This page is a work in progress
Published 1 July 2023, last updated 3 July 2023
GPU prices are affected by factors of demand such as the popularity of PC gaming, the profitability of cryptomining, and AI development, factors of supply such as manufacturing limitations, and government regulations.
The term “Graphics Processing Unit(GPU)” was first used in 1994 by Sony, in reference to the 32-bit GPU in the Playstation 1.1) NVIDIA's gaming division has been the company's largest source of revenue until it dipped in 2023 and was surpassed by data center revenue.2) The number of PC gamers around the world grew slowly but consistently between 2008 and 2019. In 2020, the number of PC gamers increased from 1.5 billion to 1.75 billion, likely because consumers were spending more time at home during the COVID-19 pandemic.3)
GPUs can be used to mine cryptocurrency, such as Bitcoin and (formerly) Ethereum. This use of GPUs first had a noticeable effect on GPU demand during the cryptocurrency bubble in late 2017 and early 2018, when street prices for GPUs rose to as much as double their MSRP. GPU prices remained closely correlated with Ethereum prices, falling after the bubble but not back to their MSRP. In 2021, another much larger crypto bubble caused GPU prices to reach quadruple their MSRP.4) Analysts at Jon Peddie Research estimated that 25% of GPUs shipped in the first quarter of 2021 went to cryptocurrency miners and speculators.5) In 2022, cryptocurrency prices crashed, causing GPUs to return to their MSRP.6) Also in 2022, Ethereum changed to a 'proof-of-stake' model that made mining with GPUs significantly less profitable.7)
The rise of AI development can be loosely tracked by NVIDIA's data center revenue, which saw a large jump in 2017 8) and continued to grow until surpassing the company's gaming revenue in 2023.9). Additionally, some of the revenue (as much as 10%, according to analysts in 201710))reported as gaming revenue by the company is likely to actually be GPUs purchased for AI development.
In 2017, a bank in Russia claims to have singlehandedly caused a GPU shortage in the country by buying them for AI development.11)
In 2023, demand for GPUs for AI development appears to be at an all-time high. Twitter, Oracle, Microsoft, and other companies are purchasing “tens of thousands” of GPUs.12)
The price-performance of GPUs has increased over time due to advances in manufacturing and chip design.13)
It takes about three months to manufacture a GPU14). It takes about 3-5 years and $10-20 billion to construct a new fabrication plant capable of producing them.15) Because of the complexity and time involved with GPU manufacture, the supply of GPUs is highly inelastic in the short-run, meaning that supply cannot quickly adjust to changes in price as it can with some other goods.
The short-run supply inelasticity described above also incentivizes GPU manufacturers to make predictions about how demand will change in the future, or at least over the 3-5 years that it takes to scale up production, NVIDIA seems to have been betting on a big increase in demand for AI development for several years. In 2016, the CEO said that the company was “going all in” on AI chips.16) In 2023, the company has ordered an additional 10,000 chips from TSMC to meet AI demand that they predict will continue through the year.17)
Government policy might affect the prices of GPUs in various ways. For example, the CHIPS Act in the US allocated $50 billion USD for American semiconductor manufacturing.18) Also, the US Commerce Department banned exports of two high-performance GPUs to China, and is considering expanding these restrictions as of July 2023.19)