Nvidia has reportedly cut the GeForce RTX 5090 off from new supplies of GB202, the full-size Blackwell die that made the card the fastest consumer GPU on the market. The claim, first raised by hardware leakers and repeated by outlets including Tom’s Hardware, VideoCardz and Guru3D, has not been confirmed by Nvidia itself, which says the card is not discontinued and will continue shipping. But the underlying pattern it describes is not new. Nvidia has quietly favored professional and data-center silicon over gaming flagships before, and looking at those earlier moves says more about where the RTX 5090 is headed than the leak alone ever could.
This piece already saw tech-insider.org’s initial report lay out the raw claim. What follows is the deeper read: the chip economics behind the decision, the eight-year history of Nvidia doing exactly this kind of reallocation, what it means for AMD and Intel, and what buyers in both the gaming and AI markets should actually expect over the next year.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the Leaks Actually Say About GB202 and the RTX 5090
The story traces back to posts from hardware leakers who claimed Nvidia had instructed board partners that no further GB202 dies would be earmarked for GeForce RTX 5090 production, with all remaining and future allocation going to the RTX Pro line. One leaker wrote that “all future GB202 supplies are RTX Pro exclusive,” a claim TweakTown reported alongside a second, separate claim describing an alleged factory notification stating Nvidia “will no longer manufacture any RTX 5090 models, including the D variants.”
Nvidia has issued no official statement confirming either claim, and the denial attributed to Jeff Fisher, who said the RTX 5090 would continue shipping, comes from press reports rather than a direct company release. Nvidia’s public position, reported October 17–24, is that the card is not discontinued, although supply is constrained. There is no official end-of-production date, no confirmed unit count tied to the change, and no public comment from board partners such as Asus, MSI or Gigabyte. What can be said with confidence is narrower than the headlines suggest: multiple independent leak sources are describing the same reallocation, several enthusiast hardware outlets consider the sourcing credible enough to report, and nothing in the allocation claims has been verified by Nvidia on the record.
Inside GB202: What Nvidia Has Actually Confirmed
Separate from the unconfirmed supply claims, Nvidia’s own Blackwell architecture documentation lays out exactly what GB202 is. The official Blackwell GPU architecture document identifies GB202 as the flagship RTX Blackwell die and confirms it powers the GeForce RTX 5090. In full configuration, GB202 carries 12 Graphics Processing Clusters, 96 Texture Processing Clusters, 192 Streaming Multiprocessors, a 512-bit memory interface and 128 MB of L2 cache. The RTX 5090 itself, as sold, uses a cut-down version of that die: 21,760 CUDA cores, 32 GB of GDDR7 memory, a 2.01 GHz base clock and a 2.41 GHz boost clock, with 96 MB of L2 cache rather than the full 128 MB.
Nvidia also lists 5th generation Tensor Cores (Tom’s Hardware counts 680 of them), 4th generation Ray Tracing Cores, 3,352 AI TOPS and 318 TFLOPS of shader throughput for the card, while the 512-bit memory bus provides 1.79TB/s of bandwidth according to Tom’s Hardware. The gap between the RTX 5090’s specification and the full GB202 die matters, because it shows Nvidia already holds back part of the chip’s capability for binning purposes. That headroom is exactly what makes GB202 flexible enough to serve two very different markets from a single piece of silicon.
Why Professional and AI Silicon Pays More Than Gaming Cards
None of this would matter if gaming and professional GPUs generated similar revenue per wafer. They do not. A single wafer of GB202 can become either a GeForce card that launched at $1,999 or a workstation-class product that reportedly sells for several multiples of that figure, depending on configuration and memory capacity. TechPowerUp reported in October 2026 that Nvidia was reserving GB202 for the 96GB RTX PRO 6000, and Nvidia introduced the RTX Pro 5500 on September 14, 2026 using a GB202 configuration reported as matching the RTX 5090’s CUDA-core configuration but with 84 GB of memory and ECC support. Exact pricing for those configurations has not been independently verified here and should be treated as reported rather than confirmed.
What is not in dispute is the broader economic logic. Enterprise buyers purchasing GPUs for AI training, inference or professional visualization work pay a premium tied to support, certified drivers, memory capacity and reliability guarantees that gaming customers do not. When semiconductor supply is tight and allocation decisions have to be made upstream at the foundry and packaging stage, a chipmaker under pressure to maximize revenue per die has an obvious incentive to steer output toward the higher-margin bucket first and backfill gaming demand second, if at all.
History Repeats: Nvidia’s Pattern of Steering Flagship Dies to Pro Markets
This would not be the first time Nvidia has blurred the line between its gaming flagship and its professional lineup. The company has used closely related silicon across both markets for at least four generations, even if it has never before publicly described a decision as a reallocation away from GeForce.
Titan RTX and the Quadro Bridge (2018)
Nvidia launched the Titan RTX in December 2018 using the full TU102 die, positioning it as a bridge product between consumer GeForce cards and the professional Quadro RTX 6000 and RTX 8000. It was not framed as a supply diversion, but it established the template: a single high-end die serving overlapping gaming and professional roles, with pricing scaled to whichever market a given SKU targeted.
Ampere’s Split Dies and Ada’s Shared Silicon (2020-2023)
The 2020 generation complicated the pattern rather than confirming it. The GeForce RTX 3090, launched that September, used the GA102 die, while the data-center A100, announced earlier that May, used the larger and architecturally distinct GA100 die. That generation is not a clean example of a shared die being redirected, since the two products were built on different silicon from the start. The clearer precedent arrived with Ada Lovelace: the RTX 4090 used the AD102 die, and Nvidia’s L40 workstation and AI-inference card, announced in October 2022, used an AD102 derivative for data-center visualization and AI workloads. Nvidia never stated publicly that GeForce supply was redirected to L40 production, but AD102’s dual use across gaming and enterprise products mirrors what the current GB202 reports describe, just without an explicit public claim of a cutoff.
Measured from the Titan RTX’s December 2018 launch to the current GB202 reports in October 2026, Nvidia has spent roughly eight years refining the same basic playbook: build one flagship die, bin it multiple ways, and let whichever market pays more pull supply away from the other when capacity gets tight.
The RTX 5080 24GB Rumor: A Possible New Flagship
If the RTX 5090 does fade from new production, the most immediate question for gamers is what replaces it at the top of Nvidia’s GeForce stack. Reports have floated an RTX 5080 variant with 24 GB of GDDR7 memory, up from the existing RTX 5080’s 16 GB, as a possible enthusiast-tier successor. This remains a rumor rather than a confirmed Nvidia product. No release date, final specification, memory bus configuration or price has been officially announced, and it should be read as a reported possibility rather than an upcoming launch.
Even if accurate, a 24 GB RTX 5080 would not necessarily match RTX 5090 performance, since it would likely retain a smaller GPU configuration than the cut-down GB202 used in the 5090. It would plug the memory-capacity gap that matters for some creative and AI workloads without closing the raw compute gap that defines flagship-tier gaming performance.
Market Impact: What This Means for PC Gamers Right Now
For anyone currently shopping for a flagship GPU, the practical effect of this story is scarcity risk rather than an immediate price change. The RTX 5090 has already been difficult to find at its official listing in many markets, a trend tech-insider.org covered in detail as secondary-market prices climbed well above MSRP over the past month. Reported retail and secondary-market prices have ranged from roughly $5,000 to $11,565 depending on region and seller, and TechSpot reportedly found most listings above $9,300 on October 19–24, against the $1,999 launch MSRP. If new GB202 allocation genuinely shifts toward RTX Pro products, existing retail inventory would need to carry demand for longer, which tends to push resale and gray-market prices higher rather than lower.
Retailers have also started treating the RTX 5090 as a card to ration rather than stock freely. Micro Center’s recent move tying RTX 5090 sales to an ID check and no-export pledge reflects a retail environment already bracing for tighter supply, independent of whether Nvidia’s reported GB202 shift turns out to be accurate.
GDDR7 Supply Adds a Second Squeeze
The GB202 allocation story is not happening in isolation. Memory supply has been its own constraint on Nvidia’s entire GPU lineup this year, with GDDR7 chip costs reportedly shaping the launch order of next-generation RTX 6080 and RTX 6090 cards. Even a GPU with a secured silicon allocation still needs enough GDDR7 modules to ship in volume, and tight memory supply compounds the effect of any chip-allocation shift, since board partners facing both a silicon constraint and a memory constraint have fewer paths to maintain full production.
Competitive Landscape: Can AMD or Intel Capitalize?
AMD’s current high-end lineup, built on RDNA 4, tops out with the Radeon RX 9070 XT and RX 9070, cards that compete with Nvidia’s RTX 5070 Ti and RTX 5080 rather than the RTX 5090. AMD has no confirmed product positioned as a direct RTX 5090-class replacement, and the company has not issued any public statement responding to Nvidia’s reported GB202 reallocation. The architecture expected to follow RDNA 4, often referred to in industry reporting as UDNA or RDNA 5, has no confirmed consumer launch date.
That absence matters. If the RTX 5090 genuinely becomes scarce, the natural beneficiary in the near term is Nvidia’s own RTX 5080, not an AMD flagship, simply because AMD does not compete at that performance tier right now. AMD’s more immediate opportunity sits a tier lower, where memory pricing pressure has already been reshaping the market; the RX 9070 GRE’s recent price movement amid a broader GPU memory price surge shows AMD competing hardest where Nvidia’s supply constraints are least relevant. Intel’s Arc lineup remains positioned at the budget and mid-range tiers and is not a factor in any RTX 5090 succession story.
What This Means for AI Developers and Enterprise Buyers
On the other side of the ledger, developers and enterprises buying GPUs for AI inference, training or professional visualization stand to benefit from the reported shift, at least in terms of availability. If Nvidia is indeed prioritizing GB202 output for RTX Pro-class products, buyers in that segment could see steadier supply at a moment when demand for AI-capable silicon across the industry remains intense. That said, professional-tier GPUs carry a meaningfully higher price floor than consumer cards even in normal supply conditions, and nothing in the current reporting suggests enterprise pricing is dropping. The practical read for a developer evaluating hardware for local model work is that the RTX 5090 may simply become a harder card to source for that purpose, pushing some buyers toward Nvidia’s purpose-built professional and workstation products instead, even for workloads that a GeForce card could technically handle.
RTX 5090 Confirmed Specifications at a Glance
The following figures come directly from Nvidia’s official specifications and architecture documentation, separate from any of the unconfirmed supply-allocation claims discussed above.
| Specification | GeForce RTX 5090 (as shipped) | Full GB202 die |
|---|---|---|
| CUDA Cores | 21,760 | Higher core count available on full die |
| Memory | 32 GB GDDR7 | 512-bit memory interface |
| L2 Cache | 96 MB | 128 MB |
| Graphics Processing Clusters | Cut-down configuration | 12 GPCs |
| Texture Processing Clusters | Cut-down configuration | 96 TPCs |
| Streaming Multiprocessors | Cut-down configuration | 192 SMs |
| Boost Clock | 2.41 GHz | Not separately specified |
| AI Performance | 3,352 AI TOPS | Not separately specified |
| Shader Throughput | 318 TFLOPS | Not separately specified |
Nvidia’s History of Shared Gaming and Professional Dies
Looking back across four Nvidia architecture generations shows how consistently the company has built one flagship die and binned it for multiple markets, even without ever publicly announcing a cutoff the way the current GB202 reports describe.
| Generation | Gaming Product | Professional/Datacenter Counterpart | Shared Silicon? |
|---|---|---|---|
| Turing (2018) | Titan RTX | Quadro RTX 6000 / RTX 8000 | Full TU102 die shared across tiers |
| Ampere (2020) | GeForce RTX 3090 | A100 | Different dies (GA102 vs. GA100), not shared |
| Ada Lovelace (2022-23) | RTX 4090 | L40 / L40S | AD102 derivatives shared across both products |
| Blackwell (2025-26) | GeForce RTX 5090 | RTX Pro line (reported) | GB202 die reportedly shared, allocation shift unconfirmed |
What Leakers and Insiders Are Saying
The GB202 reallocation claim rests almost entirely on statements from hardware leakers rather than official Nvidia channels, so it is worth reading their exact wording rather than a paraphrase. One leaker, posting as MEGAsizeGPU, stated plainly: “No more GB202 for 5090, confirmed,” before adding in a follow-up that “all future GB202 supplies are RTX Pro exclusive,” a claim TweakTown reported alongside corroborating commentary from a second source.
A separate leaker, hongxing2020, described the alleged factory-level notice in more specific terms, stating that “according to a notification from the factory, NVIDIA will no longer manufacture any RTX 5090 models, including the D variants.” Neither claim has been verified against an Nvidia source, board-partner statement or confirmed internal document, and both should be read as sourced industry chatter rather than established fact.
Five Predictions for the Next 12 Months
- The RTX 5090 will not disappear overnight. Existing board-partner inventory and already-allocated dies will likely keep trickling into retail through at least early 2027, even if new GB202 orders shift toward RTX Pro products.
- Secondary-market prices for the RTX 5090 will stay elevated or climb further if any confirmed supply reduction materializes, extending the resale pressure already documented over the past month.
- Nvidia is more likely to let the RTX 5090 quietly fade than to issue a formal discontinuation notice, consistent with how it has handled die-sharing transitions in past generations.
- If a 24 GB RTX 5080 variant does launch, expect it to be positioned as a memory-capacity upgrade for creative and light AI workloads rather than a true flagship performance successor.
- AMD is unlikely to introduce a direct RTX 5090-class competitor within this product cycle, leaving any near-term flagship-tier gap to be filled by Nvidia’s own RTX 5080 lineup rather than a rival vendor.
Frequently Asked Questions
Has Nvidia officially confirmed it stopped making the RTX 5090?
No. The claim originates with hardware leakers and has been reported by multiple enthusiast outlets, but Nvidia has not issued an official statement confirming an end-of-production date or a formal discontinuation. Jeff Fisher said the RTX 5090 would continue shipping, and Nvidia’s public position, reported October 17–24, is that the card is not discontinued.
What is GB202 and why does it matter?
GB202 is Nvidia’s flagship Blackwell architecture die, confirmed by Nvidia’s own documentation to power the GeForce RTX 5090. Reports claim Nvidia is redirecting new GB202 supply toward its RTX Pro professional line instead of GeForce cards, though that reallocation itself is unconfirmed.
Will there be a replacement for the RTX 5090?
Reports have floated an RTX 5080 variant with 24 GB of memory as a possible successor, but this remains an unconfirmed rumor with no official specification, price or release date.
Has Nvidia done this kind of reallocation before?
Yes, in a related sense. Nvidia has shared high-end dies across gaming and professional products going back to the Titan RTX in 2018 and continuing through the Ada Lovelace generation’s AD102-based RTX 4090 and L40 cards in 2022. It has not previously made a public claim of fully cutting off GeForce supply the way current GB202 reports describe.
Does AMD have a GPU that competes with the RTX 5090?
No. AMD’s current RDNA 4-based Radeon RX 9070 XT and RX 9070 compete with Nvidia’s RTX 5070 Ti and RTX 5080 tiers, not the RTX 5090. AMD has not announced a product positioned as a direct flagship-tier replacement.
Should I buy an RTX 5090 right now if I can find one at a reasonable price?
If the reported supply shift is accurate, availability at or near official pricing is likely to get harder rather than easier in the coming months, based on the retail and resale trends already visible in the market.
Why would Nvidia prioritize professional GPUs over gaming cards?
Professional and AI-focused GPUs built on the same silicon generally generate significantly higher revenue per chip than consumer gaming cards. When chip supply is constrained, that margin difference creates a clear incentive to allocate scarce dies to the higher-value market first.
Is this the same as last month’s RTX 5090 resale price story?
It is related but distinct. The resale price surge reflects existing scarcity in the market today. The GB202 reallocation claim, if accurate, describes a forward-looking supply decision that could extend or worsen that scarcity rather than resolve it.
