Leaks say NVIDIA will end GeForce RTX 5090 production and reserve all GB202 dies for the high-margin RTX PRO line. On the consumer side, a new 24 GB RTX 5080 is expected to take the crown.

The fastest gaming card on the planet may be quietly riding into the sunset. Prolific leaker MEGAsizeGPU claims NVIDIA is preparing to wind down GeForce RTX 5090 production. The supply route of GB202 , the largest graphics processor of the Blackwell design, is allegedly changing course entirely — leaving a gaping hole at the top of the GeForce stack.

MEGAsizeGPU's name carries weight in these circles: the same source called the giant 744-square-millimeter GB202 die size months in advance back in 2024, a measurement Wccftech logged as 22 percent larger than AD102. The scope of the new claim is sharp: none of the fresh GB202 dies will go to the RTX 5090 or its China-only variants; the route swings fully toward corporate models, with the RTX PRO 6000 Blackwell and RTX PRO 5000 named in the leak.

Same silicon, eightfold price tag: the corporate pull

The arithmetic is brutal. The RTX 5090 is a cut-down GB202 carrying 21,760 CUDA cores and 32 GB of GDDR7 memory. The RTX PRO 6000, by contrast, uses the full die, according to Nvidia's official figures: 24,064 CUDA cores, 96 GB of ECC memory and a 600 W power limit. At Microcenter the workstation card sits around the $16,500 mark even after a $500 cut from its $16,999.99 tag, while the gaming flagship sells for $1,999. Selling the same fab output at eight times the price, setting silicon aside for gamers starts to look like waste from the company's window.

Demand points the same way. Nvidia's data credits the corporate card with 4 PFLOPS of AI power at FP4 precision and a shared memory pool reaching 192 GB over NVLink — enough to run 70B-class systems without moving to the cloud, turning workstations into pocket data centers. With the race to feed AI servers still on, writing every GB202 die to the corporate pool instead of gaming cards looks like the supply chain's natural choice.

Shelves empty out, prices fly

Market signals agree with the leak. Per a September 2026 stock check relayed by Shattered, the RTX 5090 has nearly vanished from first-party shelves across the United States, with third-party listings stretching from $6,395 to $9,500 — up to 4.75 times the launch tag. Only regional leftovers at chains like Microcenter are said to remain. The picture suggests output is already being throttled in practice; with or without a formal EOL notice, the card has become all but unbuyable.

A successor for the empty throne is taking shape. According to the Hongxing2020 claim from a manufacturing partner, NVIDIA has told its partner the 5090 line is stopping and set its sights on a new 24 GB GeForce 5080. The SUPER dimensions Overclock3D relayed from the kopite7kimi leak — 10,752 CUDA cores, 24 GB of GDDR7 at 32 Gbps and power beyond 400 W — line up exactly with that picture. A Seasonic power-supply calculator entry logged by Club386 points at the same card with a 415 W limit and a single 16-pin plug; about 15 percent above today's 360 W 5080, it drags the PSU recommendation up to 800 W.

A 24 GB 5080 for the crown

Still, the file carries no official seal. NVIDIA has issued no confirmation; Techenclave notes the company has quietly pulled 5090 and 5080 Founders Editions from its global showcase, leaving only the 5070 Founders on the shelf — a move that alone means no farewell, possibly stock balancing or SUPER groundwork. Tweaktown's reading of a de facto exit from high-end PC gaming remains a one-sided take for now; promoting an unverified leak to headline status is premature.