Why the RTX 5090 Is Suddenly Vanishing

Why it’s suddenly everywhere

1. 01

   On Oct 10, leaker @hongxing2020 published what appears to be a factory notice: Nvidia will no longer manufacture any RTX 5090 models, including the China-specific D variants, and fellow leaker @Zed__Wang confirmed that all future GB202 supply is RTX Pro exclusive, per Tom’s Hardware.

2. 02

   The same notice names a successor: the RTX 5080 24GB would become the highest-end GeForce card, while the current RTX 5080 16GB would be discontinued.

3. 03

   The economics explain the rumor’s credibility: the RTX Pro 5500 uses the same 21,760 CUDA cores as the 5090 but carries 84GB of memory and sells into the professional AI market, and the full RTX 6000 Pro sells for around $15,000 — several times the 5090’s price for the same silicon.

4. 04

   Retail had already priced in scarcity: the RTX 5090 was selling for roughly $4,000–5,000 in the US and past €5,000 in Germany before the rumor, more than double its $1,999 launch MSRP.

The timeline

1. 2025-01Launch and instant sellout

   Nvidia launches the RTX 5090 at $1,999. Stock vanishes immediately; scalped listings on eBay pass $6,000 within days.

2. 2025A year of structural shortage

   Prices never return to MSRP. Trackers show the card holding near $4,000 through most of the year, and German retail eventually passes €5,000, up 73% from January.

3. 2025–2026The China squeeze

   US export rules kill the original 5090 for China; Nvidia ships the cut-down 5090D, then the 5090D V2 — which Beijing adds to a customs ban list while Jensen Huang is visiting China.

4. 2026GB202 finds a richer home

   Nvidia launches the RTX Pro 5500 Blackwell with the same GB202 die, 84GB of GDDR7, and a workstation price near $8,500, alongside the 96GB RTX 6000 Pro at roughly $15,000.

5. 2026Super refresh stalls

   The rumored RTX 50 Super series, including a 5080 Super 24GB, is reported delayed as GDDR7 memory prices triple — the [AI memory supercycle](/en/topics/ai-memory-supercycle/) squeezes consumer lineups.

6. 2026-10-10The factory notice leaks

   @hongxing2020 posts a notice saying GB202 is no longer allocated to GeForce; @Zed__Wang confirms RTX Pro exclusivity. Tom’s Hardware, TechPowerUp and Guru3D report it within hours; Nvidia declines to comment.

7. 2026-10-11The debate

   The story tops hardware communities worldwide, including a front-page Zhihu thread in China, as buyers weigh whether the rumor marks the end of the halo gaming GPU era.

The backstory

A flagship that never escaped scarcity

The RTX 5090’s life began with a broken promise of abundance. Announced at CES in January 2025 at $1,999, it sold out on day one and effectively never came back into stock at that price. Within weeks, scalped cards were clearing $5,000–6,500 on eBay; a Reddit user who tracked sold listings through the first quarter found prices only drifting down toward $4,000 by spring. That level held. Twenty months after launch, the card still traded at double its sticker price — a situation without real precedent for a mainstream consumer GPU, even accounting for the pandemic-era mining crunch.

The shortage was never purely about gaming demand. The 5090’s 32GB of fast GDDR7 memory and 21,760 CUDA cores made it the cheapest ticket into serious local AI work, from fine-tuning models to running large language models at home. Small AI startups and researchers bought consumer cards by the rack because the official data-center alternatives cost an order of magnitude more. In effect, the 5090 spent its entire life competing with Nvidia’s own professional lineup — sold by Nvidia, at a fraction of the price, against Nvidia.

Why killing it makes cold financial sense

Nvidia’s fiscal 2026 results, reported earlier this year, show data-center revenue of roughly $193 billion against about $16 billion from gaming — the AI business is now close to 90% of the company. Against that backdrop, every wafer of GB202 silicon is a choice: solder it onto a $1,999 gaming card, or onto an RTX Pro 5500 that professional buyers purchase around $8,500, or an RTX 6000 Pro near $15,000. The leakers’ claim — that all future GB202 goes RTX Pro exclusive — is unverified, but it describes a decision a spreadsheet would have recommended a year ago.

There is a second, quieter pressure: memory. The AI buildout has sent GDDR7 prices surging, and Nvidia’s planned RTX 50 Super refresh was reported stuck in limbo over memory costs. Retiring the 32GB 5090 frees both silicon and scarce memory for higher-margin products, and the rumored RTX 5080 24GB successor would use less of each. The same memory crunch behind the AI memory supercycle is now visibly reshaping what a consumer GPU lineup even looks like.

The China thread: banned three times over

No product line better illustrates the geopolitical squeeze than the 5090’s Chinese odyssey. The original card fell afoul of US export rules, so Nvidia built the RTX 5090D with reduced AI throughput to comply. When regulators tightened further, a second revision — the 5090D V2 — arrived in August 2025 with trimmed memory bandwidth. Then, in a pointed piece of timing, Beijing added the 5090D V2 to its own customs ban list while CEO Jensen Huang was visiting China, as the Financial Times first reported: Washington restricted the card for being too capable, Beijing for being an American lever. If the factory notice is real, the D variants now die alongside the global model — ending a two-year saga in which Nvidia redesigned the same flagship twice for one market, only for that market to close anyway.

What happens next

Two things can be true at once: the rumor is unconfirmed, and the direction of travel is unmistakable. Board partners and retailers hold inventory, so cards will not vanish overnight — but every week without new GB202 allocation tightens supply further, and prices near $5,000 suggest the market has already begun pricing in the rumor. Nvidia’s next-generation Rubin-based RTX 60 series is not rumored before late 2027 at the earliest, which means the enthusiast market may go well over a year with the RTX 5080 24GB as its ceiling. The halo card — the absurd, oversized flagship that anchored every benchmark chart — may be taking an unscheduled intermission, because the same silicon earns three to seven times more wearing a workstation badge.

Key facts

Launch

Jan 2025, $1,999

Street price now

$4,000–5,000

GB202 cores

21,760 CUDA

RTX Pro 5500

84GB, ~$8,565

RTX 6000 Pro

96GB, ~$15,000

Rumored successor

RTX 5080 24GB

What people are saying

Common questions

Has Nvidia officially confirmed the discontinuation?

No. The claim traces to a leaked factory notice posted by leaker @hongxing2020 and confirmed by @Zed__Wang. Tom’s Hardware asked Nvidia for comment and received no response; Nvidia does not typically comment on rumors. Treat it as credible but unverified until partners or Nvidia confirm.

Will RTX 5090 cards disappear from stores immediately?

Not overnight. Board partners, distributors and retailers hold inventories of chips and finished cards, so availability will decline gradually. But if no new GB202 is allocated to GeForce, supply tightens irreversibly — and prices near $5,000 already reflect that expectation.

What replaces it for high-end gamers?

The leak names the RTX 5080 24GB as the new GeForce ceiling, with the current 5080 16GB also discontinued. A true GB202-class successor is not expected until the Rubin-based RTX 60 series, rumored for late 2027 at the earliest.

Why would Nvidia do this while the card still sells at double MSRP?

Because the same silicon earns far more elsewhere. The GB202 die powers the $8,500 RTX Pro 5500 and the $15,000 RTX 6000 Pro, and data-center products now generate close to 90% of Nvidia’s revenue. Scarce GDDR7 memory adds pressure: retiring the 32GB flagship frees memory for higher-margin products.