What the leak says
The claim started with a factory notice shared by the leaker @hongxing2020. According to Tom's Hardware, the notice said Nvidia would stop making every RTX 5090 model, including the "D" variants built for China. It also said the RTX 5080 24GB would become the top GeForce card and that the RTX 5080 16GB would be discontinued. A second leaker, MEGAsizeGPU (@Zed__Wang), backed the report. VideoCardz quoted his post: "All future GB202 supplies are RTX Pro exclusive".
Most coverage repeats the same core claim. VideoCardz summed it up this way: the leaker claims that GB202 GPUs will no longer be used for new RTX 5090 production. Instead, all newly produced GB202 chips will be allocated to NVIDIA's professional RTX PRO lineup. This would affect the standard RTX 5090 as well as its Chinese variants, which use the same GPU.
Taiwan's BenchLife added some useful detail. It describes the change as moving production capacity to more profitable product lines, not as a formal end-of-life notice. Its list of affected GeForce chips is GB202-300, GB202-250 and GB202-240. BenchLife also reports that Nvidia plans the RTX 5080 24GB for the first quarter of 2027. So far, though, partners have reportedly been told about it only verbally and have not received any development materials.
What has not been confirmed
- No official statement. NVIDIA has not confirmed these plans or announced an end-of-production date. Tom's Hardware asked Nvidia for comment and had not heard back when it published. Nvidia usually doesn't comment on rumours.
- No named source document. The factory isn't named and the original notice hasn't been published.
- No timeline. France's Clubic noted that no timeline has been specified.
- The RTX 5080 24GB is still a rumour. Nvidia's own launch materials list only a 16GB RTX 5080 at $999.
Section summary: Two well-known leakers and BenchLife describe the same GB202 reallocation. All of it traces back to unverified supply-chain claims, and Nvidia has said nothing.
Halting chip supply is not the same as killing the card
Notebookcheck makes a key point here. It says it's important to distinguish between the alleged end of new GB202 chip allocation and a complete discontinuation of the RTX 5090. Nvidia and its board partners could continue selling cards built from existing inventory.
That matches how the GPU channel works. When the RTX 5090 launched, Nvidia said cards would come from many partners: ASUS, Colorful, Gainward, GALAX, GIGABYTE, INNO3D, KFA2, MSI, Palit, PNY and ZOTAC. Systems builders such as Falcon Northwest, Maingear and Origin PC would also ship it. Chips, boards and finished cards already in that pipeline don't disappear when new allocations stop.
How long that stock will last is unknown. No published data shows how much GB202 silicon or finished RTX 5090 inventory partners and retailers hold.
Why Nvidia would do this: the GB202 economics
GB202 is the largest Blackwell desktop GPU. It goes into both the gaming flagship and the professional range.
| Card | Market | Memory | Notes |
|---|---|---|---|
| GeForce RTX 5090 | Consumer | 32GB GDDR7, 512-bit | 21,760 CUDA cores (Nvidia spec); $1,999 launch price |
| RTX PRO 5500 Blackwell | Workstation (rack-mounted) | 84GB GDDR7 ECC | 1,398GB/s bandwidth, up to 600W; Nvidia lists it as "Coming Soon" with preliminary specs |
| RTX PRO 6000 Blackwell (Server, Workstation, Max-Q) | Workstation / data center | 96GB GDDR7 ECC | Server Edition aimed at inference, fine-tuning, HPC and virtual workstations |
Nvidia's specs confirm the RTX 5090's 21,760 CUDA cores. Nvidia's RTX PRO 5500 page does not publish a CUDA core count. The claim that the RTX PRO 5500 matches the RTX 5090's core count comes from Tom's Hardware, not Nvidia. Several outlets report the RTX PRO 6000's core count. Kantan News, for example, says it uses the full version of the same graphics processor, caters to professional workloads with 24,064 CUDA cores and 96 GB of GDDR7 memory capacity. Wccftech also lists the 72/48 GB RTX PRO 5000 among the GB202 professional cards.
The business case is easy to follow even without margin figures. Each GB202 die can go into a gaming card with 32GB of memory or a professional card with 84GB or 96GB of ECC memory, enterprise features and enterprise pricing. Nvidia's RTX PRO 5500 page says the card can be split into two isolated Multi-Instance GPU instances. The RTX PRO 6000 Server Edition is pitched for multi-GPU server deployments. Neither of those is a gaming feature.
Wccftech's explanation is that DRAM continues to increase in terms of prices, and demand for AI continues to jump, this might have pushed NVIDIA to refocus its current supply of GB202 chips to Pro cards. That's the outlet's reading, not something Nvidia has said. No verified margin, yield or allocation numbers are public.
There's also a cannibalisation angle. Tom's Hardware argues that dropping the GB202 GeForce cards would stop relatively cheap consumer cards from competing with Nvidia's own professional accelerators in AI work. Anyone who has seen a lab fill a rack with consumer flagships for local LLM inference knows why. A 32GB card at a gaming price cuts into the market for a 96GB card at a workstation price.
Section summary: The economic logic makes sense. Still, the specific comparisons of price and profitability are analysts' inferences, not Nvidia disclosures.
What it means for buyers
Coverage of this rumour has prompted plenty of "buy now before it's gone" advice. It's worth pausing before you act on that.
Street prices for the RTX 5090 have already moved far above launch. Wccftech says the card has already reached prices north of $6000, and this is already Pro-tier pricing. Notebookcheck reports European prices around €6,000 and suggests the card could become even more expensive once no new units are being supplied. These figures come from those outlets' own market reading and will vary by region and retailer.
Tighter supply could push prices up further. But nobody has established when supply runs dry or how big the effect would be. If you're thinking about buying an RTX 5090, here's a sensible approach:
- Check current stock and prices yourself. Don't rely on figures quoted in rumour coverage.
- Work out whether you need GB202-class hardware. For 4K gaming, an RTX 5080 may cover most of your needs. For local AI work, the 32GB of VRAM may be what you're really paying for.
- Read the warranty and return terms closely, especially from third-party sellers. Inflated grey-market prices often come with weaker support.
- Don't let a deadline that only exists in a rumour rush a four-figure purchase.
For IT and procurement teams, the leak is a reminder that consumer cards are risky as a basis for AI or visualization workstations. If GB202 does become RTX PRO-only, buyers who standardised on RTX 5090 builds would need to choose between professional cards with longer, more predictable supply and a big price jump, or lower GeForce tiers with less memory.
The rumoured new flagship: RTX 5080 24GB
According to the leaked notice, the RTX 5080 24GB would take over as the top GeForce card and replace the 16GB model, not sit beside it. Tom's Hardware links this to long-rumoured RTX 50 Super cards, which reportedly included a 24GB RTX 5080 Super delayed by memory supply problems. Notebookcheck expects the rumoured Super series to be largely unaffected by this development, since none of the expected models are said to be based on a GB202 GPU.
A 24GB RTX 5080 would be a big step down from a GB202 card in raw resources. It would still be a meaningful VRAM upgrade for people who run local models or work with heavy textures. Tom's Hardware also passes along separate rumours that Rubin-based RTX 60-series cards may not arrive until late 2027 or 2028. That timing is unconfirmed and has nothing to do with whether the GB202 report is true.
Bottom line
Here's what's established and what isn't:
- Established: Two leakers and BenchLife report that GB202 allocations for GeForce cards have stopped and the chips are going to RTX PRO products. The RTX 5090 and the RTX PRO 6000 family share GB202 silicon, according to several outlets. Nvidia's official specs confirm the memory gap: 32GB on the RTX 5090 versus 84GB and 96GB on the professional cards.
- Unconfirmed: Nvidia's actual decision, whether it's permanent or global, the RTX 5080 24GB and the end of the 16GB model, and any timeline for supply or prices.
If the report is accurate, gamers lose the top of the GeForce range to the AI boom. GB202 would still be made, just sold to workstation and data center buyers. That fits a pattern seen across the industry, where silicon and memory go to whichever customers pay the most. Until Nvidia or its partners say something on the record, it's a well-sourced rumour and should be treated as one.