RTX 5090 prices blow past $5,000 as AI memory shortage devastates GPU market

[Rogue Di]

Nvidias RTX 5090 has now blown past 5,000 dollars in some markets, and the whole GPU pricing situation has gotten genuinely severe enough that Nvidia is reportedly telling add in board partners it wont be releasing any new consumer GPU generation at all in 2026

Tom's Hardware reported this week that South Korean retailers, citing Danawa and local distributors, are now pricing the RTX 5090 at up to 7.3 million won, which works out to roughly 5,112 dollars, thats up as much as 1,050 dollars in a single month alone, and this comes after the card had already been listing for 4,329 dollars on Amazon in the US as of July, a staggering jump from its original 1,999 dollar launch price just last year

The root cause traces back to the same AI driven memory crunch thats been rippling through basically every corner of consumer electronics lately, VRAM now accounts for as much as 80 percent of a high end graphics cards total build cost, and AI data centers are absorbing such a huge share of global memory supply that Nvidia reportedly cut RTX 50 series production at its add in card partners by anywhere from 20 to 40 percent through the first half of the year

A planned RTX 50 Super refresh, which would have added an RTX 5070 Super, 5070 Ti Super and 5080 Super with higher VRAM configurations, has now been indefinitely delayed, and the reason is genuinely ironic, the exact 3GB per chip GDDR7 memory modules needed for that refresh are the very component already in shortest supply, so the product that could have eased the shortage requires the exact scarce ingredient causing it in the first place

Nvidia even released collectible paper trading cards back in July celebrating GeForce history, which the internet immediately mocked with the line fourteen cards, zero silicon, capturing pretty well how tone deaf that gesture felt given how many actual gamers cant get their hands on real hardware right now

Industry analysts say this isnt like the 2021 crypto mining shortage where availability itself was the core problem, this time the structural driver is genuinely different, AI is currently consuming around 20 percent of global wafer capacity, and real relief isnt expected until 2028 when Micron brings new HBM manufacturing capacity online, meaning gamers and PC builders are likely looking at a multi year stretch of elevated prices rather than a temporary blip

Gaming GPU Prices 2026: RTX 5090 Tops $5,000tech-insider.org

[Gary90]

Fourteen cards, zero silicon is such a perfect summary of how out of touch that trading card gesture felt, Nvidia clearly misjudged the mood of a community that just wants actual graphics cards at reasonable prices

[GhostRider]

The refresh needing the exact scarce memory component that caused the shortage in the first place is such a brutal catch 22, theres literally no way to ease the shortage with more supply because the new supply needs the same bottlenecked ingredient

[IronFist56]

No relief until 2028 when Micron's new HBM facility comes online is a genuinely long runway for gamers to just accept these prices as the new normal, thats multiple GPU generations worth of time at elevated costs

[RayOfLight]

80 percent of a high end cards build cost being memory alone really illustrates how much the actual GPU die itself has become almost secondary to the memory attached to it in terms of what determines final pricing

[NightHarbour91]

This is going to accelerate cloud gaming and game streaming adoption pretty significantly I think, if buying hardware costs this much more people are just going to rent compute instead rather than own an increasingly unaffordable GPU

[Evelyn]

Nvidia prioritizing higher margin AI accelerators over consumer gaming GPUs makes complete business sense from their perspective even though its genuinely painful for gamers, theyre going to chase the more profitable market every time given the choice

[HollywoodHogan]

There is a bit of a distinction between the official price and what a card is selling for through resellers, though. Nvidia is not necessarily collecting $5,000 for each 5090; a lot of that money is being captured by distributors, scalpers, and stores taking advantage of desperation. That does not make the situation less awful, but it does matter when assigning blame.

Still, Nvidia cannot wash its hands of the whole thing. If the company knows the supply chain is constrained and keeps presenting a flagship as the must-have gaming product, it also has a responsibility to set realistic expectations. A paper launch followed by months of artificial scarcity creates the perfect environment for price manipulation.

The sensible move is to refuse the premium. Every overpriced card that sits unsold is a tiny signal to the market. Sadly, there are always enough people with deep pockets to keep the frenzy alive, so I expect a lot of "temporary" pricing to become the new normal

[CMPunk]

This is where I disagree with the people saying gamers should simply buy a previous generation card. That advice worked when the older models were discounted properly. In a lot of places, the last-generation high-end cards are also inflated because retailers know buyers are desperate for anything with enough VRAM.

The memory issue is especially important. More raw compute does not help much if the card cannot hold the textures, models, or ray-tracing data modern software wants to throw at it. Developers are already designing around enormous memory budgets, and AI workloads have made high-capacity memory the most valuable part of the package. The silicon may be impressive, but the memory supply decides how many cards actually reach shelves.

At this point, a midrange card with a decent memory bus and sensible power draw looks far more appealing than a flagship trophy. Games are still playable when you turn down reflections; electricity bills and rent are less flexible. A shocking concept, I know

[Sam92]

The specific South Korean pricing data point is a good real time signal since that market tends to reflect global supply pressure pretty quickly, if its already hitting five figures in local currency there its probably a preview of where other markets are headed too

[Coder53]

Would love to see more competition from AMD or Intel specifically targeting an affordable mid range tier during this shortage, theres a real market opportunity for whoever can actually deliver reasonable pricing while Nvidia chases AI margins instead

[Kieran88]

A different concern is what this does to game development. If studios start assuming that a huge chunk of the audience owns hardware with ridiculous memory capacity, optimisation becomes optional. Then people with perfectly good systems get told to upgrade because a poorly managed asset pipeline is being disguised as progress.

There is a similar danger with AI features. Developers can use upscaling and frame generation to make a demanding game appear efficient, while the underlying workload keeps getting heavier. Those tools are useful when they supplement a well-optimised renderer, but they should not become a permission slip for shipping games that crawl without them.

Maybe the upside is that this pricing disaster forces buyers to think about what they actually need. A stable 90 frames per second at a sensible resolution is enough for many people. The industry has spent years turning benchmarks into status contests, and now the scoreboard costs more than the PC it is attached to. That is a pretty effective way to make me appreciate medium settings

[Hollow Coder]

The funniest part is watching people argue that a $5,000 GPU is fine because it can generate frames with AI. We have reached the stage where the solution to needing more performance is to spend the price of a complete workstation on a card that invents some of the frames anyway. Somewhere, a marketing executive is looking at this and thinking, "Yes, this is sustainable."

For actual gaming, the gains are not equally visible in every title. If you are playing competitive games at 1440p, you are likely CPU-limited long before the 5090 earns its keep. If you are doing 4K path tracing, rendering, or local model work, then the extra horsepower can be meaningful, but that is a much narrower audience than the launch hype suggests.

A lot of buyers are also underestimating the rest of the build. A flagship card needs a strong power supply, good airflow, a capable CPU, and a case that can physically swallow it. Spending five grand on the GPU and then discovering the case panel will not close would be a spectacularly expensive comedy sketch

[PaleCipher]

At $5,000, the 5090 stops being a graphics card and starts being a small used-car deposit. Even if someone has the income for it, the value proposition is completely upside down for gaming. You could build a very respectable entire PC for less than the markup on some of these cards, and it would still run pretty much every current game beautifully.

The frustrating part is that this is not just normal launch-day gouging anymore. A limited supply and a few scalpers can explain a nasty month, but prices staying at absurd levels because memory vendors are prioritising AI accelerators feels like a structural problem. Gamers are effectively bidding against companies that can justify a GPU by the revenue it generates, which is a contest retail buyers cannot win.

My 3080 is looking more attractive every day. It may not push path tracing at 4K with every slider maxed, but it plays the games I actually own without requiring a second mortgage. The best upgrade strategy might be to wait, lower a few settings, and let the market have its tantrum

[WildManJericho80]

The refresh needing the same scarce memory component is the part that makes this feel less like a normal product cycle and more like a supply-chain loop with no exit. Usually a refresh gives manufacturers a chance to change the bill of materials, improve yields, or shift to a different memory configuration. If the bottleneck is the exact thing the refresh depends on, then more demand just makes the shortage worse.

That could push board partners toward genuinely strange compromises. We might see lower-memory variants, regional allocations, or cards quietly prioritised for professional systems because those customers can pay more. None of those options is attractive to gamers, but they are predictable outcomes when every available memory package has a higher-margin destination.

The only real escape is demand destruction or a serious capacity increase, and neither happens overnight. Until then, the most powerful upgrade is patience, followed closely by the ability to close a browser tab before buying something ridiculous. My wallet is running at an impressive frame rate right now

[StoneCold]

Holy batman!

[ShawnMichaels_Botch]

People keep bringing up the crypto boom as if it proves GPU prices will inevitably crash again, but the situations are not identical. Crypto demand could evaporate rapidly when profitability changed. AI demand is tied to companies building data centres, selling services, and chasing strategic advantages, so the buyers have more capital and a stronger reason to keep purchasing.

That does not mean prices can stay at five grand forever. Hardware eventually gets replaced, competitors respond, and even AI companies have budgets. It does mean the correction may be slower and more uneven than gamers are hoping for. Consumer cards could remain squeezed while the best memory and packaging capacity gets reserved for enterprise products.

Personally, the sensible play is to skip the halo tier and watch the second-hand market carefully. There will eventually be people who bought an expensive card for bragging rights, barely used it, and then discover that their hobbies have changed. Just check warranty coverage, power connectors, and whether the seller is actually offering a GPU rather than a very fancy cardboard box