Akihabara is usually where Tokyo’s PC builders go to hunt for a deal. This month it became ground zero for the opposite. A major Japanese retailer issued a public alert warning shoppers of further graphics card price increases, days after stores across the district had already pushed through 20% to 30% hikes on many GPU models within a matter of weeks, according to a September 9, 2026 report from WebProNews. The retailer’s message was blunt: the “AI memory crunch” that has been squeezing chipmakers all year is now landing directly on consumer shelves, and there is no relief window in sight before 2028.
The Japan story is a localized flashpoint, but it is also a preview of what is happening globally. gHacks Tech News reported in early September 2026 that average GPU prices jumped 15% in a single month, with GeForce cards up 19% and Radeon models up 11%, per the WebProNews summary of that data. Nvidia’s RTX 5090 has become the poster child for the surge: PCGuide reported on September 14, 2026 that some listings now price the card above $6,500, with select overseas offers approaching $9,800 to $10,000, as AI firms reportedly buy the flagship gaming card in bulk to repurpose for server and inference workloads. TechPowerUp data cited by SiliconAnalysts put the RTX 5090 136% above its $1,999 MSRP as of September 1, 2026.
This is the latest escalation in a price cycle that has already pushed RTX 5090 pricing past $4,329 earlier this year, and it comes as the broader memory shortage has pushed GPU relief out to 2028. What makes this round of GPU inflation different from the mining-driven shortages of 2021 or the pandemic-era supply chain snarls is the root cause. This time it is not scalpers or crypto miners buying up stock. It is a structural shortage of the memory chips that every modern GPU depends on, both the GDDR6/GDDR7 that goes into gaming cards and the broader DRAM market that AI data centers are consuming at an unprecedented rate. Nvidia’s own workstation lineup shows how tight things have become: the RTX PRO 6000 Blackwell, a data-center-oriented card with 96GB of memory, jumped to $16,000, more than double its original pre-order price, PCGuide reported, as AI companies buy the professional card in volume alongside the gaming-branded RTX 5090.
Why Japan’s retail market is the clearest window into the crunch
Japan’s electronics retail sector, concentrated in Tokyo’s Akihabara district, has long served as an early-warning system for global hardware pricing because Japanese stores tend to reprice inventory faster and more transparently than big-box retailers in the US or Europe. When a major Japanese retailer issues a formal alert about further price increases, as one did on September 9, 2026 according to WebProNews, it tends to reflect wholesale pricing pressure that has not yet fully hit shelves elsewhere. The 20% to 30% increases logged across many GPU models in Akihabara in the weeks before that alert were not isolated promotions ending or seasonal demand. They tracked the same GDDR6/GDDR7 and broader DRAM price curve that TrendForce and other memory analysts have been charting since the start of the year.
That matters for a US and global audience because Japan’s retail pricing has historically moved a few weeks ahead of North American and European street prices during past shortage cycles. If the pattern holds, the double-digit percentage jumps logged in Tokyo this month are a preview of what US retailers will show on their own shelves through the rest of the fall. The underlying driver, tight graphics DRAM supply feeding into an already strained overall DRAM market, is global, not regional, which is exactly why a Tokyo retail alert carries weight for buyers in Chicago or Berlin.
The workstation GPU angle: RTX PRO 6000 as a bellwether
The gaming GPU market gets most of the headlines, but the more telling price action is happening one tier up, in workstation and data-center-adjacent cards. Nvidia’s RTX PRO 6000 Blackwell, built for professional visualization and AI development workloads with 96GB of memory, has seen its price more than double to $16,000 from its original pre-order figure, according to PCGuide’s September 14, 2026 report. That is not a card most gamers will ever touch, but its price trajectory matters because it shares the same GDDR7 memory pool as the consumer RTX 5090, and demand for one directly competes with supply for the other, a dynamic explored in a recent RTX PRO 6000 Blackwell vs. RTX 5090 vs. Radeon PRO W7900 comparison.
AI firms buying RTX PRO 6000 units in bulk for inference and fine-tuning work are, in effect, bidding against PC gamers for the same manufacturing capacity at the same memory fabs. PCGuide’s reporting notes that AI firms have started turning to the flagship RTX 5090 gaming model itself, which ships with 32GB of GDDR7, as a substitute when workstation cards are unavailable or too expensive, further tightening consumer-facing supply. This is the mechanism behind the price surge that a purely gaming-focused explanation misses: the GPU market and the AI compute market have effectively merged into a single supply pool, and gamers are now competing directly with data center procurement teams for the same silicon and the same memory packages.
The DRAM and GDDR7 supply chain, explained
To understand why graphics cards are getting more expensive rather than cheaper as their launch dates recede further into the past, it helps to separate two distinct memory markets that are both under pressure simultaneously. The first is graphics DRAM, meaning GDDR6 and GDDR7, the specialized high-bandwidth memory that sits directly on a graphics card next to the GPU die. The second is conventional DRAM, the general-purpose memory used in servers, PCs, and phones, which competes for the same fabrication capacity at memory makers like Samsung, SK Hynix, and Micron.
TrendForce’s July 3, 2026 press note on the graphics DRAM segment found a counterintuitive wrinkle: the RTX PRO 6000 Blackwell had not generated the wave of GDDR7 demand that memory suppliers originally expected, and weaker notebook shipments also cut into GDDR6 and GDDR7 demand from that channel. Despite that, memory suppliers responded by reallocating production capacity toward other mainstream products rather than expanding graphics DRAM output, which kept graphics memory supply constrained even as one source of demand softened. A follow-up summary from Igor’s Lab on July 8, 2026, citing the same TrendForce data, confirmed that graphics DRAM remained tight and that GDDR6/GDDR7 prices continued rising in line with the broader DRAM market.
The broader DRAM market tells an even starker story. Conventional DRAM contract prices were projected to rise 13% to 18% quarter over quarter in the third quarter of 2026, with NAND flash contract prices rising 10% to 15% over the same period, according to the TrendForce figures reported by Igor’s Lab. That is on top of a first-quarter 2026 spike that saw contract DRAM prices climb by a record amount quarter over quarter, a jump that outside analysts have described as the worst memory supply imbalance in roughly 15 years. When the raw material going into a graphics card’s memory chips is itself in record-short supply and rising in price every quarter, GPU assemblers have little room to absorb the cost before passing it on to retailers and, ultimately, buyers. The same pressure has already been documented on the high-bandwidth side of the market, where an HBM4 crunch has pushed DRAM inventory below 10 days, and on the consumer platform side, where DDR5 prices have climbed as much as 486%, forcing AMD to extend support for older AM4 boards.
RTX 60 series delay to 2028: what changed
For most of 2026, the working assumption among hardware leakers was that Nvidia’s next-generation RTX 60 series, expected to be built on the Rubin architecture, would begin mass production toward the end of 2027 and reach shelves shortly after. That timeline has now slipped further. PCGuide reported on September 17, 2026 that a reliable hardware leaker now claims Nvidia will not release its next-generation GeForce cards until 2028, which would leave a three-year gap between the RTX 50 series and the RTX 60 series, a much longer interval than Nvidia’s historical roughly two-year cadence. Tom’s Hardware’s reporting on the same trajectory confirms that Nvidia will not release any new RTX gaming GPUs in 2026 at all, and that RTX 60 series mass production has slipped beyond 2027 into what is now described as a likely 2028 start.
AMD’s next-generation RDNA 5 architecture, which is expected to compete directly with the RTX 60 series, does not appear positioned to fill the gap either. PCGuide’s September 17 report notes that AMD might only launch a single RDNA 5 GPU next year, a far more limited rollout than a full stack refresh, which would leave the vast majority of the market running on GPUs that are already two to three years old by the time any real next-generation alternative becomes broadly available. SetupLevel’s September 10, 2026 roundup underscores just how unsettled the official picture remains: as of that date, neither Nvidia nor AMD had announced official next-generation gaming silicon launch dates, and even the leak-based windows diverge between sources, with some still citing a second-half-2027 target for Rubin and RDNA 5 while others, including PCGuide and Tom’s Hardware, now point to 2028.
How this compares to past GPU shortage cycles
Gamers and IT buyers who lived through the 2020-2021 cryptomining boom or the pandemic-era chip shortage might assume this is a repeat of the same playbook: temporary demand spike, panic buying, and prices normalizing once the underlying driver fades. The structural difference this time is that the RTX 5090 shortage is being driven by scarcity in the memory supply chain itself rather than by resale demand. A September 1, 2026 GPU market report from GPUPoet made this point explicitly, noting that resale prices are now cheaper than retail on every RTX 50 card, the opposite of what scalping-driven shortages typically produce, because GDDR7 has become scarce enough that Nvidia is reportedly allocating it by revenue per gigabyte rather than by unit volume. That report also noted no new consumer GPU generation was expected until late 2027 at the earliest, a timeline that has since slipped further based on the newer 2028 reporting.
During the mining boom, once new mining hardware and falling cryptocurrency prices removed the incentive to hoard GPUs, retail prices corrected within a year or two as manufacturing caught up with a demand spike that was, at its core, temporary. The current shortage has no equivalent release valve. AI datacenter demand for memory is not a speculative bubble tied to a volatile asset price; it is tied to ongoing infrastructure build-outs by hyperscalers and AI labs that show no sign of slowing, and memory fabs cannot simply flip a switch to add graphics DRAM capacity without multi-year investment cycles. That is the core reason analysts are not forecasting a 2026-style price correction the way they did after past shortages faded.
Regional pricing snapshot: how far prices have moved
| Metric | Figure | Source |
|---|---|---|
| RTX 5090 price vs. $1,999 MSRP (Sept. 1, 2026) | 136% above MSRP | TechPowerUp, via SiliconAnalysts |
| RTX 5090 street listings (Sept. 14, 2026) | Above $6,500; some overseas offers near $9,800-$10,000 | PCGuide |
| Average GPU price change (single month, early Sept. 2026) | +15% overall; GeForce +19%, Radeon +11% | gHacks Tech News |
| Akihabara (Tokyo) retail GPU hikes (weeks before Sept. 9, 2026) | +20% to +30% across many models | WebProNews |
| RTX PRO 6000 Blackwell (96GB) price | $16,000, more than double original pre-order price | PCGuide |
| Conventional DRAM contract price change (Q3 2026, QoQ) | +13% to +18% | TrendForce, via Igor’s Lab |
| NAND flash contract price change (Q3 2026, QoQ) | +10% to +15% | TrendForce, via Igor’s Lab |
Next-gen GPU timeline: leaked windows compared
| GPU family | Vendor | Earlier expected window | Latest reported window |
|---|---|---|---|
| RTX 60 series (Rubin) | Nvidia | Mass production end of 2027 | 2028, per reliable leaker cited by PCGuide (Sept. 17, 2026) |
| RTX 60 series (Rubin) | Nvidia | Second half of 2027 | No official date as of Sept. 10, 2026, per SetupLevel |
| RDNA 5 | AMD | Broader stack refresh expected alongside RTX 60 | Possibly a single GPU launch next year, per PCGuide |
| RTX 50 “Super” refresh | Nvidia | 24GB RTX 5080 Super at completed-design stage | Shelved, per Tech-Insider reporting |
Market impact: who feels this first
The immediate impact falls hardest on three groups: PC gamers looking to build or upgrade a high-end rig, small and mid-size AI development shops that cannot compete with hyperscaler procurement budgets, and system integrators who quote GPU-inclusive builds to enterprise and creative customers. For gamers, the RTX 5090’s climb toward $10,000 in some overseas markets, as documented by PCGuide, effectively removes flagship-tier gaming hardware from consideration for anyone outside enthusiast or professional budgets. That pushes more buyers toward mid-range cards, which are themselves not immune, since the 19% average GeForce price increase cited by gHacks Tech News reflects broad-based inflation across the stack, not just the top SKU.
For smaller AI teams, the RTX PRO 6000’s jump to $16,000 is arguably more consequential. That card was positioned as a more accessible on-premises alternative to renting cloud GPU capacity from AWS, Azure, or Google Cloud. At double its original pre-order price, the math shifts back toward cloud rental for many workloads, which in turn increases demand on cloud GPU capacity that is itself constrained by the same underlying memory shortage. Retailers and system integrators, meanwhile, are caught in the middle: locking in quotes for enterprise clients becomes riskier when wholesale GPU pricing can move 15% to 30% within a matter of weeks, as it did in Japan.
Laptop and pre-built systems face a quieter squeeze
Discrete desktop GPUs get the most attention because their prices are visible and trackable in real time, but laptops and pre-built desktops absorb the same memory cost pressure in less obvious ways. TrendForce’s July 2026 note specifically flagged weaker notebook shipments as one reason GDDR6/GDDR7 demand softened from that channel, which suggests laptop makers are already adjusting configurations, order volumes, or pricing in response to memory costs rather than absorbing them silently. Buyers shopping for gaming laptops in the run-up to the holiday season should expect either higher prices on comparable configurations versus a year ago, or manufacturers quietly trimming memory allocations and VRAM tiers on mid-range models to protect margins, a pattern that has shown up in past DRAM shortage cycles.
What this means for the used and resale GPU market
One of the more unusual signals in this cycle is that resale pricing has decoupled from the pattern buyers expect during a shortage. GPUPoet’s September 1, 2026 market report found that resale is now cheaper than retail on every RTX 50 card, a reversal of the scalper-driven markups that defined the 2021 shortage, when used cards traded at a premium to new stock because new stock simply was not available. That inversion signals something important: current buyers are not hoarding cards speculatively in anticipation of reselling them at a profit, because new retail pricing is already climbing faster than the secondary market can track. For anyone shopping the used market right now, that means a used RTX 5090 or RTX 4090 may genuinely be the better value proposition compared with a new unit at inflated retail pricing, a dynamic that rarely held during previous shortage cycles.
Nvidia’s broader pricing strategy under scrutiny
Nvidia has reportedly told major customers that prices on servers packed with its AI chips are going up more than 15% in many cases, a signal that the company is passing through its own rising input costs rather than absorbing them to protect market share, based on Tech-Insider’s August 29, 2026 reporting. That decision, whether explicit policy or simply a function of constrained supply meeting essentially unlimited AI demand, has direct knock-on effects for consumer GPU pricing, because Nvidia’s gaming and data center product lines increasingly draw from the same wafer allocations at TSMC and the same memory packages from Samsung, SK Hynix, and Micron. When a company facing that kind of two-sided demand chooses where to allocate scarce GDDR7 and advanced packaging capacity, data center and AI customers paying enterprise prices are a more attractive allocation than consumer gaming channels paying MSRP, a dynamic that helps explain why gaming GPU supply has tightened even as Nvidia’s overall shipment volumes remain strong.
Predictions: where GPU pricing heads next
- Expect US and European retail prices to follow Japan’s pattern with a lag of several weeks to a couple of months, given the historical tendency for Tokyo retail pricing to move ahead of Western markets during memory shortage cycles.
- The RTX PRO 6000 Blackwell’s climb toward $16,000 will likely push more small AI teams back toward cloud GPU rental, adding further pressure to already-tight cloud GPU capacity at the major hyperscalers.
- With RTX 60 series and RDNA 5 both now looking like 2028 stories rather than 2027 ones, expect the RTX 5090 and RX 9070 XT to remain the practical ceiling of the consumer market for at least another 12 to 18 months.
- Laptop OEMs are likely to quietly reduce VRAM allocations or raise prices on mid-range gaming laptop configurations through the holiday shopping season, following the notebook-channel softness already flagged by TrendForce.
- Used and resale GPU pricing should remain a comparatively better value than new retail stock for as long as the resale-cheaper-than-retail inversion documented by GPUPoet persists, making the secondary market the more rational purchase path for budget-conscious buyers through early 2027.
What buyers can actually do right now
This latest wave of pricing pressure is part of a broader AI chip supply story that Tech Insider has been tracking throughout 2026 in its AI chips and GPU pricing coverage. For anyone who needs a GPU today rather than waiting out a shortage with no clear end date, a few practical steps reduce exposure to further price hikes. Buying from the secondary market makes more sense than it has in years, given that resale pricing is currently undercutting retail across the RTX 50 lineup. Locking in a purchase sooner rather than later is also rational given the documented pattern of retailers repricing inventory upward every few weeks rather than waiting for a single large jump. For enterprise and AI development teams evaluating on-premises hardware versus cloud rental, the RTX PRO 6000’s doubled price tag changes the calculus meaningfully enough that a fresh cost comparison against current AWS, Azure, or Google Cloud GPU instance pricing is worth running before committing to a capital purchase that was priced against last year’s cost baseline.
Frequently asked questions
Why are GPU prices rising in September 2026 if crypto mining is not a factor?
The current price surge is driven by a global shortage of DRAM and graphics-specific GDDR6/GDDR7 memory, not by resale or mining demand. AI data centers are competing directly with consumer GPU makers for the same memory manufacturing capacity, which has pushed contract DRAM prices up double digits quarter over quarter through 2026, according to TrendForce data.
How much has the RTX 5090 price increased?
As of September 1, 2026, the RTX 5090 was trading 136% above its $1,999 MSRP, according to TechPowerUp data cited by SiliconAnalysts. PCGuide reported on September 14, 2026 that some listings had climbed past $6,500, with select overseas offers approaching $9,800 to $10,000.
Why did Japanese retailers specifically raise prices 20% to 30%?
A major Japanese retailer issued a public alert on September 9, 2026 about further GPU price increases, following hikes of 20% to 30% across many models in Tokyo’s Akihabara district in the preceding weeks, according to WebProNews. Japanese retail pricing tends to reflect wholesale memory cost pressure earlier than markets in the US and Europe.
When will the RTX 60 series launch?
As of mid-September 2026, neither Nvidia nor AMD has confirmed an official launch date for next-generation gaming GPUs. A reliable hardware leaker cited by PCGuide on September 17, 2026 now points to 2028 for the RTX 60 series, a delay from earlier expectations of a late-2027 window, which would leave roughly a three-year gap since the RTX 50 series.
Is buying a used GPU a better option right now than buying new?
Based on a September 1, 2026 market report from GPUPoet, resale pricing on every RTX 50 card was running cheaper than retail, a reversal of the scalper-driven premiums seen in the 2021 shortage. That makes the used market a comparatively rational option for budget-conscious buyers for as long as new retail pricing continues climbing faster than the secondary market.
What is the RTX PRO 6000 Blackwell and why does its price matter to gamers?
The RTX PRO 6000 Blackwell is Nvidia’s 96GB workstation and AI development card, which reportedly jumped to $16,000, more than double its original pre-order price, per PCGuide. Because it draws on the same GDDR7 memory supply as the consumer RTX 5090, rising demand for the workstation card directly tightens supply and pushes up pricing for gaming-focused cards.
Will AMD’s RDNA 5 offer relief if Nvidia’s RTX 60 series slips to 2028?
Unlikely in the near term. PCGuide reported on September 17, 2026 that AMD might launch only a single RDNA 5 GPU next year rather than a full stack refresh, which would leave most of the market still relying on RDNA 4 and RTX 50-series hardware well into 2027 and possibly 2028.
How does this shortage compare to the 2021 GPU crisis?
The 2021 shortage was driven primarily by cryptocurrency mining demand and pandemic-era supply chain disruption, both of which eventually eased as mining profitability fell and chip production normalized. The current shortage is rooted in structural DRAM and GDDR7 memory scarcity tied to sustained AI data center buildouts, which analysts do not expect to ease on a similar timeline, since memory fabs require multi-year investment cycles to expand capacity.
