Key Takeaways
- The NVIDIA RTX 5090 launched at $1,999 MSRP but is now selling for $6,699 and above from third-party sellers — a 235%+ markup.
- Photos from Hong Kong-based outlet HKEPC show pallets of RTX 5090 cards being unpacked at AI server assembly sites in China.
- Chinese vendors are buying the standard international RTX 5090 in bulk to build AI workstations, bypassing US export restrictions on the region-specific RTX 5090D.
- The GDDR7 memory shortage is the root cause — AI data centers are projected to consume roughly 70% of global high-end DRAM output in 2026.
- The RTX 5090 carries 3,352 AI TOPS, making it genuinely useful for AI inference, especially compared to the downgraded RTX 5090D at 2,375 AI TOPS.
- RTX 50-series cards have climbed as much as 39% across the board since June 2026, with the RTX 5090 up over 41% from February prices.
- NVIDIA has not officially commented on RTX 5090 scarcity. A third round of Nvidia price hikes of up to 30% is reportedly being planned.
- Budget options like the relaunched RTX 3060 (12GB GDDR6) still exist but have also climbed to $450+ from an initial $339 relaunch price.
The RTX 5090 Is Vanishing — and It’s Not What You Think
If you’ve been trying to pick up an NVIDIA GeForce RTX 5090 lately and kept running into walls, you’re not imagining things. The card is genuinely getting harder to find, and not because there’s some random supply hiccup or a shipment got delayed at a port. The real story is a lot more annoying: AI companies are literally buying these gaming cards by the pallet and jamming them into server racks.
Images shared by Hong Kong-based media outlet HKEPC went viral in the hardware community after they showed boxes — full retail boxes, the kind with the GPU inside that’s supposed to end up in your gaming rig — stacked on pallets at what appear to be AI server assembly facilities. We’re not talking about purpose-built data center GPUs here. These are ASUS ROG, MSI, and Gigabyte branded RTX 5090 cards. The exact same products you and I would be shopping for at Newegg.
This isn’t the first time we’ve seen something like this. If you were around during the crypto boom of 2020–2021, this rhymes hard. Mining companies did the exact same thing — swept up consumer gaming cards in bulk because the compute performance-to-availability ratio made more sense than chasing professional-grade hardware. NVIDIA eventually responded with LHR (Lite Hash Rate) versions of several RTX 30-series cards that throttled mining performance. Those restrictions got mostly bypassed within a year, but the intent was clear. The question now is whether NVIDIA does something similar for AI workloads — and so far, they’re not saying a word.
Why Are AI Companies Using Gaming Cards in the First Place?
Here’s the part that makes total sense once you think about it for a second. NVIDIA’s dedicated data center accelerators — the H100, H200, and newer B200 chips — are supply-constrained to the point of absurdity. Getting allocation on those parts as a smaller AI startup or company is borderline impossible. The waiting lists are long, the prices are astronomical, and the procurement process is not built for small-to-medium operations.
The RTX 5090, on the other hand, is a consumer product that sits on retail shelves (or at least, used to). It packs 32GB of GDDR7 video memory and delivers around 3,352 AI TOPS — that’s legitimate inference performance. For a startup running local model fine-tuning or smaller-scale AI tasks, a rack of RTX 5090s isn’t an H200, but it gets the job done for a fraction of the enterprise price. Even at $5,000–$6,000 a card, it still undercuts what dedicated AI silicon costs when you factor in availability.
And then there’s the export restriction angle, which makes this whole situation even messier. The US government banned the sale of the standard RTX 5090 to China. In response, NVIDIA created the RTX 5090D — a modified version with hardware-level AI inference limitations, capped at around 2,375 AI TOPS compared to the full card’s 3,352. That 5090D has since been hit with its own sales ban in China after it was found to potentially exceed US bandwidth thresholds anyway. So Chinese vendors are now sourcing the unrestricted, full-performance, international RTX 5090 instead — the same card gamers everywhere else are trying to buy.
The GDDR7 Shortage Is the Gasoline on This Fire
Here’s the part that really stings if you’re a PC gamer. Even if AI demand cooled off tomorrow, you’d still be dealing with a GPU pricing nightmare right now because of the GDDR7 and DRAM shortage sitting underneath all of this.
IDC has projected that AI data centers will absorb roughly 70% of global high-end DRAM output in 2026 — up from somewhere between 20–30% just a few years ago. The major DRAM manufacturers, Samsung and SK Hynix in particular, are running at the limits of their production capacity. And when a company can sell a memory module to a data center at one price versus what a consumer GPU board partner pays, you can guess who wins that negotiation every single time.
GDDR7 modules — the specific type of memory that goes inside the RTX 5090’s 32GB stack — compete for fabrication capacity with HBM and server DRAM. That same bottleneck is now showing up directly on every GPU price tag. RTX 5090 prices rose 20% in a single month alone and are now sitting more than 41% higher than they were back in February 2026. The entire RTX 50-series lineup has climbed as much as 39% since June. The RTX 5070, a mid-range card, is up 36%. This isn’t just a high-end problem.
And there’s reportedly another wave coming. Reports out of Taiwan point to a third NVIDIA price hike of up to 30% hitting GeForce RTX cards before the end of the year. If that lands, we’re in really uncomfortable territory for anyone trying to upgrade their gaming PC in 2026.
What Are You Actually Paying for an RTX 5090 Right Now?
Let’s get concrete. The RTX 5090 launched at $1,999 MSRP. That’s already a premium price for a flagship card, but it was at least a somewhat reasonable ask for what the GPU delivers — 32GB of GDDR7, top-of-the-line rasterization performance, and strong ray tracing numbers for 4K gaming.
As of right now, you can’t buy one directly from NVIDIA, Amazon, or Newegg at retail. You’re strictly looking at third-party sellers. The lowest price currently listed from those sellers is sitting at $6,699 — a 235% markup before you factor in taxes and fees. Some listings have been spotted approaching $9,800 with international shipping. In Europe, prices have crossed €5,200. This is a card that’s functionally out of reach for the vast majority of PC gamers who might’ve been saving up for an upgrade.
By comparison, the RTX 5080 is still available directly from Amazon at $1,599. That’s a meaningful performance step down from the 5090 but still an extremely capable card for 4K gaming and creative workloads. If you were on the fence about whether you needed the extra headroom of the 5090, the current market is making that decision for you.
If you’re building or upgrading a gaming setup right now and the GPU situation is giving you a headache, it might be worth revisiting the rest of your peripheral game while you wait it out. A low-ping gaming router won’t fix GPU prices but it absolutely helps your online experience, and it’s actually possible to buy one at reasonable prices right now.
Budget Alternatives: The RTX 3060 Is Back (But Also More Expensive Now)
If the flagship GPU market has priced you out entirely, NVIDIA did quietly relaunch the GeForce RTX 3060 earlier in 2026. This is one of the most widely used graphics cards in the world — the kind of card that powered 1080p gaming for millions of players across its original lifecycle. The relaunched version keeps the 12GB of GDDR6 memory configuration, making it a solid pick for 1080p gaming and light creative work.
The catch is that even the 3060 hasn’t escaped the broader pricing pressure. The relaunch initially priced these cards at $339. They’ve since crept up to $450 or higher across various retailers. That’s still a far cry from $6,700, but it represents the same upward pressure affecting the whole market. Promotional deals can bring the price back down, so it’s worth keeping an eye on sales if that’s the tier you’re targeting.
If you’re in that budget-conscious upgrade space, pairing an older GPU with the right peripherals can stretch your setup further than you’d think. A solid wireless gaming headset under $100 and a reliable PC gaming controller can round out your rig while you wait for the GPU market to stabilize.
The Crypto Boom Parallel — and Why It Matters
This whole situation draws a direct line back to 2020–2021, and the parallel is uncomfortable. During that era, Ethereum mining operations bought up GeForce cards in bulk. RTX 3080s and 3090s that launched at $699–$1,499 were routinely selling for two to three times their MSRP. It took nearly two years for the market to normalize after mining profitability collapsed.
The key difference this time is the underlying driver. Crypto mining demand evaporated when Ethereum switched to Proof of Stake. AI demand isn’t going anywhere. If anything, it’s accelerating. The scale of LLM infrastructure being built across cloud providers and enterprise data centers is only growing, and the GPU appetite that comes with it isn’t slowing down. That makes this shortage structurally different — and potentially longer-lasting.
NVIDIA does have a toolkit here if they choose to use it. They’ve already proven they can reduce AI compute capability through hardware-level restrictions — that’s exactly what the RTX 5090D was. An AI-limited version of the standard RTX 5090 for western markets is technically feasible. But NVIDIA hasn’t moved in that direction, and given how central AI revenue is to the company’s growth story right now, it’s unclear they have a strong incentive to.
What About Monitor Upgrades in the Meantime?
If you’re currently gaming on a 1080p or 1440p screen and were thinking about going 4K once you grabbed an RTX 5090, the GPU situation may have given you an accidental reprieve to think that through more carefully. Worth knowing: you don’t actually need a top-end GPU to benefit from a better display. Check out whether a 144Hz monitor makes sense for your setup — the answer might surprise you depending on what you’re playing.
Where Does This Leave PC Gamers?
Honestly, in a tough spot, and there’s no clean way to sugar-coat that. The RTX 5090 was supposed to be the dream 4K gaming card for this generation — and it still is, technically. The hardware itself is excellent. The problem isn’t what the card can do. The problem is that it’s now functionally inaccessible to anyone who isn’t a scalper or an AI company.
NVIDIA holds near-complete control of the gaming GPU market. AMD has been raising prices too, but by a smaller margin — around 10% on select Radeon cards — and Intel’s Arc lineup doesn’t compete at the high end this generation. There’s no real market pressure forcing NVIDIA to protect consumer supply. Their data center business is printing money, and that business has the exact same interest in acquiring the same silicon that gamers want.
Until the GDDR7 supply situation stabilizes, or until NVIDIA steps in with some form of intervention like the LHR approach from the crypto era, the RTX 5090 shortage is likely to persist. For most PC gamers right now, the RTX 5080 at MSRP is the realistic ceiling — and even that window may not stay open forever if the broader memory market keeps tightening.
Keep your alerts set, watch for promotional pricing on mid-range cards, and maybe hold off on that flagship upgrade unless you’re prepared to pay what the market is asking. This one’s going to take time to sort itself out.
