a16z Finds GPU Rental Prices Rising

Unlike the depreciation principle that asset values fall over time, rental prices for all five GPUs were above their January levels in October on an index that set January 1 this year to 100. US venture capital firm Andreessen Horowitz (a16z) published its analysis of an index from GPU pricing data provider Ornn. H200 had the largest increase, at about 2.6 times, followed by B200, RTX PRO 6000 and H100. A100, in its sixth year since launch, also remained above its January price.

Compared with a straight-line depreciation curve based on the six-year useful life major hyperscalers assign to servers, prices for the five GPUs were 9% to 143% above the benchmark. Older models A100 and H100 spent more of the observed period below the curve, but their prices surged after April this year and crossed above it.

Growing demand for AI tokens was cited as a major factor behind the price increases. OpenRouter, an AI model marketplace, reported that weekly token use climbed from about 4.9 trillion in late October last year to about 167.7 trillion early this month, an increase of more than 30 times. Use of both open-source and closed models rose sharply, while the spread of AI agents also increased demand for older GPUs.

The Ornn index is based on real-time spot transactions, so it may differ from prices disclosed by hyperscalers and GPU cloud providers or from long-term contract rates. Future rental prices may depend on supplies of newer chips and high-bandwidth memory (HBM), data-center construction speed, and competition from alternative chips such as Google’s tensor processing units (TPUs). A growing supply of newer GPUs could accelerate price declines for older models, while continued HBM shortages could help support their prices, the analysis said. These trends may also affect depreciation standards, including GPU useful life and residual value, and judgments about long-term value.

Unlike the depreciation principle that asset values fall over time, rental prices for all five GPUs were above their January levels in October on an index that set January 1 this year to 100. US venture capital firm Andreessen Horowitz (a16z) published its analysis of an index from GPU pricing data provider Ornn. H200 had the largest increase, at about 2.6 times, followed by B200, RTX PRO 6000 and H100. A100, in its sixth year since launch, also remained above its January price.

Compared with a straight-line depreciation curve based on the six-year useful life major hyperscalers assign to servers, prices for the five GPUs were 9% to 143% above the benchmark. Older models A100 and H100 spent more of the observed period below the curve, but their prices surged after April this year and crossed above it.

Growing demand for AI tokens was cited as a major factor behind the price increases. OpenRouter, an AI model marketplace, reported that weekly token use climbed from about 4.9 trillion in late October last year to about 167.7 trillion early this month, an increase of more than 30 times. Use of both open-source and closed models rose sharply, while the spread of AI agents also increased demand for older GPUs.

The Ornn index is based on real-time spot transactions, so it may differ from prices disclosed by hyperscalers and GPU cloud providers or from long-term contract rates. Future rental prices may depend on supplies of newer chips and high-bandwidth memory (HBM), data-center construction speed, and competition from alternative chips such as Google’s tensor processing units (TPUs). A growing supply of newer GPUs could accelerate price declines for older models, while continued HBM shortages could help support their prices, the analysis said. These trends may also affect depreciation standards, including GPU useful life and residual value, and judgments about long-term value.