[Hong Jang-won’s Bull & Bear] “GPUs Are Money and Assets”: NVIDIA’s Defense
An independent exercise, not a statement by the video creator or verification of every claim in the video.
Higher rental rates can still leave less cash for debt service
This is an annual hypothetical, not actual GPU or company results. Assume a USD 30000 acquisition cost, USD 2 hourly rent, 80% utilization and 8760 hours a year: revenue is 2×8760×80%=USD 14016. Subtract USD 0.6 in power and operating costs per rented hour plus USD 1000 annual fixed maintenance, leaving USD 8811.2 before debt service. Even if rent rises 10% to USD 2.2, utilization falling to 50% reduces revenue to USD 9636 and cash under the same cost assumptions to USD 6008. Straight-line depreciation over 6 years with 0 residual value gives USD 5000 annual expense, which is neither annual principal and interest nor resale price. Assuming annual debt service of USD 7000, the first scenario covers it while the second falls short. This explains why actual rented hours, costs and repayment schedules must accompany a rental-rate chart.
Check for yourself
- Is the observed rate an executed contract price or an advertised quote, and are models and contract lengths comparable?
- How do utilization, operating costs and payback timing change alongside higher rents?
- Have forced-sale recoveries been compared separately with debt balances and collateral terms?
Limits of application
Every figure is hypothetical. Costs are simplified, excluding taxes, capital expenditures, idle power, failures and additional fees; resale value is not estimated. Depreciation is not deducted again as cash outflow. Debt-service coverage here is only a comparison of assumed annual cash, not a finding about actual covenant compliance or corporate survival.
- What to look for
- Explore different evidence in the AI investment debate through index weighting and GPU rental economics.
- Why it matters
- Match return periods and definitions, then separate rental revenue, depreciation expense and cash available for debt service to identify missing assumptions.
Timeline
3 segmentsThe video uses a chart of rental rates above a depreciation-style path to argue for durable GPU value. The a16z source reports recent rental strength but also notes older models spent extended periods below the comparison line and that these indicators are imperfect proxies. NVIDIA advances the economic-life thesis as a supplier. Comparing rents with a depreciation line alone cannot establish resale prices or debt-service capacity; actual rented hours and cash flow after costs are needed.