Older GPUs Are Still in Demand: What CoreWeave's A100 and H100 Pricing Tells Us

Key Highlights

  • Average H100 pricing in Q4 2025 was within 10% of where it started the year, indicating stable demand.
  • Average A100 pricing actually increased during 2025 as demand for older generation inference infrastructure grew.
  • CoreWeave is signing older generation GPU capacity into new reserved instance contracts ahead of availability.
  • Enterprise and AI native customers are using H100 and A100 infrastructure for proliferating inference use cases.
  • CEO Intrator stated that concerns about demand for older GPU SKUs have been put to rest by current market dynamics.

 

A persistent concern in AI infrastructure investing has been that pricing for older GPU generations, specifically NVIDIA's H100 and A100, would collapse as GB200 Supply increased. CoreWeave's 2025 data challenges that narrative directly.

H100 Pricing: Stable Through 2025

Average H100 pricing in Q4 2025 was within 10% of where it started the year. For a product category widely expected to face pricing pressure as newer GB200 systems came online, this stability is notable. The explanation lies in market segmentation. GB200 systems are being deployed primarily for frontier model Training at hyperscale. H100 systems serve a different and growing market: inference at enterprise and AI native scale.

A100 Pricing Is Rising

Even more striking is the trajectory of A100 pricing, which actually increased during 2025. Management attributed this to demand from inference use cases. A100 GPUs are efficient for inference workloads at lower batch sizes, cost less than H100 systems per unit, and are available in quantities that allow smaller AI native companies to scale inference infrastructure without competing for the most in-demand GB200 capacity. CoreWeave noted that it is signing A100 and H100 infrastructure into new reserved instance contracts ahead of when it becomes available.

The Inference Proliferation Thesis

The underlying driver of older GPU pricing resilience is the proliferation of AI inference workloads across the economy. Every AI application that reaches production deployment creates a sustained, ongoing compute demand for inference. As the number of AI applications grows, the aggregate inference compute requirement grows proportionally. A substantial portion of commercial AI inference is economically served on H100 and A100 hardware.

Implications for CoreWeave's Revenue Mix

Older GPU generations carrying stable or rising prices is positive for CoreWeave's asset utilization Economics. Infrastructure built around H100 and A100 systems, deployed extensively in 2023 and 2024, continues to generate contractual revenue at economics close to original Acquisition assumptions rather than requiring repricing. This supports the Margin visibility on the existing contracted Backlog.

Looking Forward

The H100 and A100 pricing resilience also has implications for how investors should think about GPU Depreciation curves more broadly. If inference demand continues to grow at the rates CoreWeave is observing, older GPU generations may retain value for much longer than traditional hardware depreciation schedules would suggest, improving the long-term economic profile of CoreWeave's existing infrastructure portfolio.

Disclaimer

Disclaimer: This article is for informational purposes only and does not constitute financial advice. Investing in securities involves risk, including possible loss of principal. Past performance is not indicative of future results. Please conduct your own research or consult a licensed Financial Advisor before making Investment decisions.