If you've tried to buy an RTX 5090 lately, you've seen the prices. Here's how it got here, with sources.
The launch price vs now
NVIDIA announced the RTX 5090 at CES 2025 for $1,999, available Jan 30, 2025, with 32GB of GDDR7.
On Oct 17, 2026, Tom's Hardware checked retailers:
- Amazon: about $8,500, from third-party sellers.
- Newegg: about $7,300, likely from overseas sellers.
- Best Buy: none.
- NVIDIA's own store and Micro Center: not listed.
PC Guide tracked specific models over the past month. One ASUS ROG Astral model was listed at $6,899.99 on Sept 14 and $8,979 in its Oct 19 column. That's well past "3x MSRP"; it's over 4x.
Is NVIDIA killing it?
Rumours said NVIDIA might stop making the 5090 to save its GB202 chip for pro cards. NVIDIA says no. Senior VP Jeff Fisher told 4Gamers, as reported by PC Guide:
"The supply chain (capacity) is indeed somewhat constrained, but NVIDIA is doing everything it can to get the product on the market... production has not been discontinued, and NVIDIA will continue to ship."
Why it's so tight
Three things stack up.
1. AI buyers want it. 32GB of fast memory makes the 5090 useful for running local models. PC Guide reports AI firms are buying it for server use, which pulls stock away from gamers. (How much memory a model needs is in our VRAM guide.)
2. GDDR7 is scarce. TrendForce's Sept 30 outlook says GDDR7 output remains limited as advanced capacity shifts to server DRAM, and graphics DRAM prices should keep rising.
3. One supplier is reportedly stepping back. TrendForce News (Sept 24, citing other reports) says Micron has reportedly moved its 2GB GDDR7 chips to end-of-life, leaving 3GB modules. Samsung and SK hynix still supply most RTX 50 memory, so this isn't an instant shortage, but it removes an option. The same report cites estimates that 3GB GDDR7 chips cost about $60–$70 each, roughly triple a 2GB chip.
Same report, flagged as market chatter: the next-gen RTX 60 family may have slipped from 2H27 into 2028. Treat that as a rumour.
What to do instead
- Don't pay scalper prices without a plan. For AI work, renting a GPU by the hour is often cheaper than buying at 4x. Our GPU cloud price tracker helps compare.
- Look at unified memory machines. A Mac Studio or RTX Spark laptop holds bigger models, at lower speed.
- Understand the memory. GDDR7 vs GDDR6 vs HBM explains why graphics memory is caught up in the AI boom.
