While focus is on massive supercomputers for training next-gen models, the real supply chain constraint will be 'inference' chips—the GPUs needed to run models for billions of users. As adoption goes mainstream, demand for everyday AI use will far outstrip the supply of available hardware.

AI software is improving so rapidly that older hardware, like a three-year-old NVIDIA inference chip, is now more profitable than it was when new. This phenomenon, where software advancements outpace hardware depreciation, is unprecedented and makes existing infrastructure increasingly valuable.

The current AI moment is unique because demand outstrips supply so dramatically that even previous-generation chips and models remain valuable. They are perfectly suited for running smaller models for simpler, high-volume applications like voice transcription, creating a broad-based boom across the entire hardware and model stack.

Despite the rollout of Blackwell and the announcement of Vera Rubin, cloud provider Nebius reports that pricing for older Hopper GPUs is not dropping, and in some cases is rising. This shows a persistent market for "good enough" compute for specific workloads.

Contrary to typical hardware depreciation, GPUs like NVIDIA's H100 are becoming more valuable over time. This is because newer, more efficient AI models can generate significantly more output and value on the same hardware, tying the GPU's worth to its utility rather than its age.

Andreessen highlights a unique economic phenomenon: the pace of AI software improvement outstrips hardware depreciation. This means a three-year-old NVIDIA inference chip can generate more revenue today than when it was new, a complete reversal of typical tech hardware value cycles.

Contrary to the belief that AI hardware becomes obsolete quickly, older GPUs like A100s will have a long depreciable life. As companies optimize costs, they'll use model routing to send simple queries to older, cheaper hardware, extending its utility for six to eight years.

Countering the narrative of rapid burnout, CoreWeave cites historical data showing a nearly 10-year service life for older NVIDIA GPUs (K80) in major clouds. Older chips remain valuable for less intensive tasks, creating a tiered system where new chips handle frontier models and older ones serve established workloads.

Despite reports of falling H100 spot rental prices, contract prices for sustained GPU workloads are rising. This indicates the market is shifting from short-term, experimental use to long-term, committed production deployments, reflecting stronger, not weaker, underlying demand for AI infrastructure.