The GPU market may be facing a much longer wait than expected, and gamers are not going to like what that means for their wallets. Nvidia’s Rubin-based RTX 60 series is now reportedly expected to land in 2028 instead of 2027, according to recent reporting from Notebookcheck and Tom’s Hardware (Nvidia itself has not officially announced the RTX 60 series, a release date, specifications or a price), so the current RTX 50 generation is set to stay on shelves far longer than usual, at prices that have already climbed well past their original launch numbers.
The RTX 5090 launched at a suggested price of $1,999 in January 2025, the baseline Tom’s Hardware has used to track just how far the card has drifted since. As of September 2026, reporting from AI|Expert citing The Information, PCWorld, TrendForce, and PC Gamer puts real-world retail pricing for the card at roughly $4,329, a figure Best Value GPU’s pricing tracker independently confirms on Amazon, alongside used units running about $3,999 on eBay. Even that range is far from the ceiling: Tom’s Hardware reported in September 2026 that the RTX 5090 had crossed the $5,000 mark outright, with Newegg’s cheapest listing sitting at $4,999, while Micro Center’s in-store listings range from $4,299 to $5,300, the card has largely vanished from major U.S. online retail in favor of third-party listings running $6,500 to $9,500, and TechRadar’s price-tracker update on September 14, 2026 put the cheapest available U.S. card at $6,389. That is not a one-off scalper spike. It is the product of a structural GDDR7 memory shortage that is now colliding with a delayed next-generation launch, and the combination is reshaping how long current hardware has to last.
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 outlets are actually reporting about the 2027 skip
On September 16, 2026, Notebookcheck reported that leaker Moore’s Law Is Dead claims both Nvidia’s Rubin-based GeForce RTX 60 series and AMD’s RDNA 5 GPUs “will launch in 2028,” skipping a 2027 launch window entirely. The outlet notes that Nvidia had originally planned RTX 60 for 2026, then pushed it to 2027 to let AI accelerator demand absorb more TSMC production capacity, and has now pushed it again to 2028.
Tom’s Hardware, in a report first published in February 2026 and updated September 16, states plainly that Nvidia will not release any new RTX gaming GPUs in 2026, and that RTX 60 series production is now likely to begin in 2028. WindowsForum, also reporting September 16, cites hardware leaker Kepler_L2 saying AMD’s “AT2” graphics die will be the only major new gaming GPU chip to reach market in 2027, with the rest of AMD’s next-generation Radeon family, alongside Nvidia’s RTX 60 series, pushed into 2028. This confirms and extends our earlier report on the RTX 60 series skipping 2027, with the September 16 coverage adding fresh detail on pricing fallout.
Not every outlet agrees on the exact window. TechRadar, in coverage dated January 2026 and updated in August, cited a source who put the RTX 60 flagship at “2027H2,” while separately voicing concern that a lingering memory price crisis could jeopardize that timeline. That divergence matters: the 2027-to-2028 slip is not a single confirmed date from Nvidia or AMD, it is a consensus forming across multiple leakers and outlets over the course of 2026, with the most recent reporting converging on 2028 as the more likely landing point. It also remains a rumor-level timeline: Nvidia has not announced the RTX 60 series or confirmed a delay, and the 2027 and 2028 claims both come from anonymous or leaker-based reports.
The real driver: GDDR7 and HBM are being rationed toward AI
The delay narrative usually gets framed as a TSMC capacity story, and that is part of it. But the more immediate and measurable problem is memory. ALiteq reported in August 2026 that 3GB GDDR7 modules, the kind needed for higher-capacity consumer cards, now cost $60 to $70 each, compared with roughly $20 for the 2GB modules used in existing RTX 50-series boards. On an eight-module card, that is a jump from roughly $160 in memory cost to $480-$560, before accounting for the GPU die, board, or cooler.
AI|Expert’s July 2026 reporting, citing The Information and TrendForce, says Nvidia cut RTX 50-series consumer GPU production by 30-40% in the first half of 2026 and redirected memory supply toward higher-margin AI accelerators such as H200 and Blackwell-based data center parts. TechTimes reported in June 2026 that Nvidia had already cut RTX 50 supply by about 20% by that point, pushing street prices higher months before the deeper cuts AI|Expert later described.
This is the same dynamic that has hit CPUs and laptops this year. AI|Expert’s report references comments attributed to Intel CEO Lip-Bu Tan and SK Hynix signaling that relief from HBM and GDDR7 supply constraints will not arrive until 2027-2028, as memory makers keep the bulk of new capacity pointed at data-center customers rather than consumer graphics cards. Samsung, SK Hynix, and Micron are all reported to be allocating the large majority of 2026 HBM output to AI customers, which leaves less room, and less incentive, to expand GDDR7 lines for gaming GPUs. This mirrors the broader HBM4 memory shortage squeezing DRAM inventory across the entire chip industry, not just graphics cards.
RTX 50 Super: the stopgap that keeps slipping
Nvidia’s plan to bridge the gap between RTX 50 and RTX 60 was supposed to be a Super refresh, adding memory capacity and modest clock gains to existing Blackwell silicon. Notebookcheck reported in July 2026 that Nvidia would release RTX 50 Super cards in 2027 rather than a full RTX 60 launch, citing a purported Nvidia source who said they did not think RTX 60 would arrive in 2027 at all.
That stopgap has its own problems. AI|Expert reported that Nvidia has indefinitely delayed the RTX 50 Super lineup, including a completed 24GB RTX 5080 Super design, because of the same structural GDDR7 shortage driving up prices on existing cards. TechTimes, citing BenchLife, said the most realistic window for the Super series was “early 2027, around CES 2027,” while stressing that no release date has been confirmed by Nvidia. In other words, the product built to soften the RTX 60 delay is itself running into the memory shortage that helped cause the RTX 60 delay in the first place.
AMD’s side: one die in 2027, the rest waits for 2028
AMD’s next-generation Radeon lineup, built on the RDNA 5 architecture (referred to in some leak discussions under the broader UDNA naming AMD has used for merging its gaming and compute architectures), is following a similar pattern to Nvidia’s. WindowsForum’s September 16 report on Kepler_L2’s claims describes only a single new die, internally called AT2, reaching market in 2027, with the rest of AMD’s next-generation Radeon family pushed to 2028 alongside Nvidia’s RTX 60 transition.
That narrows AMD’s 2027 desktop lineup to whatever cards can be built around the AT2 die, most likely a limited stopgap tier rather than a full top-to-bottom RDNA 5 refresh. Earlier Notebookcheck leak roundups had already flagged AIB (add-in board) partners as skeptical that RTX 60 or RDNA 5 would launch in 2027 at all, expecting a late-2027 or 2028 window for both. That skepticism, voiced as far back as late 2025, has effectively been validated by the September 2026 reporting.
What this means for current and used GPU prices
The practical effect of a next-gen delay stacked on top of a memory shortage is straightforward: current-generation and used cards have to cover a longer stretch of the market than usual. Normally a two-year architecture cycle gives buyers a predictable window before a successor arrives and pulls used prices down. With RTX 60 and most of RDNA 5 now pointing at 2028, RTX 40, RTX 50, RX 7000, and RX 9000 cards are the only realistic options for roughly three years, not two.
That extended runway, combined with tighter new-card supply, is already showing up in pricing. Beyond the RTX 5090’s climb to $4,329 on Amazon, and considerably higher across other channels detailed below, TechTimes described a broader pattern of climbing prices tied to the 20% RTX 50 supply cut it reported in June 2026. Slower depreciation on used RTX 30 and RTX 40-series cards, and RX 6000 and RX 7000-series AMD cards, is a direct consequence: buyers priced out of new RTX 50 or RX 9000 hardware have fewer fresh alternatives to wait for, so older cards hold value longer than they would in a normal cycle. That same scarcity is fueling grey-market workarounds like the modded 96GB RTX 5090 cards showing up on Alibaba, aimed at buyers who need more VRAM than official SKUs offer.
| Report / Outlet | Date | RTX 60 / RDNA 5 timing claim | Key driver cited |
|---|---|---|---|
| Notebookcheck (Moore’s Law Is Dead) | Sept 16, 2026 | Both RTX 60 and RDNA 5 launch in 2028, skip 2027 | AI demand absorbing TSMC capacity |
| Tom’s Hardware | Feb 5, 2026 (updated Sept 16) | No new RTX gaming GPUs in 2026; RTX 60 production “likely” 2028 | Delay beyond 2027 |
| WindowsForum (Kepler_L2) | Sept 16, 2026 | Only AMD’s AT2 die ships in 2027; rest of Radeon + RTX 60 in 2028 | Broad next-gen transition delay |
| Notebookcheck | July 8, 2026 | RTX 50 Super in 2027 instead of full RTX 60 launch | Nvidia prioritizing AI capacity |
| TechRadar | Jan 9, 2026 (updated Aug 23) | RTX 60 flagship targeting “2027H2” (outlier view) | Memory price crisis risk flagged |
| AI|Expert | July 26, 2026 | RTX 50 Super “indefinitely delayed” | Structural GDDR7 shortage |
RTX 5090 retail pricing in September 2026: four channels, four different prices
The single-number framing of “the RTX 5090 costs $4,329” undersells how fragmented GPU retail became by September 2026. Depending on which channel a buyer checks, the same card can show up at prices spanning nearly $5,500. Best Value GPU’s pricing tracker lists $4,329 as the going Amazon price, with used cards trading around $3,999 on eBay, both comfortably above the card’s original $1,999 MSRP but still among the cheapest realistic ways to buy one. That climb was not sudden: Tom’s Hardware, citing Shane the Gamer, had already tracked a listing at $4,381 on August 7, 2026, and by September the outlet pegged Newegg’s cheapest RTX 5090 at $4,999, right at the threshold where the card crossed $5,000 on other retail trackers. TechSpot’s August 2026 data points the same way: it found the RTX 5090 had reached nearly $4,900, or 145% above MSRP, while global RTX 5090 pricing was 136% above MSRP. Micro Center, which sells the RTX 5090 in-store only rather than shipping it online, has listed units anywhere from $4,299 to $5,300 depending on location and model, with one report citing a $4,299 cheapest in-store option and a separate September 2026 snapshot putting the low end at $4,299.99.
Move outside those two channels and pricing gets considerably worse. Tom’s Hardware reported that by mid-September 2026 the RTX 5090 had effectively disappeared from major U.S. online retailers’ own stock, replaced almost entirely by third-party marketplace sellers asking $6,500 to $9,500 per card, roughly three to nearly five times the original MSRP. That same Tom’s Hardware reporting pegged the card at a flat $5,000 even on mainstream retail trackers that month, making the third-party range look less like an outlier and more like the next step up from an already-inflated baseline. TechRadar’s price-tracking update, published September 14, 2026, found the cheapest available U.S. card at $6,389, a figure that sits much closer to the Tom’s Hardware third-party range than to Best Value GPU’s Amazon or Micro Center numbers.
| Channel | Price (Sept 2026) | Condition | Source |
|---|---|---|---|
| Nvidia MSRP (launch) | $1,999 | New | Nvidia |
| eBay | ~$3,999 | Used | Best Value GPU tracker, Sept 2026 |
| Micro Center (in-store) | $4,299-$5,300 | New | Reports citing Micro Center listings, Sept 2026 |
| Amazon | $4,329 | New | Best Value GPU tracker, Sept 2026 |
| Cheapest tracked U.S. card | $6,389 | New | TechRadar price-tracker update, Sept 14, 2026 |
| Third-party online marketplace | $6,500-$9,500 | New | Tom’s Hardware, Sept 2026 |
That spread says something specific about how the GDDR7 shortage described above is actually distributed. It is not a uniform tax that raises every listed price by the same amount. It is a scarcity that rewards whichever buyers catch an in-store restock or a rare first-party Amazon listing, while everyone else pays whatever a marketplace seller decides to charge. In-store-only retailers such as Micro Center cap purchases per customer and do not let automated checkout bots compete for stock the way online storefronts do, which keeps their listed prices closer to what Nvidia’s board partners are actually charging. Online marketplaces, by contrast, aggregate whatever third-party sellers choose to list, and when first-party stock runs thin, that number floats toward what the most desperate buyer will pay rather than what the card is nominally worth. That is the mechanism behind Tom’s Hardware finding the card missing from major online retail while TechRadar and Best Value GPU each still located listings, just at very different prices depending on whether they were tracking first-party stock or the broader resale market.
For buyers, the practical implication is that $4,329 is closer to a floor than an average, and even that floor is moving: Tom’s Hardware found Newegg’s cheapest RTX 5090 listing at $4,999 in September 2026, with the broader market having already crossed the $5,000 line the same month. Anyone shopping for an RTX 5090 in September 2026 should treat in-store retailers such as Micro Center as the first stop, check Amazon, Newegg, and other first-party listings before scrolling to third-party marketplace results, and budget for the possibility that a same-day online purchase could cost double what a patient, in-store buyer pays. Used cards on eBay, at roughly $3,999, remain the single cheapest path to RTX 5090 ownership, though that discount comes with the usual reduced-warranty risk of buying used hardware.
Framed against the rest of this reporting, the retail fragmentation looks like a preview of what buyers can expect for as long as the GDDR7 shortage and the RTX 60 delay both persist. With Nvidia’s next flagship not expected until 2028, there is no near-term supply event likely to pull these channel prices back toward each other. Buyers should expect the same gap between Micro Center, Amazon, eBay, and third-party marketplace listings to persist, and possibly widen, rather than resolve on its own before then.
Nvidia vs AMD: who is better positioned to absorb the delay
Nvidia enters this delay from a position of overwhelming market strength, which changes the calculus compared with AMD. Nvidia’s data center business, running on Blackwell and Hopper-generation accelerators, generates far more revenue per wafer than gaming GPUs, so redirecting GDDR7 and fab capacity away from RTX cards toward AI hardware is a rational margin decision even if it frustrates gaming customers. A delayed RTX 60 costs Nvidia comparatively little market share, since AMD’s own RDNA 5 lineup is delayed by roughly the same amount, and Nvidia already controls around 90% of GPU shipments heading into this delay.
AMD’s position is more exposed. With only the AT2 die confirmed for a 2027 arrival, AMD’s ability to contest the high end of the market against Nvidia’s RTX 50 and eventual Super refresh is limited to whatever performance tier that single die targets. AMD does not have the same scale of AI accelerator revenue as Nvidia to cushion a prolonged consumer GPU drought, which puts more pressure on its existing RX 9000 lineup to hold share through 2027 largely unchanged, even as AMD has recently overtaken Nvidia in unit GPU sales thanks to RTX 5090 pricing pushing buyers toward cheaper Radeon cards.
Historical context: this is not the first extended GPU cycle
Extended GPU generations are not unprecedented, but a delay driven simultaneously by fab allocation and a memory shortage is a newer combination. The 2020-2022 GPU shortage during the pandemic was driven by a mix of crypto mining demand and broader semiconductor supply constraints, and it resolved once mining demand collapsed and fab output normalized. This cycle is different because the competing demand, AI infrastructure, shows no sign of slowing, and the memory shortage is structural rather than cyclical: HBM and GDDR7 capacity is being built out specifically for data centers, not diverted temporarily from them.
That distinction matters for how long the current squeeze lasts. A demand spike from miners can vanish in a quarter. A capital allocation decision by Samsung, SK Hynix, and Micron to prioritize AI memory over consumer DRAM and GDDR7 takes years to unwind, because it is tied to multi-year fab investment plans rather than a speculative bubble.
Memory economics table: why a Super refresh got harder to ship
| Memory component | Reported cost (2026) | Comparison | Source |
|---|---|---|---|
| 2GB GDDR7 module | ~$20 | Used in current RTX 50-series cards | ALiteq, Aug 2, 2026 |
| 3GB GDDR7 module | $60-$70 | Needed for higher-capacity Super variants | ALiteq, Aug 2, 2026 |
| 8-module memory bill (3GB config) | ~$480-$560 | vs. ~$160 for equivalent 2GB configuration | ALiteq, Aug 2, 2026 |
| RTX 5090 street price (Amazon) | ~$4,329 | vs. $1,999 MSRP at launch | AI|Expert, July 26, 2026; confirmed by Best Value GPU tracker, Sept 2026 |
| RTX 5090 street price (third-party marketplace) | $6,500-$9,500 | vs. ~$4,329 first-party Amazon price | Tom’s Hardware, Sept 2026 |
| RTX 50-series production cut | 30-40% | H1 2026, redirected to AI accelerators | AI|Expert, July 26, 2026 |
What buyers should actually do right now
For anyone shopping for a GPU this fall, the practical takeaway from this reporting is that waiting for RTX 60 or a full RDNA 5 lineup is no longer a short-term bet. If the 2028 timeline reported by Notebookcheck, Tom’s Hardware, and WindowsForum holds, that is roughly two more years on current architectures. Buyers hoping for a Super refresh to land at anything close to current RTX 50 pricing should also temper expectations, given that AI|Expert’s reporting describes the Super lineup as indefinitely delayed over the same GDDR7 shortage inflating today’s prices.
That does not mean prices are guaranteed to keep climbing in a straight line. TrendForce and other market trackers cited across this reporting note that HBM and GDDR7 capacity expansions are underway at Samsung, SK Hynix, and Micron, just aimed primarily at AI customers rather than gaming. Any slack that AI demand leaves behind, whether from a slowdown in accelerator orders or new fab capacity coming online, would flow toward consumer GPU memory first, since that is the segment currently being rationed hardest. Enthusiast segments that depend on high-end cards, including the sim racing community already absorbing a 39% GPU price jump, are among the first to feel that rationing.
Five predictions for the GPU market through 2027-2028
- RTX 50 Super, if it ships, arrives late and thin. Based on TechTimes’ CES 2027 window and AI|Expert’s “indefinitely delayed” characterization, expect a narrow launch, likely limited to one or two SKUs rather than a full stack, if Nvidia ships it at all before RTX 60.
- AMD’s AT2 die becomes the default upper-mid-range option for 2027. With the rest of RDNA 5 pushed to 2028 per WindowsForum’s Kepler_L2 reporting, AT2-based cards will likely absorb most of AMD’s 2027 desktop gaming volume.
- Used GPU prices stay firmer for longer than a typical cycle. With no full next-gen replacement until 2028, RTX 40-series and RX 7000-series cards should depreciate more slowly than in previous generational transitions.
- GDDR7 pricing stays the binding constraint on new SKUs, not die availability. ALiteq’s cost breakdown suggests memory, not GPU silicon, will keep determining which cards Nvidia and AMD can price competitively through 2027.
- 2028 becomes a genuine reset point, not just another leak cycle. With multiple independent outlets (Notebookcheck, Tom’s Hardware, WindowsForum) converging on the same year for both companies, 2028 now looks like the most durable target for the next full architecture change on both sides.
The bigger picture: gaming GPUs are now a secondary priority
What this reporting adds up to is a market structure where consumer gaming GPUs are no longer the primary product driving Nvidia’s or AMD’s roadmap decisions. AI accelerator demand is setting the pace for fab allocation and memory supply, and gaming hardware is being scheduled around whatever capacity is left over. That is a meaningful shift from the last two decades, when GPU architecture cycles were driven primarily by gaming and, more recently, crypto mining demand.
Whether that shift is permanent depends largely on how long the AI infrastructure buildout keeps expanding. If capital spending on AI accelerators plateaus in 2027 or 2028, gaming GPUs could see a faster return to predictable, less inflated release cycles. Until then, the RTX 60 delay rumors that have circulated since September 2026, which Nvidia has not confirmed, look like a preview of how future GPU generations may be scheduled: around AI demand first, gaming demand second.
Frequently asked questions
When will the Nvidia RTX 60 series actually launch?
Reporting from Notebookcheck, Tom’s Hardware, and WindowsForum published September 16, 2026 converges on 2028 as the most likely launch window for Nvidia’s Rubin-based RTX 60 series, having slipped from an earlier 2027 target, though Nvidia has not announced the series or confirmed any date. TechRadar’s earlier reporting cited a “2027H2” estimate, but more recent coverage suggests that window has since moved later.
Is AMD’s RDNA 5 also delayed to 2028?
Most of it, according to WindowsForum’s September 16, 2026 report citing leaker Kepler_L2. Only a single die, referred to as AT2, is expected to reach market in 2027, with the rest of AMD’s next-generation Radeon family following in 2028 alongside Nvidia’s RTX 60 series.
Why is the RTX 5090 so much more expensive than its $1,999 launch price?
AI|Expert’s July 2026 reporting, citing The Information, PCWorld, TrendForce, and PC Gamer, attributes the RTX 5090’s climb from its January 2025 launch price primarily to a structural GDDR7 memory shortage and a 30-40% cut to RTX 50-series production in the first half of 2026, as Nvidia redirected memory supply toward AI accelerators. By mid-September 2026, that shortage had pushed prices even further apart by channel: as low as roughly $3,999 for a used card on eBay, $4,999 for Newegg’s cheapest listing, and as high as $6,500 to $9,500 on third-party marketplaces once first-party stock disappeared, per Tom’s Hardware, which separately confirmed the card had cleared the $5,000 mark that month, with TechRadar separately tracking a $6,389 floor for the cheapest available new U.S. card.
Will the RTX 50 Super series still launch?
It has not been officially confirmed by Nvidia. TechTimes cited BenchLife’s estimate of an early-2027, CES-timed window, while AI|Expert described the lineup, including a completed 24GB RTX 5080 Super design, as indefinitely delayed due to the GDDR7 shortage.
What is causing the GDDR7 and HBM shortage?
Reporting cited by AI|Expert says Samsung, SK Hynix, and Micron have allocated the large majority of their 2026 HBM production to AI data-center customers, leaving less capacity for GDDR7 modules used in consumer gaming GPUs. Comments attributed to Intel CEO Lip-Bu Tan and SK Hynix in that same reporting suggest meaningful relief is not expected until 2027-2028.
Should I buy a GPU now or wait for RTX 60 or RDNA 5?
Based on the reporting above, waiting specifically for RTX 60 or a full RDNA 5 lineup means waiting roughly two more years, into 2028. Buyers who need a GPU sooner are more likely to be shopping among current RTX 40, RTX 50, RX 7000, and RX 9000-series cards, or a limited RTX 50 Super or AMD AT2-based refresh, if either ships in 2027 as some reports suggest. Within the current generation, channel matters as much as timing: Micro Center’s in-store listings and Amazon’s tracked price both sit well below third-party marketplace asking prices, so checking first-party stock before resorting to resale listings can meaningfully change the price paid.
Does this delay affect laptop GPUs too?
The reporting cited here focuses on desktop RTX 60 and RDNA 5 parts. Given that the underlying constraint is memory supply rather than a desktop-specific issue, it is reasonable to expect mobile GPU refreshes to face similar scheduling pressure, though none of the outlets cited in this piece give a specific mobile GPU timeline.
