Nvidia denies the RTX 5090 is discontinued while conceding supply is constrained

Nvidia’s senior vice president of personal computing, Jeff Fisher, said in Taipei on October 16 that the GeForce RTX 5090 has not been discontinued and that the company will keep shipping it. The remark came in a post-event interview after the launch party for RTX Spark PCs. Taiwanese outlet 4Gamers posted it at 9 p.m. Taipei time; Taiwan’s Central News Agency filed at 21:55 and updated at 22:06. A crawl of Nvidia’s newsroom index on October 18 found no occurrence of “5090” across its first three pages, which run from October 17 back to September 30, so what exists today is a spoken answer relayed by Taiwanese media, with no matching written statement.

Key points

• Fisher’s quoted answer, per 4Gamers (translated from the Chinese): “The supply chain (capacity) is indeed somewhat constrained, but NVIDIA is doing everything it can to get the product on the market. Although there are other variants of the RTX 5090 (chip), production has not been discontinued, and NVIDIA will continue to ship, which is very important.” He did not explain what the other variants are.
• The card is missing from Nvidia’s own October newsroom: October 16 carried the RTX Spark PC launch, October 17 a global RTX Spark celebration, and the GeForce blog’s article list has no 5090 headline either. Tom’s Hardware wrote on October 17 that it asked Nvidia for comment and “received no reply.”
• Retail tells a different story. In Tom’s Hardware’s October 17 spot check, Amazon third-party sellers listed the card at about $8,500, Newegg at about $7,300, Best Buy had none, and the card was absent from Nvidia’s own marketplace and from Micro Center.

Background

Around October 10, hardware leaker hongxing2020 said a factory notice showed Nvidia ending RTX 5090 production, including the China-only 5090 D; leaker Zed__Wang backed the claim. Dexerto turned the leak into a report on October 12, with the industrial logic that GB202 wafers were moving to more profitable professional and data-center parts, and that a 24GB RTX 5080 might take over as flagship. This site’s October 11 digest carried the same thread under a single-source label and kept it out of the archive, noting that a lone leak could not carry an entry on its own.

The rumor left marks at retail before Fisher spoke. On October 15, Tom’s Hardware reported that the RTX 5090 had vanished from Nvidia’s marketplace and that Micro Center had reclassified it as “Not Carried,” with partner cards out of stock at every physical location. The next day Fisher took questions at the RTX Spark launch: Nvidia announced RTX Spark laptops on October 16, and its October 17 celebration post gives a starting price of $2,399, with events in Taipei, Berlin, Seoul, San Francisco, Istanbul and Beijing. Taiwan was one of the few markets with a physical party.

A second supply line runs alongside this one. Dexerto notes that Nvidia reportedly cut RTX 50 series production in February because of a global memory shortage, and this site recorded Micron’s October 1 view that the shortage lasts into 2028. The 5090 carries 32GB of GDDR7, so rising memory prices feed straight into its cost.

The facts

Nvidia’s own RTX 5090 product page was still live on October 18. Its structured data lists $1,999 with availability “InStock”; the visible buy module on the same page renders placeholder text, “Starting at $XXX.00, Available January XX,” and is not a working path to stock. Those two signals point at different things: the structured data is a template field, while the buy button reflects the real listing state. A maintained product page shows the SKU has not been deleted from the site. It is not evidence that cards are shipping.

Specs come from the same page, checked October 18: 21,760 CUDA cores, 32GB GDDR7, a 512-bit memory interface, 575W total graphics power and a recommended 1,000W system supply. Those numbers explain why two kinds of buyer watch this card at once: players want frame rates, and people running local models want 32GB of VRAM.

Channel	Price or status	Checked
Nvidia list price	$1,999	nvidia.com product page, 2026-10-18
Nvidia marketplace	Card absent	Tom’s Hardware, 2026-10-17
Amazon	About $8,500, third-party sellers	Tom’s Hardware, 2026-10-17
Newegg	About $7,300, likely Chinese sellers	Tom’s Hardware, 2026-10-17
Best Buy	No stock	Tom’s Hardware, 2026-10-17
Micro Center	“Not Carried,” stores out of stock	Tom’s Hardware, 2026-10-15
China RTX 5090 D v2	From ¥16,499	nvidia.cn product page, 2026-10-18

Price reporting diverges. In an October 17 opinion piece, TechRadar put third-party pricing near $10,000, more than a thousand dollars above Tom’s Hardware’s $8,500. Neither account includes a checkable listing snapshot, so this entry quotes the figures as a range and does not average them.

The China angle

Nvidia’s China site still showed “￥16,499 元起” on the RTX 5090 D v2 page as of October 18, returning HTTP 200 with a last-modified header of October 16, 14:13 GMT. That sits against the leak’s claim that both the 5090 D and D v2 were being cut: on the public web, at least, Nvidia has not taken the China-specific model down. Dexerto reported that the D series is sold only in China.

The comparison has limits. A product page is a marketing asset and can lag production by months. A live page shows that no delisting notice was issued; it does not show that supply has resumed. For buyers in China, ¥16,499 is the official anchor, and street pricing detaches from that anchor first when stock runs out.

What others say

Tom’s Hardware’s Bruno Ferreira framed the denial with Nvidia’s financials: gaming revenue is now a small slice of an ever-larger pie, and statements like this one read as a way to keep the brand in front of gamers. The same piece notes that Intel and AMD have also repeatedly said their consumer GPUs are not dead, that none of the three has a new GPU launch, and that their quarterly launch figures keep leaning toward AI.

Darren Allan’s October 17 opinion piece at TechRadar carries the strongest counter-reading. Citing the leaker’s claim that Nvidia is diverting all GB202 dies to RTX Pro cards, he argues the company is reallocating capacity to higher-margin products and that the RTX 5090’s de facto exit is already under way. He puts the fallback at about $1,800 for an RTX 5080 on Newegg and £1,450 at Overclockers UK, and warns that the 5080 will climb further once 5090 stock runs out. The numbers come from the author’s own browsing, not a spot check.

Dexerto’s Callum Doyle fills in two gaps: Fisher did not explain the “other variants,” and he gave no timeline for how long supply stays tight. Doyle also notes this was Nvidia’s first public response after a week of rumors. TechPowerUp’s headline pairs the denial with an admission that supply is tight; its summary credits 4Gamers.

The Central News Agency’s on-site report adds a different emphasis. Fisher described the supply crunch as a global condition that Nvidia also operates under, and tied GeForce to CUDA and then to today’s AI market: “If there were no gaming business and no GeForce, there would be no Nvidia of today” (translated from the Chinese).

Nvidia’s own results give the story its scale. The fiscal 2027 second-quarter release on August 26 reported $96.2 billion in revenue, up 106% year over year, with data center at $89.0 billion, up 117%; on those two numbers data center is about 92.5% of revenue. A GB202 die sold as a professional or data-center part earns far more than the same die sold as a gaming card, which makes the direction of any allocation shift easy to infer. It remains an inference: Nvidia has not published allocation figures.

Our take

For anyone building a local inference box, the actionable fact here is extended scarcity plus a premium. Whether the card is formally discontinued matters less. The $1,999 list price no longer clears at retail, and an $8,500 street price puts a single card in whole-system territory: this site has covered the DGX Spark at $4,999 for 64GB and $6,950 for 128GB, and the Ryzen AI Halo at $7,299.99 for 192GB. Those are complete machines. A 32GB RTX 5090 still has a role in multi-GPU builds, and at today’s premium the total-cost arithmetic is worth redoing.

For anyone waiting for prices to fall, Fisher’s answer contains no executable commitment: no date, no volume, no channel. “Doing everything it can to get the product on the market” reads as easily as a recovery next month as it does continued tightness next year. He also left “other variants” unexplained. According to the leak TechRadar repeats, the same GB202 wafers are being routed to RTX Pro cards, and Nvidia’s site lists the RTX PRO 6000 workstation edition at 96GB of GDDR7 with ECC and 600W. The phrase works as a denial and as an omission at the same time.

Three questions remain open: whether “other variants” means the 5090 D and D v2 or the professional cards that share GB202; whether Micro Center’s “Not Carried” becomes normal at other channels; and whether Nvidia follows up with anything in writing. Those answers will shape next year’s build budgets more than a single denial does.

The evidence deserves its own note. The hardest part of this story is the retail record: prices, missing listings, an empty shelf. The softest is the spoken answer. The “other variants” wording appears only in 4Gamers’ account; the denial and the admission of tight supply were also recorded independently by the Central News Agency on the day, and Tom’s Hardware notes that a second Taiwanese site, SJTV, quoted the same post-event interview. Nvidia’s newsroom carries no matching written text. The October DLSS 5 update had a first-party blog post behind it; these two stories do not carry the same evidentiary weight. The supply pressure starts further upstream, and the memory shortage running to 2028 explains the empty shelves better than any denial does.