A consumer card made for server racks
HKEPC reports that the modifications are intended to meet the space requirements of densely packed AI servers. The cards take up less room than standard consumer versions, but the route from the factories to their eventual buyers is hard to trace: intermediaries sit between the two.
A post accompanying the video claimed the cards were being sent to China for AI training and inference servers. It also said an RTX 5090 could not be found for less than $7,000 and called the situation a national-security failure. Those are claims made with the footage, not details established by the video itself.
For PC builders and gamers, the pressure is more direct. A finished RTX 5090 can cost well over $10,000, several times Nvidia’s official $1,999 suggested price. Nvidia says reports that production has stopped are false, but senior vice president Jeff Fisher acknowledged earlier this month that supply is limited. The card is also absent from Nvidia’s own online store.
What the footage does not show
Tom’s Hardware reports that Chinese businesses buying powerful GPUs and converting them for AI use may be one reason for the shortage. The video shows a large operation, but it does not show how much of the RTX 5090 supply it represents. Nor does it establish whether the modified cards are going to the training and inference servers described in the accompanying post.
I think that gap matters more than the dramatic footage. The conversion is plausible as a way to fit consumer GPUs into dense server racks; the harder question is how much it contributes to prices that can exceed $10,000. Nvidia’s admission that supply is limited gives buyers a reason to worry, but not a way to separate production constraints from demand by AI operators.
The squeeze leaves ordinary buyers competing for components against a market willing to pay for AI capacity. If that demand keeps drawing available cards into server infrastructure, PC builders may be left choosing between unaffordable new hardware and older parts that no longer meet modern needs.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X