Graphics card shoppers hoping October would bring relief from last year’s pricing chaos are getting the opposite. New market data published this month shows GeForce RTX 50-series cards trading well above their original list prices, with the flagship RTX 5090 now fetching nearly two and a half times its sticker price on the resale market. The culprit, according to multiple hardware outlets tracking the trend, is not a shortage of gamers wanting GPUs, it is a shortage of GPUs, period, as AI infrastructure buildouts from Nvidia, AMD, and their cloud customers eat into the same silicon and memory supply that consumer cards depend on.
The numbers come from GPUPoet’s October 2026 GPU Price/Performance Rankings, a monthly market report that tracks both retail and resale pricing across dozens of current cards. The October edition, compiled from August-to-September pricing data, found that 59 of 81 tracked GPUs got more expensive month over month, while only 17 got cheaper. That is not a story about one card having a bad month. It is a market-wide trend, and it lines up with a string of hardware announcements in the first week of October that make clear why: Nvidia, AMD, and their partners are racing to put more silicon into AI racks, AI PCs, and local inference boxes, and every wafer and every gigabyte of memory that goes into those products is one that does not go into a gaming GPU.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
RTX 5090 Resale Prices Jump 23% in a Single Month
The headline number is the RTX 5090. Nvidia’s flagship carries a suggested price of $1,999, but GPUPoet’s tracking found the best resale deal for the card in September reached $4,795 — about 2.4 times MSRP, and a 23% jump from August’s best resale figure. New RTX 5090 cards sold on Amazon told a similar story: the average new-card price climbed to roughly $5,258 in September, up from about $4,262 in August. In other words, buying a brand-new RTX 5090 in September cost more, on average, than the resale premium itself would have a year ago. Retail listings for partner cards kept climbing after that: PCGuide recorded a Gigabyte AORUS RTX 5090 at $6,899.99 on September 14, 2026, and the same model at $7,055.90 on September 29, 2026.
The RTX 5090 is the extreme case, but it is not alone. The RTX 5080, which carries a $999 MSRP, saw its best September resale price hit $1,233, while new units on Amazon averaged about $1,623 in September versus roughly $1,245 in August. The RTX 5070 Ti, at a $749 MSRP, posted a best resale price of $895 in September, with new Amazon listings averaging around $1,212 compared to about $914 in August. Every RTX 50-series card GPUPoet tracked rose in price for a second consecutive month, and in five of six cases, new-card prices on Amazon rose faster than resale prices, a sign that retail supply, not just the secondhand market, is tightening.
Why Gaming GPUs Are Losing the Supply Fight to AI Hardware
None of this is happening in a vacuum. The same week GPUPoet published its pricing data, Nvidia and Microsoft held a joint event in San Francisco where Jensen Huang and Satya Nadella detailed how the two companies are co-engineering hardware and software so AI agents can run locally on Windows PCs, according to Nvidia’s newsroom. Centerpiece of that push is RTX Spark, a new Nvidia-Microsoft platform aimed at putting local AI compute directly into consumer and workstation machines. Nvidia also introduced a 64GB configuration of its DGX Spark desktop AI system, built on the GB10 platform, with a starting price of $4,999 and availability through Acer, ASUS, Dell, Gigabyte, HP, and MSI.
On the data-center side, Nvidia’s next-generation Vera Rubin NVL72 platform has already appeared in MLPerf Inference v6.1 results published on the NVIDIA Blog, meaning the generation after Blackwell is already being benchmarked for enterprise inference workloads even as Blackwell-based products are still filtering into the consumer channel. AMD is pushing a parallel track with its Helios AI Rack systems, which pack 72 MI455X GPUs per rack alongside EPYC CPUs, Pensando networking, and direct liquid cooling. HPE disclosed a $1.2 billion order from cloud provider Vultr for Helios AI Rack systems, the first order of its kind, according to a September 2026 industry summary from StorageNewsletter, underscoring how much enterprise demand is now competing for the same advanced packaging and memory capacity that feeds consumer GPU production.
AMD CEO Lisa Su addressed the supply question directly on October 6 in Taipei, telling press the company plans to substantially increase chip output to meet demand. AMD has also been promoting its own local-AI PC push: ahead of Nvidia’s RTX Spark launch, the company published Gorgon Halo AI benchmarks and said it has shipped over half a million agentic PCs to date, per Tom’s Hardware’s October 2026 news archive. Every one of those pitches, from both companies, points toward the same underlying reality: the silicon, substrate, and memory that used to go almost entirely into gaming and workstation GPUs is now being split three ways, between consumer graphics, local AI PCs, and data-center AI racks.
RTX 50-Series Pricing: MSRP vs September Resale
| Card | MSRP | Best Resale (Sept.) | Premium Over MSRP | Resale Change vs. Aug. |
|---|---|---|---|---|
| GeForce RTX 5090 | $1,999 | $4,795 | ~2.4x | +23% |
| GeForce RTX 5080 | $999 | $1,233 | ~1.2x | Rose month over month |
| GeForce RTX 5070 Ti | $749 | $895 | ~1.2x | Rose month over month |
Source: GPUPoet, October 2026 GPU Price/Performance Rankings (data compiled from August-to-September listings).
New Cards vs. Resale: The Gap Retail Buyers Actually See
The resale premium tells one part of the story, but the more painful number for ordinary buyers is what new, boxed cards cost on mainstream retail channels like Amazon. GPUPoet’s data shows new-card pricing on Amazon climbing faster than resale pricing for five of the six tracked configurations between August and September, which suggests retailers themselves are passing through higher component costs rather than scalpers alone driving the gap.
| Card | New-Card Avg. (Aug., Amazon) | New-Card Avg. (Sept., Amazon) | Month-Over-Month Change |
|---|---|---|---|
| GeForce RTX 5090 | ~$4,262 | ~$5,258 | Up |
| GeForce RTX 5080 | ~$1,245 | ~$1,623 | Up |
| GeForce RTX 5070 Ti | ~$914 | ~$1,212 | Up |
Source: GPUPoet, October 2026 GPU Price/Performance Rankings.
Historical Context: This Is the Third Major GPU Squeeze in Six Years
Anyone who shopped for a graphics card during the pandemic-era crypto boom or the initial generative-AI buildout of 2023 will recognize the pattern, even if the cause has shifted each time. The first squeeze was driven by cryptocurrency miners buying up consumer cards for proof-of-work mining. The second was driven by early large language model training runs pulling data-center GPUs away from gaming SKUs. What is different this time, based on the announcements accompanying the GPUPoet report, is that the pressure is now coming from three directions simultaneously: traditional data-center AI training and inference (Vera Rubin, Helios AI Rack), a new category of local AI desktop hardware aimed at consumers and developers (DGX Spark, RTX Spark), and ordinary consumer gaming demand that has not gone away. That three-front demand is harder to resolve with a single supply expansion, because it is not one customer segment competing with gamers, it is several, each growing at once.
Market Impact: Who Gets Hurt and Who Benefits
The immediate losers are PC gamers and small system builders who do not have the purchasing leverage of a hyperscaler. A builder planning an RTX 5090 rig in September paid roughly the price of two RTX 5080 systems’ worth of markup just in the premium over MSRP. For budget-conscious buyers eyeing the RTX 5070 Ti as a reasonable mid-range option, the jump from a sub-$900 part to one regularly listing above $1,200 pushes it into a different price bracket entirely, competing with cards that used to sit a tier above it. At the top end the asking prices are even steeper: PCGuide listed the ASUS TUF Triple Fan RTX 5090 at $8,499.99 on October 22, 2026, and reported an INNO3D RTX 5090 X3 OC listing at $11,565 in October 2026.
The winners, at least in the near term, are retailers and resellers who already hold inventory, along with Nvidia and AMD themselves, since higher average selling prices on existing silicon improve margins even without volume growth. Cloud providers like Vultr, which locked in large AI rack orders such as the $1.2 billion Helios deal with HPE, also benefit from securing capacity ahead of further tightening. The losers beyond individual gamers include smaller system integrators and boutique PC builders, who now have to either absorb margin compression or pass costs on to customers who are already seeing sticker shock at major retailers.
Competitive Comparison: Nvidia vs. AMD Strategy Right Now
Nvidia and AMD are responding to the same supply crunch with notably different public strategies. Nvidia is leaning into premium, high-margin local AI hardware, the $4,999 DGX Spark 64GB and the Microsoft co-developed RTX Spark platform, effectively asking consumers and developers who want local AI capability to pay for dedicated AI hardware rather than compete for gaming GPU inventory. AMD, by contrast, is emphasizing scale and existing shipments, pointing to over half a million agentic PCs already shipped and publishing Gorgon Halo AI benchmarks to argue its current Ryzen AI hardware can handle local AI workloads without needing buyers to purchase separate AI-specific boxes.
| Approach | Nvidia | AMD |
|---|---|---|
| Flagship local AI product | DGX Spark 64GB (GB10 platform), from $4,999 | Existing Ryzen AI-based PCs, benchmarked under “Gorgon Halo” |
| Consumer PC platform push | RTX Spark, co-developed with Microsoft for Windows PCs | Over half a million agentic PCs shipped to date, per AMD |
| Data-center scale product | Vera Rubin NVL72 (MLPerf Inference v6.1 debut) | Helios AI Rack, 72 MI455X GPUs per rack |
| Notable enterprise order | Partner OEMs: Acer, ASUS, Dell, Gigabyte, HP, MSI (DGX Spark) | HPE’s $1.2B Vultr order for Helios AI Rack systems |
The practical effect for a shopper comparing an RTX 5090 against an older RTX 4090 or weighing whether to wait for the rumored RTX 6080 is that neither company currently has an incentive to flood the consumer channel with cheaper inventory while AI-adjacent demand is absorbing supply at higher margins.
Memory Prices Are Making the Problem Worse
GPU pricing does not move in isolation from memory pricing, and 2026 has already seen repeated warnings from memory makers about tightening supply. Micron has said its memory supply is largely sold out for 2027, and separate reporting has tracked GDDR and DRAM cost increases hitting other hardware categories, from SSDs facing NAND price jumps to AMD’s own RX 9070 GRE being squeezed by GPU memory costs. Every graphics card, whether it is a $749 RTX 5070 Ti or a flagship RTX 5090, depends on GDDR memory that is now competing with the same AI buildout pulling DRAM and HBM into data centers. When memory suppliers are sold out more than a year in advance, graphics card makers have little room to absorb cost increases without either raising prices or cutting margins, and the GPUPoet data suggests they are choosing to raise prices.
What This Means If You’re Shopping for a GPU Right Now
For anyone currently comparing RTX 5090 price against a build budget, the GPUPoet data points to a few practical takeaways. First, waiting for prices to drop in the next month or two looks like a weak bet given that every RTX 50-series card tracked rose for a second consecutive month in September. Second, the gap between new-card and resale pricing has narrowed uncomfortably for the RTX 5090 specifically, meaning the usual advice to avoid resale markups is less useful when new units are also trading well above MSRP. Third, buyers considering a jump to the rumored RTX 6080, whose launch order reportedly hinges on GDDR7 chip availability, should not assume a new generation automatically resets pricing to MSRP given how tight memory supply already is.
For builders specifically targeting AI or local inference workloads rather than gaming, the emergence of dedicated platforms like Nvidia’s RTX Spark and the 64GB DGX Spark may actually be worth evaluating as an alternative to overpaying for a gaming GPU repurposed for AI tasks, since those products are built and priced specifically around that use case rather than inheriting gaming-card scarcity pricing.
Where the RTX 5090 Supply Story Stands
Reports on October 16–20, 2026, described severe retail scarcity for the RTX 5090, including Micro Center listing the card as “Not Carried,” while NVIDIA denied discontinuation. Jeff Fisher, identified as senior vice president of NVIDIA’s personal-computing business, said on October 16, 2026, that RTX 5090 production had not stopped and shipments would continue. For shoppers, the denial settles the question of whether the card is being retired, but it does not make cards easier to find.
The GeForce RTX 5090 remains an active, shipping product, although supply is tight. NVIDIA still lists it on its official site, while real-world supply appears severely constrained and prices are far above MSRP. The card launched with 32 GB of GDDR7 memory and a $1,999 launch price, and no acquisition, renaming, or successor announcement is confirmed by the available official-source results. NVIDIA’s own blog also still referenced the card on October 21, 2026, saying the FLUX 3 Image Klein model could generate a one-megapixel image in about four seconds on an RTX 5090.
Price trackers are best read as snapshots rather than final sale prices. GPUPoet’s methodology averaged the three cheapest eBay and Amazon listings, so its best resale deal figure reflects what sellers were asking, not a confirmed market-wide price on any single day. Buyers who need the card should compare several listings and confirm that a retailer actually has stock before paying a premium.
Which GPU Tier Is Getting Hit Hardest
Looking at the three tiers GPUPoet tracked side by side, the pain is not evenly distributed. In dollar terms, the RTX 5090 takes the largest hit: its $2,796 gap between MSRP and best resale price dwarfs the roughly $234 gap on the RTX 5080 and the roughly $146 gap on the RTX 5070 Ti. But in percentage terms, all three cards are moving in the same direction, and the RTX 5090’s 2.4x multiplier is an outlier even by flagship-GPU standards, exceeding the premiums typically seen even during the 2023 AI-training GPU crunch. For shoppers on a tighter budget, the RTX 5070 Ti’s jump from a sub-$900 part to one regularly priced above $1,200 on Amazon is arguably more disruptive in practice, since it pushes a card marketed as a mid-range option into price territory that used to belong to the RTX 5080.
That tier compression matters for anyone comparing current cards against older stock. Buyers who already own an RTX 4090 may find it makes more financial sense to hold onto last-generation hardware for 3D rendering and gaming workloads rather than upgrade into a market where the newer card costs multiples of its original price. That calculus is a reversal from the normal upgrade cycle, where waiting a generation usually means paying less for more performance, not paying a premium just to stay current.
The Bigger Picture: AI Demand Is Reshaping the Entire PC Hardware Market
Graphics cards are just the most visible symptom of a broader reallocation happening across the PC hardware industry in 2026. The same memory and advanced-packaging capacity that GPU makers compete for is also squeezing SSD pricing, system memory, and even some pre-built PC and laptop lines, as component costs rise faster than manufacturers can pass them through without hurting sales. Analysts covering the space increasingly describe this as less a temporary shortage and more a structural shift: AI compute, whether in a hyperscale data center or a local desktop AI box, is now a permanent, growing claimant on the same supply chains that have historically served gaming and general-purpose computing almost exclusively. That shift does not necessarily mean gaming GPUs disappear or stop improving generation to generation, but it does mean the assumption that consumer graphics cards get first claim on available memory and packaging capacity no longer holds the way it did five years ago.
Predictions: Where GPU Pricing Goes From Here
- RTX 5090 pricing is unlikely to return to anywhere near its $1,999 MSRP while Micron and other memory makers report supply sold out well into 2027.
- Expect Nvidia and AMD to keep steering marketing toward dedicated AI desktop hardware (DGX Spark, RTX Spark, Ryzen AI PCs) rather than promising near-term consumer GPU price relief.
- Watch the rumored RTX 6080 launch closely: if its release genuinely hinges on GDDR7 chip costs as reported, its debut pricing could arrive higher than past x080-tier launches.
- Resale and new-card price gaps will likely keep narrowing for high-end cards, making resale marketplaces a less reliable source of savings than in previous shortages.
- Enterprise AI rack orders, like HPE’s $1.2 billion Vultr deal, are likely to keep growing in scale through early 2027, continuing to compete with consumer GPU supply chains.
Frequently Asked Questions
Why is the RTX 5090 price so much higher than its MSRP in October 2026?
According to GPUPoet’s October 2026 market report, the RTX 5090’s best resale price reached $4,795 in September, about 2.4 times the $1,999 MSRP Nvidia set when the card launched on January 30, 2025, driven largely by GPU and memory supply being pulled toward AI data-center and local AI PC hardware rather than gaming cards.
Are RTX 5080 and RTX 5070 Ti prices also rising?
Yes. GPUPoet’s tracking found the RTX 5080’s best September resale price hit $1,233 against a $999 MSRP, and the RTX 5070 Ti reached $895 against a $749 MSRP, with new-card Amazon averages for both cards rising faster than resale prices between August and September.
Is this GPU price surge caused by scalpers or by retailers?
The data suggests both are involved, but retail pricing itself is a bigger factor than in prior shortages. GPUPoet found that new-card prices on Amazon rose faster than resale prices for five of six tracked configurations, indicating retailers are passing through higher costs rather than markups being driven solely by resellers.
What is driving GPU memory shortages in 2026?
Memory makers including Micron have reported that supply is largely sold out well into 2027, as DRAM, GDDR, and HBM capacity is increasingly allocated to AI data-center hardware and AI PC products, leaving less room to absorb rising demand for consumer graphics cards.
Should I buy an RTX 5090 now or wait?
Based on current trends, prices rose for a second consecutive month in September across the entire RTX 50-series lineup GPUPoet tracked. There is no reported indication of an imminent supply correction, so buyers should budget for current pricing rather than waiting for a short-term drop.
What is RTX Spark and how does it relate to gaming GPU prices?
RTX Spark is a platform Nvidia and Microsoft introduced for Windows PCs built around local AI agent workloads. It is a separate product category from gaming GPUs, but it draws on the same underlying silicon and memory supply chains, which is part of why analysts are watching it as a factor in overall GPU and memory pricing.
How does AMD’s GPU pricing compare to Nvidia’s right now?
AMD has also seen price increases across its lineup, including its RX 9070 GRE, which has been affected by the same GPU memory price surge impacting Nvidia’s cards. AMD CEO Lisa Su said on October 6 in Taipei that the company plans to substantially increase chip output, though no specific pricing relief timeline was given.
Will the RTX 6080 be cheaper than the RTX 5090 at launch?
Reports indicate the RTX 6080’s launch order and pricing could hinge heavily on GDDR7 chip availability and cost, meaning its debut price is not guaranteed to undercut current RTX 5090 pricing even though it would sit in a lower product tier.
