NVIDIA B200 GPUs Trade at 58% Premium Over Launch Price

NVIDIA's B200 GPUs are currently trading at a residual value that is 158 percent of their original launch price. This represents a 58 percent premium nearly a year after the hardware became widely available, according to new data from Silicon Data.

The analytics firm reports that the residual values for NVIDIA's older A100 and H100 GPUs also remain significantly higher than standard depreciation schedules would predict. In a typical straight-line depreciation model, an asset's value declines evenly over its estimated useful life. The current market prices defy those expectations.

Defying Standard Depreciation

Silicon Data's analysis shows the AI GPU market operates outside normal financial models. For equipment, residual value is calculated by subtracting cumulative depreciation from the original cost. The firm states that NVIDIA's current-generation and previous-generation GPUs are commanding values "well in excess of what 3 or 5-year straight-line schedules would imply."

This sustained high value is largely driven by ongoing scarcity and relentless demand. The B200 GPU remains highly sought after for running large language models and other AI workloads.

Efficiency Drives Demand

A key factor behind the B200's strong residual value is its efficiency for inference tasks. According to estimates from SemiAnalysis, the GPU can process AI inferences at a cost of just $0.20 per 1 million tokens. This performance is measured at a rate of 76 tokens per second per user on the DeepSeek R1 model.

Such efficiency makes the B200 a cost-effective choice for companies deploying AI applications at scale. This practical utility underpins the sustained market demand that keeps prices raise.

Rental Market Reflects Value

The premium is also visible in the GPU rental market. Silicon Data compiled hourly rental rates that show a similar appreciation. In January, B200 GPUs rented for just under $5 per hour. By August, those rates had climbed to between $5.50 and $5.80 per hour.

This represents a price increase of 50 to 80 percent over an eight-month period. The rental market data reinforces the trend of rising value for NVIDIA's flagship AI hardware.

The report from Silicon Data uses GPU residual values as a barometer for the health of the AI industry. Despite calls from some AI leaders to pace development, the data suggests strong demand continues. The B200's market performance remains strong as the AI supercycle persists.