Nvidia’s older GPUs continue to generate rental income, signalling durable demand for legacy hardware in the AI compute market
Executive summary: Nvidia's older GPU models continue to generate rental income, indicating sustained demand for legacy hardware despite newer product releases. This shows slower depreciation of GPUs, affecting revenue forecasts for cloud providers like CoreWeave and influencing capital allocation decisions.
Who is involved: Nvidia, CoreWeave, cloud customers, investors.
Likely next: Nvidia may continue monetizing its legacy GPU inventory while monitoring demand for newer architectures and any shifts in AI spending patterns.
The continued rental income from Nvidia’s previous‑generation GPUs shows that legacy hardware retains value in the AI compute market longer than many anticipated. This persistence suggests that depreciation curves for these chips are flatter than expected, providing a steady cash flow for owners of the equipment. For cloud‑focused firms such as CoreWeave, the ability to earn from older models lessens the immediate pressure to replace every server with the newest architecture as soon as it launches. From a business perspective, the durable demand for older GPUs can smooth capital‑expenditure planning and create a more predictable baseline of revenue amid the rapid rollout of newer generations. It also introduces a competitive nuance: providers that can mix legacy and cutting‑edge hardware may offer tiered pricing or flexibility that appeals to cost‑sensitive AI workloads. While newer chips will still drive performance gains, the sustained utility of older models indicates that the AI hardware ecosystem will likely support a mixed‑generation approach for the foreseeable future.
What's next — scenarios
Legacy Resilience (Base Case) (55%)
Cloud providers maintain higher profit margins by extending server lifecycle depreciation schedules by 12 to 18 months.
- Secondary market pricing for H100 and A100 GPUs remains stable over the next quarter
- Cloud providers announce extended depreciation timelines in upcoming earnings reports
Rapid Obsolescence (Downside Case) (25%)
Surplus older hardware suddenly floods the resale market, forcing asset write-downs for specialized cloud providers.
- Major enterprise customers cancel legacy GPU rental contracts in favor of Blackwell-class clusters
- Secondary market lease rates drop by more than 20% within a 30-day window
Hybrid Compute Equilibrium (Upside Case) (20%)
Software optimization layers successfully abstract hardware differences, cementing a permanent tier-based pricing model for AI compute.
- Major AI framework releases explicitly optimize inference workloads for older GPU architectures
- CoreWeave or similar neo-cloud providers report record utilization rates on previous-generation fleets
What to watch
- CoreWeave and secondary cloud provider Q3/Q4 capital expenditure guidance on hardware upgrades
- Spot rental prices for A100 and H100 GPU instances on major broker marketplaces over the next 60 days
- Nvidia management commentary regarding legacy chip demand durability during the next earnings call
- Enterprise adoption rates of Blackwell architecture versus continued reliance on Hopper-generation chips
Timeline
- — NVIDIA’s Older GPUs Still Earn Rent. What That Does and Doesn’t Prove for CoreWeave (Yahoo Finance)
Key entities
Sources
Related cases
- Nvidia's 2 GW AI infrastructure push in Australia strengthens its lead in AI chip supply and data‑center capacity
- A Yahoo Finance article argues that an overlooked, low‑priced Magnificent Seven stock could become the biggest winner in the trillion‑dollar AI race
- Nvidia eyes massive $10 billion stake in Anthropic's landmark $2.3 trillion IPO
- Nvidia may take a multi‑billion‑dollar stake in Anthropic’s upcoming IPO
- Nvidia partners with Australian data center providers to deliver up to 2 GW of AI infrastructure by 2027
- Nvidia’s inaugural year‑ahead forecast signals a trajectory that could overtake Apple and Alphabet