Nvidia’s best-selling gaming card may be losing a fight it never had to have in public. Reports circulating since October 10, 2026 claim the company is steering future GB202 chip output away from the GeForce RTX 5090 and toward its far more profitable RTX Pro workstation cards, a move that would leave the $1,999 flagship without fresh silicon once existing inventory clears. Nvidia has not confirmed the shift. But the reporting chain is wide enough, and the underlying economics are clear enough, that the story is worth examining as a market event rather than a one-day rumor.
What makes this moment different from a typical leak cycle is the math behind it. GB202 is a single, enormously expensive die that Nvidia can sell two ways: inside a $1,999 gaming card, or inside a professional GPU that commands several times that price. When a chipmaker has a limited number of wafers and two buyers competing for the same silicon, the buyer paying more usually wins. This piece looks at what is actually confirmed about the RTX 5090 and GB202, what remains an unverified leak, how the professional GPU market factors into Nvidia’s decision, and what this likely means for gamers, system builders, and Nvidia’s closest competitor over the next year.
What’s Being Reported About GB202 and the RTX 5090
The core claim, as described by VideoCardz.com, is that Nvidia has stopped allocating newly produced GB202 chips to the GeForce RTX 5090 and will direct all future GB202 output to the professional RTX Pro lineup instead. The claim traces back to posts from hardware leakers identified in various reports as MEGAsizeGPU, hongxing2020, Zed__Wang, and Jukan. None of them are Nvidia employees, and none of the outlets covering the story, including Tom’s Hardware, have identified an on-record Nvidia statement confirming the allocation change.
That absence of confirmation is the single most important fact in this entire story, and it is worth repeating before going further: Nvidia has not announced that RTX 5090 production is ending, has not confirmed any change to GB202 allocation, and has not said whether current retail inventory represents the last of the card. Everything downstream of that point, including every number in this article about future supply, is reported speculation layered on top of a leak, not a company disclosure.
Still, the claim is plausible precisely because it fits a pattern chipmakers follow when a single die serves multiple product tiers. TweakTown and Hardware Busters both frame the move as a straightforward allocation decision rather than a conspiracy: wafer capacity is finite, and Nvidia gets to choose which finished product absorbs that capacity first.
The RTX 5090’s Confirmed Specifications
Separating rumor from record starts with what Nvidia has actually published. The GeForce RTX 5090 ships with 21,760 CUDA cores, 32 GB of GDDR7 memory, a 2.01 GHz base clock and a 2.41 GHz boost clock. Nvidia rates its fifth-generation Tensor Cores at 3,352 AI TOPS and its fourth-generation Ray Tracing Cores at 318 TFLOPS. Those figures come directly from Nvidia’s own product listing, not from any leaker, and they are the baseline against which every rumor about the card’s future has to be measured.
The RTX 5090 launched as the flagship of the GeForce RTX 50 series and, according to Nvidia’s architecture documentation, was the first RTX 50-series product to ship using the GB202 die. That detail matters because it establishes the RTX 5090 as GB202’s original home. Whatever is happening with allocation now is a change from that starting point, not a continuation of an existing arrangement.
Inside GB202: The Chip Both Product Lines Are Fighting Over
GB202 is Nvidia’s largest Blackwell-generation GPU die, and its scale is a big part of why this story has legs. Nvidia’s own Blackwell architecture document, published at images.nvidia.com, puts the full chip at 92.2 billion transistors, arranged across 12 Graphics Processing Clusters (GPCs), 96 Texture Processing Clusters (TPCs), and 192 Streaming Multiprocessors (SMs). The full die carries a 512-bit memory interface and 128 MB of L2 cache.
The RTX 5090 does not use every part of that die. Nvidia’s architecture document confirms the gaming card ships with 96 MB of L2 cache, below the full chip’s 128 MB ceiling, which is a standard practice for consumer cards built on a die that also serves higher-margin products. Binning and partial enablement let Nvidia sell lower-yield or lower-spec dies into the gaming market while reserving fully enabled dies for workstation and data-center customers who pay a premium for every extra core and megabyte of cache.
Why a Single Die Serving Two Markets Creates Tension
This is where the economics of the rumor start to make sense independent of any leaker’s credibility. A foundry only produces so many GB202 wafers in a given quarter. Every wafer Nvidia turns into an RTX 5090 is a wafer it cannot turn into an RTX Pro card. If professional and AI infrastructure customers are willing to pay several multiples of the RTX 5090’s price for the same underlying silicon, a rational supplier allocates scarce wafers toward the higher-margin buyer first and lets the lower-margin product absorb whatever capacity is left. That is not a scandal. It is how component allocation has worked across the chip industry for decades, and it is the same basic logic that has periodically squeezed gaming GPU supply whenever data-center or professional demand spikes.
The RTX Pro Lineup Competing for the Same Silicon
Reports naming the professional cards reportedly built on GB202 point to three models: the RTX Pro 6000 Blackwell, the RTX Pro 5500, and the RTX Pro 5000. These are workstation and AI-development cards aimed at studios, researchers, and enterprise buyers rather than gamers, and they are positioned well above the RTX 5090 on Nvidia’s internal price ladder. It is worth being precise about confirmation status here: claims that the RTX Pro 6000 carries 96 GB of memory, that the RTX Pro 5500 carries 84 GB, or that professional cards sell for four to five times the RTX 5090’s price are not established by the official Nvidia specifications identified for this story. They circulate in outlet reporting, not in a verified Nvidia price sheet, so they belong in the “reported” column rather than the “confirmed” one.
What is not in dispute is the category these cards occupy. Professional GPUs built on flagship dies have historically sold into markets, AI training clusters, rendering farms, scientific computing labs, that treat GPU cost as a line item against a much larger budget. A studio pricing out a render farm or a lab speccing an AI workstation is not shopping on the same price sensitivity as someone buying a single gaming PC, which is exactly why a chipmaker facing a wafer shortfall gravitates toward that buyer first.
RTX 5090 vs. the Full GB202 Die: What’s Actually Confirmed
The table below isolates confirmed, Nvidia-published specifications only. It shows how much of the full GB202 die the RTX 5090 actually uses, and why there is still headroom in the chip that a fully enabled professional card could claim.
| Specification | GeForce RTX 5090 (as shipped) | Full GB202 Die |
|---|---|---|
| CUDA Cores | 21,760 | Not separately published by Nvidia |
| Streaming Multiprocessors (SMs) | Subset of 192 total | 192 SMs |
| Graphics Processing Clusters | Subset of 12 total | 12 GPCs |
| Texture Processing Clusters | Subset of 96 total | 96 TPCs |
| Memory | 32 GB GDDR7 | 512-bit interface (full die) |
| L2 Cache | 96 MB | 128 MB |
| Transistor Count | Same physical die | 92.2 billion |
| Boost Clock | 2.41 GHz | Die-dependent, not separately listed |
| AI Performance (Tensor Cores) | 3,352 AI TOPS | Varies by product configuration |
| Ray Tracing Performance | 318 TFLOPS | Varies by product configuration |
That gap between 96 MB and 128 MB of L2 cache, and the fact that the RTX 5090 does not draw on the die’s full 192-SM, 12-GPC, 96-TPC configuration, is precisely the room Nvidia has to build a more fully enabled card for a buyer willing to pay more for it. The RTX 5090 was never using every transistor on the wafer to begin with.
Market Impact: What This Means for Buyers Right Now
Even as an unconfirmed rumor, this story changes behavior in the channel. Retailers, system integrators, and AI workstation builders do not wait for a company press release before adjusting purchasing plans when enough independent outlets converge on the same claim. If GB202 wafers are genuinely being redirected, RTX 5090 cards currently sitting in warehouses and on store shelves become more valuable by the week, simply because replacing them gets harder. That dynamic has already played out with past Nvidia GPUs facing supply constraints: cards nearing the end of their production run routinely hold or gain value on the secondary market precisely because buyers price in scarcity before a company confirms it.
For gamers specifically, the practical risk is not that RTX 5090 cards vanish overnight. It is that the window to buy one at anything resembling the $1,999 launch price narrows, while enthusiasts who already own one find their card appreciating rather than depreciating, an unusual position for a nine-month-old GPU. For AI hobbyists and small studios who buy the RTX 5090 specifically because it is the cheapest way to get 32 GB of GDDR7 and Blackwell-class Tensor Cores, the calculus gets harder: wait and hope supply holds, buy now at whatever price is available, or shift attention to the RTX 5080 class of cards that several outlets, including Tom’s Hardware, suggest could become Nvidia’s new flagship-adjacent option by default.
Where AMD Fits Into the Supply Squeeze
Any constraint on Nvidia’s top consumer card is, almost by definition, an opening for AMD’s Radeon lineup. AMD does not currently field a direct RTX 5090 competitor in raw specification terms, but its RX 9070 series occupies the upper-mid tier of the discrete GPU market and has already shown it can gain share when Nvidia pricing or availability wobbles. A reported RTX 5090 supply squeeze gives AMD a straightforward marketing angle even without a new product launch: highlight price-to-performance at the tiers where Radeon already competes, and let Nvidia’s scarcity do the rest of the persuading.
The bigger strategic question is whether AMD has the wafer allocation and die design to capture buyers who get priced or supply-constrained out of the RTX 5090 tier, rather than just the mid-range shoppers it already serves well. That is a 2027 roadmap question more than an October 2026 one, but it is the kind of opening AMD has historically been slow to fully convert into sustained market-share gains against Nvidia’s high end.
Historical Context: Nvidia Has Favored Data-Center Demand Before
This would not be the first time Nvidia’s gaming lineup absorbed the cost of surging demand elsewhere in its business. Across prior GPU generations, periods of high demand for data-center and professional compute have repeatedly coincided with tighter supply and higher street prices for Nvidia’s consumer cards, as the company’s limited wafer allocation gets pulled toward whichever buyer generates more revenue per chip. The pattern is consistent enough that hardware outlets treat “data-center demand squeezes gaming supply” as a recurring storyline rather than an isolated event tied to any single GPU generation.
What makes the current moment notable is the scale of the gap between the two markets. AI infrastructure spending has grown so large relative to the consumer graphics card market that even a modest reallocation of wafer capacity toward professional and data-center products can meaningfully affect gaming GPU availability. Nvidia’s gaming division, historically its most visible consumer brand, is increasingly a smaller slice of a company whose growth is now driven by AI compute at a scale the GeForce brand was never built to generate.
Where RTX 5090 Sits on Today’s GPU Price Ladder
Context matters when assessing how painful a supply squeeze would actually be. The table below places the RTX 5090’s confirmed $1,999 launch price alongside recent street pricing reported elsewhere in hardware coverage for competing tiers, to show how much room exists below Nvidia’s flagship before buyers would need to look at AMD alternatives.
| Card | Vendor | Approximate Recent Street Price | Market Position |
|---|---|---|---|
| GeForce RTX 5090 | Nvidia | $1,999 (official launch MSRP) | Consumer flagship, GB202-based |
| GeForce RTX 5070 Ti | Nvidia | ~$1,100 (recently reported) | Upper mid-range |
| Radeon RX 9070 XT | AMD | ~$800 (recently reported) | Upper mid-range, Nvidia’s closest price competitor |
| Radeon RX 9070 GRE | AMD | ~$570 (recent promotional pricing) | Mid-range value option |
The roughly $1,200 gap between the RTX 5090 and the RX 9070 XT is already wide under normal supply conditions. If RTX 5090 availability tightens further while that gap holds or grows, the pressure shifts not just toward AMD’s existing price tiers but toward whatever Nvidia designates as its next de facto flagship, likely a higher-memory RTX 5080 variant, several outlets have speculated could reach 24 GB.
Why Nvidia Would Make This Trade Even Without Announcing It
From a pure business standpoint, the logic behind the leaked claim is hard to argue with even if the sourcing is thin. Nvidia does not need to formally discontinue the RTX 5090 to achieve the same practical effect as discontinuing it. Quietly slowing new GB202 allocation to the gaming card, while continuing to build RTX Pro units from the same wafers, produces a gradual supply decline that looks identical to a wind-down from the outside, without Nvidia ever having to issue a statement that invites scrutiny of its consumer pricing or its AI-driven margin strategy. Silence, in this case, may be the more profitable choice regardless of whether the underlying leak is accurate in every detail.
That also explains why Nvidia has reportedly pushed back informally on some of the leaker claims, per Startup Fortune’s reporting, without issuing a formal denial through investor or press channels. A company facing a credible rumor that happens to align with its own financial interests has limited incentive to definitively confirm or kill the story either way.
What to Watch Over the Next Few Months
Several concrete signals would confirm or undercut this story faster than any leaker’s social media post. Retail listings quietly disappearing without restock dates would be one. A sudden gap between RTX 5090 and RTX 5080 pricing narrowing, as retailers try to push buyers toward the lower tier, would be another. A formal RTX Pro price list or spec sheet from Nvidia, which has not yet appeared for this generation’s GB202-based professional cards, would resolve most of the “unconfirmed” figures circulating in current reporting. Until one of those happens, this remains a well-sourced but officially unverified supply story.
Predictions: Where This Likely Goes From Here
- RTX 5090 street prices are likely to drift upward over the next two to three months as retailers price in scarcity risk, regardless of whether Nvidia ever confirms the allocation shift.
- Nvidia will probably continue to avoid a direct statement on GB202 allocation, since neither confirming nor denying the rumor currently costs the company anything commercially.
- Expect Nvidia marketing attention, and possibly a refreshed RTX 5080 variant with more memory, to shift toward filling the gap left by a tighter RTX 5090 supply, rather than Nvidia announcing an RTX 5090 successor early.
- AMD is likely to lean on price-to-performance messaging around the RX 9070 series through the holiday shopping season without launching a new high-end die fast enough to directly capture displaced RTX 5090 buyers.
- A formal RTX Pro Blackwell price sheet, if and when Nvidia publishes one, will be the clearest signal of how much premium the professional lineup commands over GeForce, and will likely confirm that the economics favor professional allocation exactly as the current rumor claims.
The Bottom Line
Nothing here is as simple as “Nvidia killed the RTX 5090.” What is actually confirmed is that GB202 is a massive, expensive die that already serves a professional product line paying meaningfully more per chip than the gaming market does, and that the RTX 5090 does not even use the full die’s resources today. Whether or not the specific leaker claims about allocation turn out to be accurate in every detail, the underlying incentive for Nvidia to favor its highest-margin buyer when wafer supply is tight is not a rumor. It is basic chip economics, and it has reshaped GPU availability before. Buyers shopping for an RTX 5090 this quarter should treat today’s price and stock as less guaranteed than it looked a month ago.
Frequently Asked Questions
Has Nvidia officially confirmed it is ending RTX 5090 production?
No. Nvidia has not issued any statement confirming that RTX 5090 production is ending or that GB202 allocation has changed. The claim originates from hardware leakers and has been reported by multiple outlets, but it remains unconfirmed by Nvidia as of this writing.
What is GB202 and why does it matter?
GB202 is Nvidia’s largest Blackwell-architecture GPU die, confirmed by Nvidia’s own architecture documentation to contain 92.2 billion transistors, 192 Streaming Multiprocessors, and a 512-bit memory interface. It is the chip used in the GeForce RTX 5090 and reportedly in Nvidia’s RTX Pro professional cards as well.
What are the confirmed specs of the RTX 5090?
Nvidia officially lists the RTX 5090 with 21,760 CUDA cores, 32 GB of GDDR7 memory, a 2.01 GHz base clock, a 2.41 GHz boost clock, 3,352 AI TOPS from its Tensor Cores, and 318 TFLOPS of ray-tracing performance.
Will RTX 5090 prices go up if this report is accurate?
It is likely. If GB202 wafer allocation genuinely shifts away from the RTX 5090, existing retail inventory becomes harder to replace, which historically drives street prices above MSRP for GPUs nearing the end of their production life.
What is the RTX Pro lineup Nvidia is reportedly prioritizing?
Reports name the RTX Pro 6000 Blackwell, RTX Pro 5500, and RTX Pro 5000 as the professional cards reportedly sharing GB202 with the RTX 5090. Specific memory capacities and pricing for these cards circulating in outlet reporting are not confirmed by official Nvidia specifications.
Should I buy an RTX 5090 now or wait?
If the allocation shift is accurate, waiting carries real risk of higher prices or limited availability. Buyers who need the card’s specific combination of 32 GB of memory and Blackwell Tensor Core performance today have more reason to act than buyers who can flexibly consider the RTX 5080 tier instead.
Does this affect the RTX 5090 D variant sold in China?
Reports group the China-market D variant with the standard RTX 5090 because both use the same GB202 die. As with the main story, this detail comes from leaker claims rather than an Nvidia confirmation.
Could AMD benefit from an RTX 5090 supply squeeze?
AMD does not currently sell a direct RTX 5090 competitor, but tighter Nvidia flagship supply typically increases attention on AMD’s upper mid-range Radeon RX 9070 series, especially among price-sensitive buyers who get priced out of Nvidia’s top tier.
