The RTX 5090 may be dead, and its replacement has less memory
Key takeaways
- Leaker MEGAsizeGPU claims all future GB202 silicon is going to RTX Pro workstation and AI cards, which would end GeForce RTX 5090 production
- A second leaker says a factory notice names the RTX 5080 24GB as the new top GeForce, with the 16GB model discontinued
- Retail RTX 5090 listings already sit between roughly 5,000 and 9,000 dollars, and a halt would push them higher
- Nvidia has not commented, so treat the timing and the replacement card as unconfirmed until it does
A month ago we wrote that the RTX 5090 cost as much as a car. This week the story is that you may not be able to buy a new one at all. Two separate leakers say Nvidia is pulling its biggest consumer chip out of the GeForce line entirely.
What the leaks say
The first claim comes from MEGAsizeGPU, a leaker with a decent track record on Nvidia supply. The line is blunt: all future GB202 supplies are RTX Pro exclusive. GB202 is the silicon inside the RTX 5090 and its China-specific variants. If every wafer goes to workstation and AI cards, there is no RTX 5090 left to build.
The second claim, from hongxing2020, cites what is described as a factory notice. It says the top GeForce model becomes an RTX 5080 with 24GB of memory, and the existing 16GB RTX 5080 is discontinued. MEGAsizeGPU has not backed that part up. The 24GB configuration is not new as a rumour, since it has been floating around since 2024 as a possible RTX 5080 Super, built on the smaller GB203 chip with 3GB GDDR7 modules.
Nvidia has not said a word. That matters. Supply chain whispers about Nvidia are often right in direction and wrong in timing, so the sensible reading is that GB202 is being prioritised for professional cards, not that a date is set.
Why Nvidia would do this
Follow the money. An RTX 5090 sells for 1,999 dollars at list, and Nvidia's cut is a fraction of that. The same GB202 die in an RTX Pro 6000 sells for several times more, to buyers who order in racks rather than ones. With TSMC's advanced capacity rationed and AI demand still climbing, every GB202 that goes into a gaming card is money left on the table.
This is the same logic that has already pushed memory makers to lock up 2027 supply and nudged Intel and Qualcomm into raising CPU prices. Consumer hardware is now what gets built with whatever is left over.
What it means if you want one
Existing stock does not vanish. Board partners and retailers can keep selling RTX 5090 cards until they run out, and some will hold inventory precisely because a halt would push prices up. Listings already sit between roughly 5,000 and 9,000 dollars in the US, far above the 1,999 dollar launch price, and a confirmed end of production would lift them again. If you have been waiting for a sane price on a 5090, that wait probably ends in disappointment.
The RTX 5080 24GB, if it exists, is a different proposition. It would fix the one complaint about the current 5080 (16GB is tight for 4K and local AI) without the GB202 price tag. Speculative pricing of 1,999 to 2,999 dollars has been floated, but nobody outside Nvidia knows.
For most people the practical answer has not changed. A current RTX 5080 or a Radeon RX 9070 class card does nearly everything a 5090 does at a fraction of the cost. The 5090 was always a halo product. It may now be a collector's item.
