Mumbai: For decades, buying technology meant owning the machine and watching it depreciate. AI is making that rather simple idea increasingly inconvenient.
Amazon is reportedly exploring a transaction involving roughly $8 billion worth of Nvidia Grace Blackwell chips already deployed across more than a dozen US data centres. Under the proposed structure, the chips could move into a special-purpose vehicle, which would raise financing from investors, while Amazon would lease the hardware back and continue using it.
In other words, Amazon could stop treating some of its AI hardware as something it simply owns and start treating it more like infrastructure.
The machine stays in the building. The economics move somewhere else. Finance has entered the server rack.
The New AI Question Is Who Owns The Compute
The proposed Amazon structure arrives at a revealing moment. The AI infrastructure race has become extraordinarily capital intensive, with hyperscalers spending tens of billions of dollars on servers, networking equipment and data centres before the resulting capacity necessarily produces an equivalent return.
Amazon itself has already made the scale of that commitment clear.
The company expects to invest approximately $200 billion in capital expenditure during 2026, with a substantial portion going toward AWS and AI-related infrastructure. During the first half of 2026, Amazon reported $96.3 billion in cash capital expenditure, compared with $55.6 billion during the same period a year earlier.
The proposed $8 billion chip transaction therefore represents roughly 8.3% of Amazon’s first-half 2026 capital expenditure. It is not an additional $8 billion investment by Amazon; rather, it is a reported attempt to change how part of that infrastructure is financed.
That distinction matters.
Nvidia Is Already Turning Compute Into An Asset Class
Amazon’s reported move does not exist in isolation.
In August, Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR designed to mobilise more than $500 billion of third-party capital over time for AI infrastructure.
Nvidia has been unusually explicit about the philosophy behind the strategy: computing capacity can be treated as productive infrastructure because it generates revenue, can serve multiple customers and can potentially be redeployed.
That is a significant conceptual shift.
The AI industry once sold chips.
Now it is beginning to sell the financial logic surrounding those chips.
And suddenly a GPU has acquired something resembling a mortgage.
Why Amazon Could Like The Structure
For Amazon, leasing hardware through a separate vehicle could provide several advantages without requiring the company to abandon the infrastructure itself.
The potential benefits include:
- Less capital tied up directly in physical AI hardware.
- Greater flexibility in financing rapid infrastructure expansion.
- Potential access to institutional investors looking for long-duration infrastructure assets.
- A structure that could allow Amazon to keep operating equipment while shifting some financing obligations elsewhere.
Amazon has also said its 2026 spending is supported by substantial customer commitments, particularly within AWS. That gives the company an important economic argument: AI infrastructure is not being built entirely on speculation.
There is already demand waiting for the machines.
But Moving An Asset Does Not Make The Risk Disappear
This is where the financial engineering becomes more interesting — and slightly less glamorous.
The Bank for International Settlements has warned that AI infrastructure financing is increasingly moving toward debt, private credit and structures involving special-purpose entities. Such arrangements can shift borrowing away from a hyperscaler’s conventional balance sheet while leaving it with long-term lease, capacity or guarantee obligations.
The BIS has described some of these structures as a form of “shadow borrowing”, because economically they can resemble debt even when the financing sits elsewhere.
That does not make the model inherently dangerous. It does, however, make the real level of leverage harder to understand.
And GPUs are not buildings.
A data centre can remain useful for decades. AI accelerators can become technologically outdated much faster, particularly when every new generation promises better performance and efficiency. An investor financing today’s hardware therefore has to care about what that hardware will be worth tomorrow.
Technology moves considerably faster than accounting departments enjoy.
The Bigger Story Is A Financial Infrastructure Race
Amazon‘s reported $8 billion transaction could ultimately prove to be simply another financing arrangement.
But the direction is important.
AI companies are increasingly discovering that building compute capacity requires not just engineers, chip suppliers and electricity, but banks, private-credit funds, infrastructure investors and carefully constructed ownership structures.
That creates a new ecosystem around AI:
chips → data centres → leases → debt → institutional capital → computing revenue.
The positive case is straightforward. Financial markets could help AI infrastructure expand faster without forcing every company to fund enormous hardware purchases entirely from its own balance sheet.
The less comfortable question is what happens if AI revenues fail to grow quickly enough to support all that infrastructure.
The BIS has already noted that AI-sector financing is becoming more interconnected, including investment relationships between companies that also have commercial relationships with one another.
That makes the industry’s financial plumbing worth watching almost as closely as its GPUs.
Because the next AI bubble, if one ever forms, may not look like a warehouse full of useless machines.
It could look perfectly modern, perfectly productive, and impeccably financed.
Which, admittedly, is how financial problems usually prefer to dress.
Read More: 2026 AI Glasses Are Quietly Redrawing The Odd Privacy Line
