Why Are RTX 5090 & 5080 Still Out of Stock?
A year and a half after launch, the RTX 5090 sells for more than double its list price and the 5080 carries a 40% premium. This is not scalpers alone. Here is what is actually going on.
The numbers first, from our own August 2026 GPU Price Index: 16 of the 17 priority GPUs we track sell above MSRP, and the tracked panel averages 26.0% over list, up from 10.4% in July. The cheapest in-stock RTX 5090 we could find was $4,299.99 against a $1,999 list price. The cheapest RTX 5080 was $1,399.99 against $999. Finding either card "at MSRP" is, for practical purposes, not a thing that happens in August 2026.
Several causes stack on top of each other, and the biggest one has nothing to do with gamers at all.
1. AI Datacenters Eat First
Nvidia's datacenter business now dwarfs its gaming business, and both compete for the same inputs: advanced-node wafer capacity, packaging, and above all memory. Every wafer and every gigabyte of high-bandwidth memory allocated to AI accelerators is capacity that does not become a GeForce card. When one customer segment pays datacenter margins, consumer allocation becomes the leftover, and in 2026 the leftovers are thin.
This is why the shortage will not resolve the way the 2021 crypto crunch did. Crypto demand collapsed on its own. AI demand keeps setting capex records.
2. The Memory Shortage Is the Choke Point
The specific bottleneck in 2026 is DRAM. Memory makers have shifted production toward the high-bandwidth memory that AI accelerators consume, and contract prices for conventional DRAM and GDDR have risen sharply as a result. A graphics card is, in bill-of-materials terms, a slab of increasingly expensive memory attached to a GPU die. When GDDR7 module prices climb, board partners either raise card prices or stop building the card. Both are happening.
Reporting through 2026 has also described output cuts on some consumer SKUs as Nvidia and its partners steer scarce memory toward the products with the best margins. Treat the specifics as reporting rather than confirmed policy, but the street prices in our index are consistent with supply being deliberately thin at the high end.
3. The Refresh That Is Sitting in a Warehouse
The relief valve was supposed to be the RTX 50 Super refresh: a reported 5070 Super with 18GB, a 5070 Ti Super with 24GB, and a 5080 Super with 24GB, all built on 3GB GDDR7 modules. By most accounts the silicon is done and board partners have working designs. The launch is reportedly on hold anyway, because those 3GB modules are exactly the part whose price exploded. Current reporting points to a launch around CES in January 2027, and that date is a rumor, not a promise.
The practical effect for buyers: there is no new mid-cycle supply coming this year, and everyone who was waiting for the Super cards is now bidding on the same existing inventory.
4. Scalpers, Import Pricing, and the Premium Spiral
Scarcity invites the usual opportunists. Bots clear MSRP drops in seconds, and marketplace listings price against desperation rather than list price. On top of that, a meaningful share of high-end inventory reaching US retail in 2026 arrives through import channels carrying tariff and logistics costs that get passed straight through. When the only in-stock RTX 5090s are marketplace and import listings, $4,300 becomes the market price whether or not anyone thinks it is fair.
What a Buyer Can Actually Do
- Consider last-gen seriously. A used RTX 4090 at around $2,073 on eBay is less than half the price of a new 5090 and still a monster. The RX 6800 XT was the only card in our August panel selling below MSRP.
- Buy the dips, not the hype. Prices move in windows. July was meaningfully cheaper than August across almost the whole panel; the market does produce better weeks, and patience is a real strategy.
- Use stock alerts and drop trackers. MSRP stock still appears in small bursts at Best Buy, Micro Center, and direct from board partners. Being early matters more than being persistent.
- Check prices before you assume. Our RTX 5090 and RTX 5080 pages track live retailer prices, so you can see what today's floor actually is instead of anchoring on MSRP.
Bottom Line
The RTX 5090 and 5080 are hard to buy at list because the memory they are built from is worth more to AI datacenters than to gamers, and the mid-cycle refresh that would have added supply is reportedly delayed to around CES 2027 by that same memory economics. Scalpers and import premiums amplify the problem, but they did not create it.
Until memory supply loosens, the winning moves are patience, alerts, and a willingness to look one generation back. The monthly GPU Price Index is the fastest way to see when the tide actually turns.