Jensen Huang and Elon Musk: GPU Useful Life Far Exceeds Financial Depreciation Assumptions
nashnova research
Nvidia's Jensen Huang and Tesla's Elon Musk have reached a rare consensus: GPU useful life is more than double the standard depreciation assumption — and secondary-market data backs them up, with direct implications for cloud providers' margins and financing.
What is Musk's argument — why does the same GPU grow more valuable over time?
At SpaceX's Q3 2026 earnings call, Musk argued that as underlying AI models improve, the same GPU hardware delivers ever-higher "effective compute per watt."
This means → a GPU's commercial value hinges not on hardware age, but on how much useful intelligence each watt of power produces — better models make older cards more valuable.
In plain terms = think of a truck whose fuel cost stays flat, but whose cargo value per mile keeps rising because the roads improve every year. This also explains why the H100, launched years ago, still leases above its original price.
What specific claim did Huang make?
At a GTC Berlin media session, Huang stated that GPUs built on Blackwell and the upcoming Vera Rubin architecture have an effective useful life of eight-plus years.
This means → that is twice the four-year depreciation cycle most corporate finance models assume — if the depreciation window is wrong, the income statement is wrong.
He offered two supporting reasons: first, Nvidia's programmable software stack lets it continuously optimise performance without a hardware refresh; second, strong secondary-market demand allows GPUs to be redeployed across different workloads.
What does the secondary-market data show?
A new report from Silicon Data confirms that real-world valuations of the A100, H100, and B200 all sit well above the residual values implied by standard three-to-five-year straight-line depreciation.
The numbers: the A100, roughly six years old, still trades at 25% of original cost; the H100, about four years since launch, retains roughly 58%; the B200, shipping for under two years, has appreciated to 158% of its original price.
In plain terms = under textbook depreciation, these cards should be nearly worthless by now — but the market says compute assets hold value far better than the accounting assumes.
If the depreciation assumption changes, how big is the financial impact?
For cloud providers, extending GPU depreciation from four years to eight would cut annual depreciation expense in half, lifting operating margins directly.
This means → Wall Street's worry that massive AI capex "takes too long to pay back" eases significantly when the amortisation window doubles — same investment, half the balance-sheet drag.
This reflects a deeper shift: the market previously assumed AI chips depreciate to near-zero, but financial institutions are starting to recognise that Nvidia compute assets carry real equity value and can serve as collateral for debt financing. The reassessment of GPU depreciation will ultimately reshape cloud providers' capital-allocation decisions and cost of capital.
市场有风险，内容仅供研究参考，不构成投资建议。