Nvidia's GeForce chief says the RTX 5090 is not being killed off. He also confirmed why nobody can buy one at list price. If you are budgeting a 32 GB card for local inference, plan around street prices, not the $1,999 MSRP.
What Nvidia actually said
Jeff Fisher, senior vice president of Nvidia's personal computing business, spoke to Taiwanese media at the RTX Spark launch event in Taipei on October 16. According to the Startup Fortune write-up, which cites Tom's Guide, he said reports of the RTX 5090's discontinuation are not true and the card will keep shipping.
He also said supply chain capacity is "indeed somewhat constrained", even as Nvidia is "going all out to bring products to market".
That is a promise of continued existence. It is not a promise of availability at any particular price.
Why people believed it was dead
Two things fed the rumor, both unconfirmed by Nvidia:
- Micro Center marked numerous RTX 5090 listings from Asus, MSI, Gigabyte, PNY and Zotac as "not carried", per Club386 and Overclock3D. The source describes that as language the retailer uses for products it does not plan to restock.
- Leakers MEGAsizeGPU and hongxing2020 claimed a factory notice halted GB202 production for GeForce, redirecting silicon to the professional RTX Pro line. They also claimed a 24GB RTX 5080 would become the top consumer card. Nvidia never confirmed this, and Fisher's comments say the opposite.
Treat the leak as unverified. The price data below is not.
The price numbers
| Data point | Price | Source |
|---|---|---|
| Launch MSRP (January 2025) | $1,999 | Startup Fortune |
| Cheapest new unit, September | $4,999.99 (about 150% over MSRP) | select.keenable.ai, US retail listings |
| Tom's Hardware tracker | past $6,000, more than triple MSRP | Tom's Hardware, via Startup Fortune |
| Peak listing during the discontinuation panic | $8,500 on Amazon | Tom's Hardware |
The $8,500 figure is a peak, not a typical price. The $4,999.99 September floor is the number I would plan around. The source says these are listings at mainstream retailers, not fringe marketplaces.
Why it is a memory story
The source ties the shortage to memory, not to the GPU die:
- IDC estimates put AI data centers at roughly 70% of world memory output in 2026, up from 20 to 30% in 2022.
- SK Hynix has confirmed its DRAM, NAND and HBM production is sold out through the rest of 2026.
- Micron's CEO said this month that memory supply will stay tight through 2028.
- The GDDR7 on an RTX 5090 competes for wafer allocation with the HBM going into Nvidia's own data center chips.
- TechSpot's trackers show GPU prices broadly up around 15% in recent months, with Nvidia cards among the hardest hit.
The article also mentions a seventeen-plaintiff antitrust suit filed in June against Samsung, SK Hynix and Micron. It alleges coordinated DRAM supply restriction and a price rise of roughly 700% over four years. Those are allegations, not findings.
What I would do
I would not pay $5,000 for 32 GB of VRAM to run inference. At that price the card costs more than twice its list price, and nothing in the source suggests relief before 2028 if Micron's timeline holds.
If you already own a 5090, keep it. Resale value is probably better than at any point since launch, though the source gives no resale data.
If you need memory capacity rather than raw CUDA throughput, look at unified-memory machines first. I covered one in the RTX Spark N1X laptop benchmark post. For the earlier round of RTX 50 pricing, see the June RTX 50 price hike post.
Tradeoffs and unknowns:
- The source gives no availability data beyond "constrained". It cites no lead times.
- Regional prices outside the US are not stated.
- Whether Micro Center restocks is unknown.
Verify it yourself in 15 minutes
- Check the Micro Center product pages for 5090 models from Asus, MSI, Gigabyte, PNY and Zotac and see whether they read "not carried".
- Compare against the Tom's Hardware price tracker.
- Search new, in-stock 5090 listings at your regional retailers and compare the cheapest against $1,999.
Sources: