RTX 5090 Reportedly Set to Be Discontinued, with All Good Chips Reserved for Professional Cards

This morning, multiple foreign media outlets reported that Nvidia is going to halt production of the RTX 5090. GB202 is the largest desktop GPU in the Blackwell architecture, and among consumer-grade products, only the 5090 uses it.

The source of the news is a factory notice screenshot leaked from a foundry channel, which states "Nvidia will no longer sell any RTX

💡 What You Will Learn

RTX 5090 Reportedly Set to Be Discontinued, with All Good Chips Reserved for Professional Cards This morning, multiple foreign media outlets reported that Nvidia is going to halt production of the RT

📜 Table of Contents

• Same Chip, Two Ways to Sell It
• The Bottleneck Is VRAM
• The Successor Is Nowhere in Sight
• Who Gets Hit
• It Won't Get Cheaper

RTX 5090 Reportedly Being Discontinued — All the Good Chips Are Going to Professional Cards

This morning, multiple foreign media outlets reported that NVIDIA is halting production of the RTX 5090. GB202 is the largest desktop GPU in the Blackwell architecture, and among consumer-grade cards, only the 5090 uses it.

The source is a factory notice screenshot leaked through foundry channels. The notice states: "NVIDIA will no longer sell any RTX 5090 models, including the D variant. Going forward, the highest-end GeForce model will be the RTX 5080 24GB. The RTX 5080 16GB version will be discontinued. Please be advised."

Same Chip, Two Ways to Sell It

The full GB202 has 24,064 CUDA cores, and it powers the RTX PRO 6000 Blackwell with 96GB of GDDR7. The 5090 uses a cut-down GB202-300 with 21,760 cores and 32GB of VRAM.

Same chip — put it in a PRO 6000 and it sells for five figures in USD; put it in a 5090 and the launch MSRP was $1,999. The price gap has only widened over the past two years. The PRO 6000 went from $8,565 in March 2025 to an official $16,000 this October. On the consumer side, prices long ago stopped following MSRP. On October 10, the median asking price across 88 listings on eBay was $6,758, while on Xianyu in China the median was ¥44,688 — roughly triple the MSRP of about ¥14,000.

The Bottleneck Is VRAM

Chip capacity may not be the tightest constraint — GDDR7 is. Samsung, SK Hynix, and Micron have shifted the bulk of their advanced capacity to HBM, leaving consumer-grade GDDR7 at the back of the line. A single 2GB GDDR7 module costs over $20 on the spot market, and the 5090 carries 16 of them — that's over $320 in VRAM cost alone.

NVIDIA's moves are even more direct. In the first half of 2026, it proactively cut RTX 50 consumer-grade production by 30% to 40%, and quietly raised wholesale prices to channels by about $300.

Micron was even more decisive — it outright killed its consumer-facing Crucial brand, stating plainly that AI-driven data center demand has surged and the company decided to exit the consumer business to serve major clients. Micron's stock has risen considerably since.

The Successor Is Nowhere in Sight

Rumors say the RTX 5080 24GB will replace the 5090, with speculated specs of GB203, 10,752 CUDA cores, and 24GB GDDR7. That's 50% more VRAM than the current 5080, with the same core count.

BenchLife says the target is Q1 2027, and board partners have received verbal notice but not yet development materials. I'm skeptical of that timeline. The 50 SUPER was already put on hold once due to the high cost of 3GB GDDR7.

Also worth clarifying: cutting off GB202 allocation and killing the 5090 are two different things. AICs and channels still have existing silicon in hand and can keep assembling for a while.

Who Gets Hit

For people building PCs, it's the most visible. The flagship slot will sit empty for the better part of a year, and the highest model you can buy becomes a 5080 24GB — a card whose development materials haven't even been distributed yet.

Local AI users are hit most directly — what they want is exactly the 5090's 32GB. After the PRO 6000 jumped to $16,000, many switched to buying consumer-grade 5090s to build AI workstations, since the cost per unit of VRAM is better. With GB202 fully diverted to PRO cards, this group will be competing with enterprise clients for the same supply.

In China there's an extra layer. The 5090D series was already a compute-crippled version, and the export-compliant 5090D V2 also has locked features. This has spawned a gray-market supply chain: smuggled 5090s get their cooling and PCBs reworked and stuffed into AI servers — over a thousand units are already in circulation.

It Won't Get Cheaper

Three independent channels corroborate the same direction, so this rumor has decent credibility. It's still a rumor — NVIDIA hasn't said a word. How fast it plays out remains to be seen; how long existing inventory can hold and how much GB202 board partners still have on hand — there's no public data on either.

One thing is certain: the 5090 won't get cheaper in the foreseeable future. Launch MSRP and actual transaction prices decoupled two years ago, and now the eBay median is $6,758 — a production halt will only make existing stock scarcer.

The 2021 mining boom ended overnight when Ethereum switched to PoS. There's no such switch this time.

If you're buying, be clear about whether you need frame rates or VRAM — don't rush to order just because of discontinuation rumors.