RTX 5090 End-of-Production Rumor Breakdown: GB202 Shifting to Workstations, the Flagship Gap, and PC Builder Decisions#

Zixing | October 12, 2026

Hi everyone, I’m Zixing.

Last week, in the RTX 5090 retail market “extinction” piece, my lens was on retail prices — first-party stockouts, third-party five-figure listings, street price medians surging past twice the launch price. On October 10, two leaks plus a suspected factory notice screenshot pushed the question from “how expensive” to “whether there will be any at all”: reportedly, NVIDIA may have stopped supplying GB202 chips to the consumer market, redirecting all of it to the RTX PRO workstation line.

First, let me pin down this article’s information framing: as of the October 12, 2026 retrieval time, all of the above is rumor, with zero official confirmation from NVIDIA. Two leaks and one fakeable screenshot are the entire carrier of this rumor. NVIDIA’s official website still lists the RTX 5090 for sale. So in this article, every statement touching on “end of production” carries a “reportedly” or “the leaker claims” prefix — that is not writing timidly, it is the basic discipline of hardware reporting: a leak is a leak, a fact is a fact, and between them sits one official announcement of distance.

The division of labor with the previous piece also needs stating: the October 5 piece covered out-of-control pricing (a retail-side phenomenon); this one covers the rumor and product supply (supply-side logic) — why GB202 might be fully rerouted, who takes over the flagship slot, what the gap period market looks like, and what Q4 builders should do. Read together, the two pieces form the complete current picture of this graphics card.

Table of Contents#

1. The Rumor Itself: Two October 10 Leaks and One Screenshot
2. Credibility Assessment: Leaker Tier, Media Uptake, and the Official Posture
3. Business Logic: One GB202 Die, a Several-Fold Profit Gap
4. Product Line Restructuring Scenario: the Empty Flagship Slot and the 5080 24GB Positioning Contradiction
5. Market Impact of the Flagship Gap: Inventory, Secondhand, and the 5090D Question
6. Q4 Build Decisions: Three Types of Buyers, Find Yourself
7. Risk Notes: Rumor Grading and Data Timeliness

1. The Rumor Itself: Two October 10 Leaks and One Screenshot#

First, the three components of the rumor itself, laid out one by one. They come from two different leakers, with different carriers and different reliability — an important point that Section 2 will expand on.

Leak one: MEGAsizeGPU’s “GB202 goes all-workstation.” On October 10, 2026, hardware leaker MEGAsizeGPU (X platform account @Zed__Wang) posted: “All future GB202 supplies are RTX Pro exclusive” — from now on, all GB202 chip supply will go exclusively to the RTX Pro line. He then posted again to emphasize: “No more GB202 for 5090, confirmed.” Note the context of that word “confirmed”: it is the leaker emphasizing confidence in his own sources, not an official NVIDIA confirmation — the two differ by an order of magnitude.

Leak two: hongxing2020’s “EOL” and the factory notice screenshot. The same day, leaker hongxing2020 posted “5090 EOL, 5080 24GB ready,” attaching a screenshot of a suspected factory notice. Per the screenshot’s content (obtained and translated by VideoCardz), NVIDIA has notified at least one contract manufacturer that it will no longer sell any RTX 5090 model — note the scope: including the 5090D variant built for China’s regulatory-limited market. This leak also carried a product-line judgment: the RTX 5080 24GB will replace the 16GB version as the new GeForce flagship, with the 16GB version discontinued.

Here is a comparison table of the two leaks; note the third column — the informational strength of the two leaks is not equal:

This rumor cluster was picked up the same day by mainstream hardware outlets including Tom’s Hardware, VideoCardz, TechPowerUp, and Notebookcheck — but collective coverage is not collective endorsement; every outlet used rumor qualifiers in their reporting. That is precisely how professional outlets handle it: the heat is worth recording, the facts are not yet settled.

2. Credibility Assessment: Leaker Tier, Media Uptake, and the Official Posture#

Readers often ask: with so many hardware leaks, how do you judge which ones deserve trust? This rumor happens to be a textbook-grade sample, so let me break the assessment framework into three layers and walk through each.

Layer one: the leaker’s track record and evidence form. MEGAsizeGPU has a track record in hardware leak circles and his claims have been cited by multiple outlets; but his two posts are, in essence, assertions with no attached documentary evidence. hongxing2020’s leak came with a screenshot of a suspected factory notice — a screenshot looks more “solid,” but The Gamer Couch, while relaying it, stated the plain truth: “Notices get faked.” A harder-looking evidence form is not the same as more credible evidence — the cost of faking a notice image is far below the cost of faking an entire supply-chain story. That is the baseline instinct for reading leaks.

Layer two: media uptake and each outlet’s stance. Mainstream outlets covered it collectively, but their stances differ subtly. Here is a table of the verified key stances:

Layer three: the official posture. This is currently the clearest layer: Tom’s Hardware sought comment from NVIDIA, and NVIDIA did not respond — consistent with its standing practice of not commenting on rumors; NVIDIA’s website still lists the RTX 5090 for sale. Zero response plus on-sale-on-site means that officially, it has been neither confirmed nor denied. Any inference that reads “the site still lists it” as “the rumor collapsed,” or “no response” as “tacit admission,” is over-interpretation.

Stacking the three layers: this is a rumor with heat, with plausibility, and without hard evidence. The business logic holds (Section 3 does the math), and outlets were willing to report it, but the final link in the evidence chain — official confirmation — does not exist. The practical advice for ordinary readers: you may adjust your decision weights per the rumor, but do not act as if it were established fact.

3. Business Logic: One GB202 Die, a Several-Fold Profit Gap#

Why are so many people willing to believe this rumor? Because the math works. GB202 is NVIDIA’s largest Blackwell-family chip; the RTX 5090 pairs it with 21,760 CUDA cores and 32GB of GDDR7; the same die also powers RTX PRO workstation cards. The rumor is not that GB202 is being retired, but that the die’s flow is being rerouted wholesale — note the framing: the rumor speaks only of “rerouting,” never of “expanding output”; any derivative claim that “NVIDIA is increasing GB202 production” has no basis.

First, put the sibling products’ specs side by side and focus on the spread in the memory column:

Two notes: first, the RTX PRO 5500 and the RTX 5090 are the same die scale (both 21,760 CUDA) — the same silicon potential, with different value extracted on two product lines. Second, HotHardware’s original text wrote the PRO 6000’s memory as GDDR6, a typo in that piece; this article goes with GDDR7 (consistent with 4Gamers and other sources). The 96 / 84 / 72-or-48GB memory tiers for the PRO 6000 / 5500 / 5000 come from 4Gamers’ compilation.

Now the prices — this is the rumor’s engine. Launch pricing versus current market conditions:

One wafer, cut into an RTX 5090, launches at $1,999; cut into an RTX PRO 6000, it launches at $8,565 — a factory-price gap of more than four times. And the top Gold edition has been scalped past $16,000, eight times the consumer card. HotHardware’s October Best Buy spot check added a retail footnote: RTX 5090s from nearly all board partners and NVIDIA’s own FE edition are sold out, with the four in-stock cards priced from $8,044.99 up past $9,000 — far above the $1,999 launch MSRP. In other words, even if NVIDIA never actively reroutes, the actual profit headroom the consumer channel leaves for GB202 already cannot compare with the workstation line.

On the demand side there is a second push: per artvoice, AI customers are buying workstation cards in bulk, and the RTX PRO 5500 is a recently released new workstation product. On one side, big AI customers willing to place bulk orders; on the other, scattered, price-sensitive consumer gamers steadily driven away by scalping — where the chips flow is not a hard commercial call.

And this script has precedent. Per HotHardware, Micron previously announced its exit from the Crucial consumer storage business, ceding capacity to data-center customers (with an official statement from that business’s head, Sumit Sadana), and Micron’s stock surged after the news. Media outlets widely cite this case as an analogy for NVIDIA’s choice: when AI demand and consumer demand fight over the same capacity, it is always the earnings logic that sends consumer products out first.

4. Product Line Restructuring Scenario: the Empty Flagship Slot and the 5080 24GB Positioning Contradiction#

If the rumor holds, what is affected is not just the 5090 but the entire top structure of the GeForce line. Assembled from the leak framing, the scenario runs as follows (a rumor-based scenario, not established fact):

• The RTX 5090 and 5090D exit the consumer market;
• The RTX 5080 24GB replaces the 16GB version as the new GeForce flagship, with the 16GB version discontinued;
• The full-fat big GB202 die henceforth appears only on Pro-badged products — The Gamer Couch’s relay put it bluntly: if the leaks hold, “the 5090 Ti doesn’t exist.”

The 5080 24GB part happens to be the weakest link in the entire rumor. VideoCardz explicitly noted that MEGAsizeGPU did not confirm the 5080 24GB claim — it exists only in hongxing2020’s leak. And what hongxing2020’s “5080 24GB” actually is has no industry consensus: the prior RTX 5080 SUPER rumor spec was a GB203 small die with 24GB of GDDR7, and per The Gamer Couch’s relay, the SUPER card was delivered to board partners in July and then shelved because GDDR7 prices spun out of control. The “5080 24GB” in the factory notice could be that shelved SUPER SKU revived, or could be a simple memory-spec bump — neither possibility can currently be confirmed.

Assuming it does arrive, pricing is also a problem. 4Gamers’ expectation framing is as follows (media expectation, not official pricing):

This surfaces the most contorted part of the rumor, a product positioning contradiction: a 24GB card on a GB203 small die is expected to price above the big-die GB202 RTX 5090’s launch price ($1,999). Gamers paying more money get a smaller core — this is not NVIDIA’s first time letting the word “flagship” float with the market, but when a small die occupies the flagship slot, the gap between the “GeForce flagship” name and the “Pro-exclusive big silicon” reality gets laid on the table. As for when NVIDIA will formally put these changes on the table, 4Gamers’ projected window is: as early as the end of this month, or at CES 2027 (January next year) — note, this too is media speculation, not an official schedule.

5. Market Impact of the Flagship Gap: Inventory, Secondhand, and the 5090D Question#

Even if the rumor is true, the market will not flip overnight on October 10. Consolidating the verified judgments from all sides, the gap period most likely takes three forms.

First, inventory drawdown takes time; the 5090 will not vanish overnight. Per Tom’s Hardware’s prediction, board partners, distributors, and retailers still hold already-built inventory, and the 5090 will not immediately evaporate from the market; but as inventory drains, availability falls and prices will be affected — note this is media extrapolation, not official framing. Layered on top of last week’s retail snapshot (early-October US street price median around $4,700, roughly 135% above launch; Amazon third-party range $6,300-$11,500; Newegg third-party $6,549-$9,300 — see the retail “extinction” piece), toward empty inventory, price has exactly one natural accelerant.

Second, a “formal EOL” may never be announced. This is the most insightful part of TechPowerUp’s stance: because the chip is merely “rerouted” rather than the product line retired, NVIDIA may well never issue any discontinuation notice — product pages stay up, channels hold scattered stock, the site still lists it for sale, but the chip supply has already been diverted. The practical implication for players: those waiting for “official confirmation” before acting may never see that day come. To judge real supply, watch channel spot stock and price trends, not announcements.

Third, the secondhand market will “jump the gun” immediately. The Gamer Couch explicitly warns: the EOL leak itself stimulates secondhand 5090 price gouging — the hotter the rumor, the more holders hoard, and the higher secondhand asking prices climb. This requires no official confirmation to happen, because it depends only on market sentiment. If the secondhand market has moved since October 10, the cause is not that discontinuation was proven, but that people believe it will be discontinued.

Finally, the 5090D question. The factory notice screenshot says the discontinuation scope includes the China-market 5090D variant, but this detail currently rests on a single hongxing2020 source with no independent second-source cross-verification. Considering that the 5090D line already went through a V2 revision this year plus that May price hike targeting the 5090 / 5090D V2 specifically (per vendor notice framing — see our October 5 piece), China’s flagship supply was already more fragile — if the rumor holds, that market’s gap will be more complete than the US market’s. But until a cross-source appears, the 5090D portion must stay suspended.

6. Q4 Build Decisions: Three Types of Buyers, Find Yourself#

Now for recommendations. The overarching principle first: in the rumor stage, the most expensive move is acting as if it were established fact — panic buying, paying premiums for secondhand, paying an emotional tax for a “last chance” — all of it is the result of treating a rumor like the evening news. The media consensus compiled by EJS Computers matches my own judgment; three buyer types, one by one:

Three execution details:

• Move your decision anchor from “official announcements” to “channels.” Section 2 noted TechPowerUp’s view that NVIDIA may never formally announce an EOL — so “wait for official confirmation” most likely never gets its signal. Two observable indicators are enough: availability of first-party spot stock at mainstream retailers (cf. HotHardware’s Best Buy spot-check method), and the slope of secondhand asking prices.
• Budget across two lines, not one. Q4 builds are not seeing only GPU inflation — I broke down the memory line in the DDR5 build decision guide, where 32GB became the new mainstream while prices doubled, and the DDR5 supply pause warning noted even contract quotes had been suspended. With GPUs and memory both on an uptrend, the sequencing question of “which to buy first” deserves more attention than “whether to buy” — the general rule is to lock in first whichever component has the steeper price slope and is untouched by the rumor.
• A time anchor for the “wait for next gen” crowd. If you choose to skip the 5080 24GB and wait for the next generation, 4Gamers’ projected announcement window (end of this month or CES 2027) can serve as an observation point — but it is only speculation. We have previously discussed next-gen consumer timing in our RTX 60 delay analysis, so I won’t repeat it here.

One line summarizing the decision matrix: mid-range buys as usual, flagships wait cautiously, need-driven buyers watch spot stock, and nobody panics.

7. Risk Notes: Rumor Grading and Data Timeliness#

As always, let’s lay out the quality of the evidence before closing:

• Rumor is not fact: whether the RTX 5090 / 5090D has truly ended production has no official NVIDIA confirmation — only two leaker posts and one suspected factory notice screenshot; screenshots can be faked (The Gamer Couch’s exact warning: “Notices get faked”). Every “end of production” statement in this article is in rumor framing.
• Single-source limits: the 5090D discontinuation detail comes from the single hongxing2020 screenshot with no independent cross-verification; the 5080 24GB part lacks MEGAsizeGPU’s endorsement, and its relationship to the 5080 SUPER rumor (GB203 + 24GB GDDR7, shelved in July) is unsettled.
• Expectations are not pricing: the 5080 24GB’s $2,000-$2,500 is a 4Gamers media expectation; the end-of-this-month or CES 2027 announcement window is media speculation. Neither is official information.
• “Rerouting” is not “expanding output”: the rumor concerns only GB202 chip flow shifting to the workstation line; there is no basis whatsoever for “NVIDIA increasing GB202 output,” and derivative claims should all be dismissed.
• Price timeliness: the Best Buy four-card $8,044.99-$9,000+ figures are HotHardware’s October spot check; the US street price median of $4,700, Amazon’s $6,300-$11,500, and Newegg third-party’s $6,549-$9,300 are all early-October snapshots (October 5 retrieval) — on purchase day, go by the live listing. Taiwan market figures (5090 ~NT$250,000; Gold-edition PRO 6000 ~NT$520,000) are 4Gamers reporting.
• No timetable for inventory drawdown: when built inventory runs out and when prices rise again are both media projections with no official data.

Final Words#

From the GPU and storage ledger in late September, to the October 5 retail-side “extinction” piece, to today’s end-of-production rumor — reading the three together, my own feeling is: the RTX 5090’s story is turning from “an expensive card” into “a disappearing product slot”. Out-of-control pricing can still be called market behavior; chip rerouting is a product decision — if GB202 really does all end up wearing the Pro badge, then GeForce’s flagship is henceforth just a small-die card priced above the old flagship’s launch MSRP. The “gaming flagship” concept that gamers’ real money voted into existence over twenty years — faced with AI’s purchase orders, this is about the weight it can command.

But back to the act of building a PC itself, and my advice is actually simpler than a month ago: rumors cannot change demand, and demand is what decides what to buy. The 1080p and 2K world keeps turning — buy the mid-range card when you need it; if you truly need 32GB of VRAM, watch inventory, price the risk, and don’t pay the rumor a premium; and for the wait-for-a-flagship crowd — every extra month your old card holds on is the most silent veto of this pricing regime cast.

I’m Zixing — see you in the next hands-on piece.