For one week in September 2026, AMD outsold Nvidia in graphics cards at one of Europe’s biggest PC parts retailers. According to sales data from German e-commerce site Mindfactory, AMD captured roughly 56% of GPU unit sales in the first week of September 2026, while Nvidia slipped to about 40%, according to Gigazine’s report on the Mindfactory rankings. It is a startling number for a company that, by nearly every other measure, still trails Nvidia badly in the graphics card market. The reason is not that AMD suddenly built a better chip. It is that Nvidia’s flagship RTX 5090 has effectively vanished from US retail shelves, with third-party sellers now charging as much as $9,500 for a card that launched at $1,999, according to a Tom’s Hardware investigation.
The gap between one retailer’s weekly sales chart and Nvidia’s overall market position tells the real story of the GPU market in late 2026. Nvidia still commands somewhere around 90% of global discrete GPU shipments, per Jon Peddie Research (JPR), a dominance that has barely budged in more than a year. But that shipment share measures units that left the factory months ago, not what a gamer can actually buy today at a sane price. The Mindfactory numbers are a real-time snapshot of what happens when a GPU maker’s flagship product becomes functionally unpurchasable: budget-conscious buyers route around it, and AMD’s Radeon lineup is the only mainstream alternative sitting at the checkout counter.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What actually happened at Mindfactory
Mindfactory is not a niche storefront. It is one of Germany’s largest PC component retailers and one of the few outlets that publishes granular, near-real-time sales rankings by brand and model, which makes its data a useful barometer for European PC gaming demand even though it is a single retailer in a single country. In the week covered by Gigazine’s report, AMD Radeon cards outsold Nvidia GeForce cards by a wide margin, a reversal of the pattern that has held for most of the past two years. AMD’s share landed at approximately 56% of GPU units sold, versus roughly 40% for Nvidia, with the remainder split among smaller vendors and Intel’s Arc lineup.
That flip did not happen because AMD launched a new architecture or dropped prices. It happened because Nvidia’s own supply chain problems pushed buyers away. The RTX 5090, the card most associated with Nvidia’s Blackwell generation, has been reported missing from major US online retailers’ in-stock listings, forcing shoppers toward third-party marketplaces where prices have climbed past $9,500 for a card with a $1,999 launch MSRP, according to Tom’s Hardware. Even setting aside those extreme outlier listings, more moderate secondary-market prices have hovered in the $4,600 to $5,000 range for months, a level tech-insider.org covered in its own reporting on sim-racing GPU price increases as the card became nearly unaffordable for enthusiast PC builders.
Why the RTX 5090 disappeared: GDDR7 is the bottleneck
The root cause traces back to memory, not GPU silicon. The RTX 5090 uses 32GB of GDDR7 memory, and that memory type has become one of the most supply-constrained components in the entire PC industry because Samsung, SK Hynix, and Micron have all prioritized wafer capacity for AI-datacenter HBM production over consumer GDDR7 output, according to reporting from Dopodomani that cites the wider GPU memory shortage.
Nvidia has already passed some of that cost onto its board partners twice in 2026. In May, the company told partners to expect roughly a $300 increase in the cost of building RTX 5090 units, effective around May 13, driven by GDDR7 memory costs that had risen approximately 40% since late 2025, according to a report from RunAIHome. That was not an isolated adjustment. In July, Nvidia raised kit costs across its entire GeForce lineup for the second time in 2026, following an earlier price notice tied to the RTX 5090 and its China-market variant, the RTX 5090D V2, as covered by Tech Times. Each round of cost increases works its way downstream into retail pricing, and each round pushes a few more buyers toward AMD’s cheaper Radeon cards.
The price trajectory tells its own story. In March 2026, GDDR7 shortages had already pushed the RTX 5090’s street price to $3,500 to $4,000-plus against its $1,999 list price, according to Compute Market’s pricing analysis. Every subsequent month added more pressure, and by September the card had effectively left first-party retail entirely in some US markets, replaced by third-party listings at more than double that March price.
AMD’s Radeon RX 9070 XT: the accidental beneficiary
AMD did not need to do much to benefit from Nvidia’s supply crisis. The Radeon RX 9070 XT, built on RDNA 4 and the card most directly competing for the budget-to-midrange buyers priced out of Nvidia’s high end, was tracked at a lowest average price of $718 in September 2026, ranging from $675 on the cheapest day to $755 on the most expensive, according to GPU Poet’s pricing data. A separate tracker, TheComparator, listed the cheapest new RX 9070 XT offer at $790 as of September 13, with Newegg and CentralComputer both hovering near the $790 to $800 range. Those prices sit only modestly above the card’s original 2025 launch pricing, a sharp contrast to the RTX 5090’s multiple-hundred-percent markup.
That price stability matters more than any single benchmark chart right now. A gamer weighing an RTX 5090 at $4,600-plus against an RX 9070 XT at roughly $750 is not making a performance decision anymore. They are making an availability and affordability decision, and AMD’s mainstream lineup, including cards like the RX 9070 XT and the tier tech-insider.org detailed in its RTX 5070 vs RX 9070 XT comparison, is simply present on shelves in a way Nvidia’s flagship is not.
The bigger picture: Nvidia still owns roughly 90% of the market
It would be a mistake to read the Mindfactory numbers as evidence that AMD is winning the GPU war. Jon Peddie Research’s shipment data, which tracks add-in-board (AIB) graphics cards across the entire industry rather than one retailer’s weekly sales, shows Nvidia holding roughly 90% of the discrete GPU market through the first half of 2026. Nvidia’s share stood at 91% in the fourth quarter of 2025 and eased only slightly to 90% in the first quarter of 2026, with AMD holding steady at 8% and Intel at 1%, according to Windows Central’s analysis of the JPR data. Tech-insider.org covered the following quarter in detail in its report on Nvidia’s 90% GPU share as shipments hit 12.5 million units in Q2 2026, the highest quarterly shipment total in four years.
Those two data sets are not contradictory. They are measuring different things at different points in the supply chain. JPR’s shipment figures count everything that left a factory and entered the channel over a full quarter, including the millions of lower-end and mid-range Nvidia cards that remain widely available and reasonably priced. Mindfactory’s sales ranking captures a single week of actual purchases at a single retailer, and it is dominated by whatever is in stock and affordable right now, which in September 2026 increasingly means AMD. Put simply, Nvidia still ships roughly nine out of every ten graphics cards sold worldwide, but a shrinking number of those cards are the ones gamers actually want to buy, because the one everyone wants is scarce and absurdly expensive.
| Quarter | Nvidia AIB share | AMD AIB share | Intel AIB share | Source |
|---|---|---|---|---|
| Q3 2025 | 92% | 7% | 1% | Club386 / JPR |
| Q4 2025 | 91% | 8% | 1% | Windows Central / JPR |
| Q1 2026 | 90% | 8% | 1% | Windows Central / JPR |
| Q2 2026 (12.5M units shipped, 4-yr high) | ~90% | ~9% | ~1% | Tom’s Hardware / JPR |
| Mindfactory (single retailer, week of Sept 9, 2026) | ~40% | ~56% | n/a | Gigazine |
Historical echoes: this is not the first GPU shortage
PC gamers who lived through 2020 and 2021 will recognize the pattern, even if the underlying cause is different this time. During the cryptocurrency mining boom of that period, GPU street prices climbed to two to four times their MSRP as miners and gamers competed for the same limited chip supply, a shortage that took roughly two years to fully unwind. That crisis was driven by demand, specifically mining profitability pulling cards off shelves as fast as retailers could stock them. The 2026 shortage is driven by supply, specifically a memory bottleneck upstream of the GPU itself, as GDDR7 fabrication capacity gets diverted toward AI-datacenter memory products, a squeeze tech-insider.org has also tracked in its coverage of the broader RTX PRO 6000 Blackwell workstation card market, where the same memory constraints show up in professional-grade hardware.
The difference matters for how long this lasts. A demand-side shortage, like the mining boom, eventually resolves once the external demand driver collapses, which is roughly what happened when mining profitability crashed in 2022. A supply-side shortage tied to memory fabrication capacity resolves only when chipmakers build more capacity or reallocate existing capacity back toward consumer products, a process that typically takes multiple quarters rather than months. Nvidia’s own board-partner pricing notices in May and July 2026 suggest the company does not expect GDDR7 costs to normalize quickly, since it has now raised kit costs twice within a single year rather than absorbing the increase and waiting for prices to settle.
The China grey market adds another layer
Nvidia’s supply squeeze has also spawned an entirely separate, unofficial market. Modified RTX 5090 cards carrying 96GB of VRAM, triple the standard card’s 32GB, have appeared on Alibaba, reportedly selling for roughly $3,888, well below what a standard RTX 5090 costs on the beleaguered US retail market, a development tech-insider.org detailed in its earlier report on China’s modded RTX 5090 hitting 96GB VRAM for $3,888. These cards are not sanctioned by Nvidia, and buyers take on real risk around warranty support and driver compatibility, but their mere existence underscores how distorted the pricing environment has become: a hand-modified, grey-market card built by a third-party shop in China is now cheaper than an official RTX 5090 bought through legitimate US retail channels.
What this means for gamers deciding what to buy right now
For anyone building or upgrading a gaming PC in September 2026, the practical calculus has shifted. Chasing an RTX 5090 at current market prices no longer makes sense for the vast majority of buyers, since the card’s price-to-performance ratio has collapsed under the weight of its own scarcity. AMD’s RX 9070 XT, priced around $718 to $800 depending on retailer, delivers most of what a mainstream gamer needs at 1440p and competitive 4K settings without asking for a five-figure budget. Nvidia’s own mid-range cards, including the RTX 5070 tier, remain far more reasonably priced than the 5090 and have not seen anywhere near the same level of shortage-driven inflation, since GDDR7 supply constraints hit the highest-memory-capacity cards hardest.
The Mindfactory data is also a reminder that regional markets can diverge sharply from global averages. A US buyer facing bare shelves and $9,500 listings is in a very different position than a German buyer where AMD inventory is plentiful and competitively priced. Shoppers should check local retailer stock and pricing directly rather than assuming global market-share statistics reflect what is actually available to buy in their own country.
| Card | Launch MSRP | September 2026 price | Approximate increase |
|---|---|---|---|
| Nvidia RTX 5090 | $1,999 | $4,600 – $9,500 (third-party) | 130% – 375%+ |
| Nvidia RTX 5090 (grey-market 96GB mod, China) | n/a | ~$3,888 | Below standard RTX 5090 retail price |
| AMD Radeon RX 9070 XT | ~$599 – $649 | $718 – $800 | ~15% – 25% |
| Nvidia RTX 5070 (comparison tier) | ~$549 | Relatively stable, no major shortage reported | Modest |
Where the market goes from here: five predictions
First, expect the Mindfactory-style sales flip to spread to other retailers and regions through the fourth quarter of 2026, as more buyers in more markets get priced out of the RTX 5090 and default to AMD or lower-tier Nvidia cards instead. This will not show up as an AMD shipment-share jump in JPR’s quarterly data right away, since that metric lags actual retail sell-through by weeks.
Second, Nvidia will likely announce a third GDDR7-related cost adjustment before the end of 2026 if memory prices continue climbing, following the pattern set by its May and July price notices to board partners. Each adjustment will further widen the gap between RTX 5090 MSRP and street price, unless Nvidia opts to formally raise the card’s official price instead of letting street markups absorb the difference.
Third, AMD’s overall JPR shipment share, currently around 8% to 9%, is likely to tick upward modestly over the next two quarters, even if it never approaches anything close to parity with Nvidia. A one- or two-point gain would still represent AMD’s best shipment position in several years.
Fourth, grey-market and modified GPU listings, like the 96GB Chinese RTX 5090 variants, will keep growing in volume as long as the price gap between official and unofficial channels stays this wide. Expect more coverage of warranty disputes and compatibility problems tied to these unofficial cards as adoption increases.
Fifth, the GDDR7 shortage will remain the dominant storyline in GPU pricing through at least the first half of 2027, since memory fabrication capacity expansions typically take several quarters to come online, and AI-datacenter demand for high-bandwidth memory shows no sign of easing in the near term.
How the RTX 5090 shortage compares to past GPU crunches
Every major GPU shortage of the past decade has had a distinct root cause, and understanding which type of shortage the market is in matters for predicting how long it lasts. The 2017-2018 shortage was driven by Ethereum mining demand pulling mid-range cards off shelves. The 2020-2021 shortage combined pandemic-driven PC demand with a second, larger wave of mining demand, plus general semiconductor supply chain disruption. The 2026 shortage is different again: it is not driven by external demand for the GPU itself, but by a memory component shortage that has nothing to do with gaming demand at all. GDDR7 capacity is being consumed by the same AI infrastructure boom that has made high-bandwidth memory scarce for AI accelerators, a dynamic tech-insider.org has also tracked in its coverage of the broader memory market. That distinction is why cutting RTX 5090 demand through pricing has not solved the problem the way it eventually did during past mining-driven shortages: gamers pulling back does not free up more GDDR7 wafer capacity, because that capacity is being allocated to entirely different, non-gaming products.
AMD’s priorities look different from Nvidia’s right now
Part of why AMD has more consumer GPU inventory to sell is that its manufacturing priorities diverge from Nvidia’s. Nvidia’s data-center AI accelerator business dwarfs its gaming GPU revenue, and when the company has to choose where to allocate scarce TSMC wafer capacity and GDDR7 memory supply, gaming cards are not the first priority. AMD’s AI accelerator lineup, built around its Instinct data-center chips, is smaller relative to its overall business than Nvidia’s AI silicon is to Nvidia’s, which leaves relatively more manufacturing bandwidth available for consumer Radeon products. That structural difference, not a sudden AMD engineering breakthrough, is a big part of why RX 9070 XT supply has stayed healthy while RTX 5090 supply has not.
It also explains why AMD has little incentive to chase Nvidia’s flagship-tier performance crown right now. A card that competes directly with the RTX 5090 would face the same GDDR7 constraints, since high-end cards with large memory footprints are exactly the segment getting squeezed hardest. AMD’s decision to concentrate its current lineup in the mid-range, where GDDR6 and smaller GDDR7 configurations are easier to source, looks less like a strategic retreat and more like a company building where the memory supply chain actually allows it to build.
Beyond Germany: signs the pattern is spreading
Mindfactory is the clearest single data point available right now, but it is unlikely to be an isolated case. Tom’s Hardware’s own reporting on the RTX 5090’s disappearance from US retail describes the same underlying dynamic driving the German numbers: first-party listings drying up while third-party resellers fill the gap at inflated prices. When a flagship product effectively exits legitimate retail channels in the world’s largest PC gaming market, not just in one European storefront, it becomes far more plausible that AMD’s relative sales strength at Mindfactory reflects a broader trend rather than a local quirk. Buyers should expect similar reporting to surface from other major retailers, including US-based outlets, as September and October data becomes available and analysts look for confirmation that the German pattern is not an anomaly.
What board partners are absorbing behind the scenes
Board partners, the companies that build and sell Nvidia and AMD reference designs under their own brands, have been the ones absorbing and passing along Nvidia’s cost increases throughout 2026. Nvidia’s internal guidance to those partners in May, flagging a roughly $300 per-unit cost increase for RTX 5090 production tied to GDDR7 pricing, was not a one-time notice. The July follow-up, which extended cost increases across the entire GeForce lineup rather than just the flagship card, signals that Nvidia expects memory costs to remain elevated broadly, not just at the high end. That has practical implications for buyers eyeing mid-range cards too: even cards that have not seen dramatic street-price spikes yet could see gradual price creep if board partners continue passing along higher GDDR7 costs across the full product stack.
Frequently asked questions
Did AMD actually overtake Nvidia in the GPU market?
Not globally. AMD overtook Nvidia in unit sales at one retailer, Mindfactory in Germany, for one week in September 2026, capturing roughly 56% of sales versus Nvidia’s 40%. Globally, Nvidia still holds around 90% of discrete GPU shipments according to Jon Peddie Research.
Why is the RTX 5090 so expensive right now?
The card uses 32GB of GDDR7 memory, and GDDR7 supply has been squeezed because memory makers Samsung, SK Hynix, and Micron have prioritized wafer capacity for AI-datacenter memory production. Nvidia has passed rising memory costs onto board partners twice in 2026, and third-party sellers in the US have listed the card for as much as $9,500 against its $1,999 launch price.
Is the AMD Radeon RX 9070 XT a good RTX 5090 alternative?
For most gamers, yes, on value grounds. The RX 9070 XT sells for roughly $718 to $800 in September 2026, only modestly above its 2025 launch price, and delivers strong 1440p and competitive 4K performance without the extreme markup affecting Nvidia’s flagship.
Are the modified 96GB RTX 5090 cards from China safe to buy?
They carry real risk. These cards are not officially sanctioned by Nvidia, so buyers lose standard warranty coverage and may run into driver or compatibility issues. They are cheaper than official retail RTX 5090 units, around $3,888 versus $4,600-plus, but that discount comes with reduced support and accountability if something goes wrong.
When will GPU prices return to normal?
There is no confirmed timeline. Because the current shortage is tied to GDDR7 memory fabrication capacity rather than a temporary demand spike, the shortage is generally expected to persist longer than past mining-driven shortages, likely into at least the first half of 2027, unless memory makers reallocate capacity back toward consumer GDDR7 production.
Is Nvidia losing market share to AMD long-term?
Not meaningfully yet. JPR’s shipment data shows Nvidia’s AIB share easing only slightly, from 91% in Q4 2025 to roughly 90% through the first half of 2026, while AMD has held in the high single digits. The Mindfactory sales flip reflects short-term buying behavior around one scarce product, not a structural shift in market share.
Should I wait for RTX 5090 prices to drop before buying a GPU?
Given that the shortage is supply-driven and tied to a memory bottleneck with no clear near-term resolution, waiting indefinitely is not a reliable strategy for most buyers. Gamers who need a card now are generally better served by evaluating mid-range options from both Nvidia and AMD rather than banking on a near-term RTX 5090 price correction.
Related Coverage
- China's Modded RTX 5090 Hits 96GB VRAM for $3,888 [2026]
- RTX PRO 6000 Blackwell vs RTX 5090 vs W7900: 96GB Gap [2026]
- Sim Racing GPUs Jump 39%: RTX 5090 Nears $5K [2026]
- NVIDIA Grabs 90% GPU Share as Shipments Hit 12.5M [2026]
- NVIDIA Gives Away 4 RTX 5090s in NBA 2K27 Contest [2026]
