Amazon is exploring an unusual way to finance its massive AI infrastructure buildout: sell billions of dollars worth of Nvidia chips to investors, then lease them back.
According to people familiar with the matter, Amazon is discussing a deal that could involve around $8bn of Nvidia’s advanced Grace Blackwell chips.
The idea is straightforward. Amazon would transfer thousands of chips into a special-purpose vehicle, or SPV, funded by outside investors. Amazon would then lease those chips back and continue using them in its data centres.
On the surface, it may look like another financing transaction.
But there is a bigger story here.
The AI infrastructure race is becoming so capital intensive that even one of the world’s largest and strongest technology companies is looking for new ways to fund it.
Why would Amazon sell chips it still needs?
Amazon is not getting rid of the computing power.
It would continue using the Nvidia chips after transferring them to the investment vehicle. The difference is who owns the assets on paper and how the financing is structured.
That matters because buying and deploying advanced AI chips requires enormous amounts of capital.
Amazon is expected to spend around $220bn in capital expenditure this year, with the majority directed towards AWS and the construction and expansion of AI data centres.
That means the company is not just spending on servers.
It is spending on:
- Advanced Nvidia chips
- Data centre construction
- Networking equipment
- Power infrastructure
- Cooling systems
- Land and facilities
- The wider infrastructure needed to run AI workloads
The more aggressively Amazon builds out its AI capacity, the more capital gets tied up in physical assets.
The proposed chip transaction is essentially an attempt to make that balance sheet more flexible.
The deal structure is the interesting part
Amazon would place the Nvidia chips into an SPV, or special-purpose vehicle.
Outside investors would provide financing to that vehicle, potentially through debt issuance.
Amazon would then lease the chips back.
This gives Amazon access to the same computing infrastructure without having to keep all of the ownership tied directly to its balance sheet.
The proposed structure could also attract a much wider pool of investors if the vehicle receives an investment-grade credit rating.
Investors are reportedly expecting the entity to benefit from Amazon’s double-A credit profile, which could make the debt attractive to institutions such as:
- Insurance companies
- Pension funds
- Institutional investors
- Other fixed-income investors
Amazon is also considering an equity stake of up to 10% in the vehicle.
The discussions are still ongoing, so the final structure could change.
This is bigger than Amazon
The important takeaway is that Amazon is not operating in isolation.
The entire technology industry is trying to figure out how to finance the enormous cost of the AI buildout.
The first wave of AI investment was largely about buying chips and building data centres.
Now the question is becoming:
Who ultimately owns all these assets, and who is financing them?
That is where structures such as chip-backed financing, leasing arrangements and special-purpose vehicles are becoming increasingly important.
Companies want access to enormous amounts of computing power without putting all of the associated financing burden directly onto their own balance sheets.
And investors are increasingly willing to finance those assets because AI infrastructure has become one of the biggest investment themes in the market.
Nvidia’s chips are becoming financial assets too
There is another interesting angle here.
Nvidia GPUs are no longer simply pieces of hardware sitting inside a data centre.
They are increasingly becoming collateral for financing.
Companies such as CoreWeave have already used access to Nvidia chips to support borrowing.
Nvidia itself has also been getting involved in the financing ecosystem.
In August, the company offered to backstop up to $125bn of debt through a broader $500bn financing platform involving major Wall Street firms.
That tells you something about how quickly the AI infrastructure market is evolving.
The financing model is starting to develop around the chips themselves.
The five-year question
Amazon expects each generation of semiconductors to last at least five years, according to its regulatory filings.
That makes the economics of these transactions particularly interesting.
A chip may be extremely valuable when it is deployed, but technology moves quickly.
Nvidia’s Grace Blackwell generation is among its most advanced today, but the company’s newer Vera Rubin platform is already on the horizon.
That creates an important question for investors:
How much is an AI chip worth several years from now?
That question matters for anyone financing chips.
If the underlying hardware loses value faster than expected, the economics of the financing structure become more complicated.
On the other hand, older chips can continue to be useful for running applications even after newer chips arrive.
So the value of these assets does not necessarily disappear the moment a new generation is launched.
Amazon already has a huge financing requirement
The proposed transaction also comes at a time when Amazon is already tapping debt markets to help fund its investment plans.
In March, the company announced plans to raise around $50bn through corporate bonds, up from an earlier $37bn target because of strong investor demand.
But conditions became less favourable when Amazon sold $25bn of bonds in July.
Investors demanded higher yields for the debt, particularly on longer-dated maturities.
That makes alternative financing structures more interesting.
If Amazon can finance specific infrastructure through an investment vehicle rather than relying entirely on traditional corporate borrowing, it potentially gives the company another source of capital.
The goal is not necessarily to spend less on AI.
It is to find more efficient ways to pay for it.
Why investors should pay attention
For investors, this raises an important distinction.
Amazon’s enormous AI spending is not automatically a bad thing.
AWS needs computing capacity to serve customers building and running AI applications. Amazon also has significant exposure to the broader AI ecosystem through its investments and partnerships.
But spending hundreds of billions of dollars on infrastructure creates a financial question alongside the growth opportunity.
How quickly will those investments generate returns?
The answer will depend on how much demand Amazon can generate through AWS and other AI-related services.
If AI demand continues to grow rapidly, investing heavily in infrastructure could put Amazon in a strong position.
If returns take longer to materialise, the financing burden becomes more important.
That is why transactions like this deserve attention.
They offer a window into how the largest technology companies are managing the financial side of the AI boom.
The balance sheet is becoming part of the AI story
For years, the AI conversation was dominated by one question:
Who has the best chips?
Then it became:
Who has enough data centre capacity?
Now another question is emerging:
Who can finance all of it efficiently?
Amazon’s proposed $8bn Nvidia chip transaction sits directly in that shift.
The company wants to keep deploying massive amounts of computing infrastructure while finding ways to avoid putting every dollar of that infrastructure investment directly onto its balance sheet.
That is a very different kind of AI competition.
It is no longer just about technology.
It is about capital, financing, asset values and returns on investment.
The bigger message for the AI trade
The AI boom is creating an enormous infrastructure cycle.
Nvidia sells the chips.
Cloud providers buy them.
Data centre operators deploy them.
Banks and investors finance them.
And companies are increasingly finding creative ways to turn those physical assets into financing opportunities.
Amazon’s proposed transaction shows just how interconnected that ecosystem has become.
The AI race is expensive. The next phase may be about figuring out who ultimately carries the cost.
For investors, that could become just as important as tracking chip demand, data centre capacity or AI revenue growth.
Because at some point, the market will want to see not just how much companies are spending on AI, but how efficiently that spending turns into cash flow.