1

. The discussions remain ongoing and subject to change.

Finance Massive AI Infrastructure Spending Through Creative Structures

The move reflects mounting pressure on hyperscalers to finance massive AI infrastructure spending while preserving creditworthiness and managing balance sheet risks2

2

. Amazon expects to spend $220 billion in capital expenditure in 2026, up from $131 billion in 2025, with the majority allocated toward AWS for purchasing advanced chips and building AI data centers2

2

5

5

. CEO Andy Jassy stated in July that despite raising the spending target by $20 billion, Amazon would likely remain unable to meet all customer demand through 2027, with strong demand already visible for 20285

5

. Tech companies are pursuing various strategies to move debt off their balance sheets, including residual value guarantees that provide lenders with assurances about the future value of chips or data centers without directly borrowing cash1

1

. Nine of the largest U.S. tech companies had about $3 trillion in off-balance sheet commitments as of August, indicating spending plans may be larger than publicly disclosed3

3

.

Remove From Its Balance Sheet While Maintaining Access

Investors expect the entity will receive an investment-grade credit rating based on Amazon's current double-A rating, paving the way for institutional investors such as insurance companies and pension funds to participate in the deal1

1

5

5

. The proposed structure would let Amazon remove the depreciating assets from its balance sheet while continuing to use the chips through the leaseback agreement3

3

. Amazon has been tapping multiple financing tools to fund AI infrastructure expansion: it entered into a $17.5 billion term loan in June and raised £4.25 billion through a debut sterling bond offering, making it the largest bond issuer among major hyperscalers this year with cumulative sales equivalent to nearly $100 billion2

2

. In March, Amazon laid out plans to raise about $50 billion through corporate bond issuances, increasing the offering from $37 billion following strong demand, though it faced weakening interest in long-dated debt when it sold $25 billion in bonds in July1

1

.

GPU-Backed Bonds Emerge as AI Infrastructure Financing Trend

The deal represents part of a growing trend of using Nvidia GPUs as financial assets, with GPU-backed bonds becoming another avenue in the industry4

4

. Financing backed by graphics processing units has proven popular among companies such as CoreWeave, which have used their ready access to chips to fund borrowing1

1

. Nvidia itself entered the GPU financing arena in August by offering to backstop up to $125 billion of such debt through a $500 billion financing platform with major Wall Street groups, though recent reports suggest the initiative has made Wall Street wary of the firm's plans1

1

4

4

. Depreciation concerns remain a key debate factor surrounding Nvidia GPUs, as residual values of advanced AI chips have proven higher than initially expected, meaning hyperscalers need more financing to cover their investments4

4

.

EC2 Capacity Blocks Price Increase Signals Demand Pressures

Separately, Amazon announced a roughly 15% price increase on its EC2 Capacity Blocks for ML service, which lets customers reserve access to Nvidia chips ranging from the older A100 to the newer B300, with updated rates taking effect next week2

2

. The pricing page notes that current rates are scheduled to be updated next in October 2026, with pricing consisting of an upfront reservation fee plus an operating system fee2

2

. For example, the p5.48xlarge instance—which runs eight Nvidia H100 chips—is listed at $41.528 per hour in major U.S. regions2

2

. The two moves together reflect the pressure Amazon faces in financing its AI infrastructure expansion while managing costs and maintaining competitiveness in the cloud computing market. KeyBanc Capital Markets argued that Amazon's heavy AI infrastructure spending could pressure near-term margins but strengthen its AI moat and support durable AWS earnings growth through the end of the decade5

5

. AWS Chief AI and Technology Officer Matt Wood stated on Thursday that he expects lower AI costs and fewer technical barriers as computing capacity expands and efficiency improves5

5

.