[e.g., "One asset, two economies: training vs inference on a GPU cluster"]
[one line: the same fleet underwritten two ways — capex-heavy training vs throughput-driven inference]
![[e.g., "One asset, two economies: training vs inference on a GPU cluster"]](/__l5e/assets-v1/530d3cbc-7386-4174-8bdd-ec351c3fc2ce/gpu-economics-hero-v2.png)
Project Snapshot
- Asset
- [GPU cluster, e.g., 10k accelerators]
- Mode
- [Training / Inference toggle]
- Capex
- [$/GPU]
- Power
- [MW, PUE]
- Useful life
- [obsolescence assumption]
The Challenge
[The same fleet has two different economic profiles — training is capex- and utilization-driven with obsolescence risk; inference is throughput- and cost-per-token-driven. How do you underwrite an asset whose returns depend on its use case?]
The Approach
- 01
[Build the shared cost base: capex, power, PUE, depreciation/obsolescence]
- 02
[Training case: cluster utilization, time-to-train, cost per training run]
- 03
[Inference case: throughput, cost per token, utilization, opex]
- 04
[Add a toggle/scenario switch]
- 05
[Sensitize to the GPU obsolescence curve]
Inside the Model
[Describe the toggle switching training/inference outputs.]
Results at a Glance
What This Demonstrates
- GPU/accelerator economics
- Capex vs opex underwriting
- Scenario/toggle modeling
- Depreciation & obsolescence handling
Key Takeaway
[Idea + mechanism, two sentences.]
Illustrative case built on representative data; not based on any confidential or client transaction.