a16z Report: NVIDIA H200 Rental Prices Up 143% This Year
A latest report from venture capital firm a16z shows that as of October 2026, the rental price of NVIDIA H200 GPUs had surged 143% compared with the start of the year, the largest increase across its entire product line; the B200 released in 2024 rose 111%, and the RTX 6000 Pro rose 73%. The report says the price increases were driven by a surge in global AI token consumption, but it does not specify the exact source of demand, nor does it address supply-side factors such as data center capacity and power constraints. The article also mentions that the Tesla V100, nearing its 10th anniversary of release, has not depreciated along the traditional depreciation curve; a four-card configuration paired with NVLink can outperform the M5 Ultra and DGX Spark by up to 230%, and high-memory versions command a premium of about $200 in the second-hand market.
Why read this
AI score 60
The a16z report shows H200 rental prices rose 143% this year, and the article also notes that the report does not explain the source of demand or supply constraints.
Topics
Body · AI translation
On October 26, news emerged that, driven by the surge in global AI token consumption, the Nvidia data center graphics card rental market has seen a rare price inversion, with multiple high-performance graphics cards having effectively become appreciating assets.
The latest report released by venture capital firm a16z shows that, as of October 2026, the rental price of Nvidia's H200 graphics card had soared 143% compared with the beginning of the year, the largest increase across its entire product line.
The B200 graphics card released in 2024 followed closely behind, with rental rates rising 111%, and even the RTX 6000 Pro, positioned as a professional workstation card, recorded a 73% increase.
Market monitoring data shows that this round of price increases did not break out suddenly; most GPU rental prices fell to a low point in the first quarter of 2026, with H100 and B200 at one point about 23% lower than their levels at the beginning of the year, before rebounding sharply in April and May.
H200 in particular climbed to a peak in late May, pulled back for a few weeks in the middle, and then began rising again in October.
However, although the report mentions the rapid growth of AI token consumption, it does not explain the specific source of demand, such as whether it comes from agent workloads or the scaled deployment of enterprise applications, nor do the charts address supply-side factors, including data center capacity and power constraints and other key variables that may affect rental prices.
It is worth noting that even older GPUs approaching the 10th anniversary of their release have not depreciated along the traditional hardware depreciation curve; instead, they have found a new role in mid-range LLM operation scenarios.
Taking the Tesla V100 as an example, when paired with NVLink board-level interconnect, the performance of a four-card combination can exceed that of the M5 Ultra and DGX Spark by as much as 230%, which has directly driven high-memory versions to command a premium of about $200 (about 1,440 yuan) in the second-hand market.