Gamers have spent months battling high prices and empty shelves for graphics cards. A new wave of reports and analysis suggests the shortage may not be as simple as demand outstripping supply.
The accusation, led by tech critic Ed Zitron and picked up by outlets including RespawnFirst, is that Nvidia has pushed its manufacturing capacity towards enterprise AI hardware that its biggest customers can’t actually use yet. If that’s true, gamers have been squeezed out to make room for chips that are sitting in warehouses.
The “dark GPU” problem
Much of the claim comes from a Business Insider report. Citing internal Amazon emails from August, it says Amazon set up an “Early Rack Receipt” program, taking delivery of around $5 billion worth of Nvidia GB300 server racks before its data centres had the power to run them. Amazon reportedly expected to take on about $3 billion more each month.
Several AWS sites apparently lack the utility capacity to switch the servers on, so some of the hardware is sitting idle in warehouses. Business Insider calls these “dark GPUs”. Zitron’s analysis also points to Microsoft CEO Satya Nadella admitting he had chips “sitting in inventory” that he couldn’t plug in.
What Nvidia is accused of
Critics argue that Nvidia has sent most of its TSMC wafer supply to data centre chips such as the H100, B200 and GB300, at the expense of GeForce gaming cards. They also claim it has encouraged giants like Amazon, Google, Microsoft and Meta to order hardware far ahead of when they need it. Those orders lift Nvidia’s revenue figures, while buyers take on the storage, power and depreciation costs.
That fits with earlier reports that Nvidia has scaled back gaming GPU supply, including claims it paused RTX 5090 production to focus on AI and that no new GeForce cards are expected for some time.
The other side
Nothing here has been proven. The claims are largely analysts’ readings of the available information rather than documented wrongdoing. Amazon has defended the practice, describing staging hardware early as a routine “best business practice” so equipment can be installed as soon as power is available. Nvidia has not responded to the accusations.
Critics warn, though, that if AI demand cools, Nvidia could be left with a glut of hardware, much like the GPU crash after the crypto mining boom.