GoKawiilStudy finds older NVIDIA GPUs like A100 retain long-term earning power via open-weight AI models
A new Ornn Data paper argues that NVIDIA's older Ampere-generation GPUs, particularly the A100, remain economically useful far longer than assumed because open-weight models can be run cheaply on them. The analysis found that self-hosted open-weight models like gpt-oss-120b produce output more cheaply on A100 chips than on newer H100 hardware, and that A100 rental prices have stayed unusually stable over a five-year contract period compared to Hopper and Blackwell chips.
GoKawiil's interpretation of the reporting above, not reported fact.
The findings push back against the common assumption that each new GPU generation instantly devalues the previous one, suggesting depreciation schedules for AI infrastructure may be too aggressive. If price-elastic, latency-tolerant workloads like agents and batch evaluation keep routing to older, cheaper hardware, GPU rental markets and capital expenditure forecasts across the AI industry could need rethinking.
- Self-hosted open-weight models can cost one-fifth or less than comparable closed models per task.
- On certain workloads, older A100 GPUs beat newer H100 chips on cost-per-output-token.
- Rental market data shows A100 pricing holding value far better than Hopper or Blackwell chips over five-year terms.
NVIDIA RTX 4090 GPU — For readers interested in the economics of running open-weight models, an RTX 4090 offers substantial VRAM and compute for local inference experiments without renting datacenter GPUs. It's a practical way to explore self-hosting sparse open-weight models like those discussed in the article, right from a home workstation.
See NVIDIA RTX 4090 GPU on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: data.ornn.com, 2026-09-22
Published there as: “The Economics of Open-Weight Inference”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.