Graphics card buyers hoping for a price correction have run out of runway. New pricing data tracked through September 2026 shows GPU costs climbing for the third time this year, and the underlying cause, a historic memory chip shortage, is now pushing the next generation of gaming GPUs from Nvidia and AMD out to 2028. There is no meaningful relief window left in 2027.
The math is stark. Nvidia’s RTX 5090 launched at $1,999 in early 2025 and now sells for more than $5,000 at US retail, according to BigGo Finance. An Asus RTX 5080 that sold for $1,199 has jumped to $1,595 on Amazon. AMD’s RX 9070 XT, once a $830 card, has crossed $1,037 after a Q3 2026 MSRP hike of up to 20%, BigGo Finance reported. None of this is a one-off spike. It is the third coordinated round of GPU price increases from Nvidia and AMD in 2026 alone, following hikes in January and a narrower RTX 5090-only adjustment in May.
Behind the sticker shock sits a memory market that TrendForce describes as historically tight. Conventional DRAM contract prices rose 90% to 95% quarter-on-quarter in Q1 2026, the steepest quarterly jump the research firm has recorded, while NAND flash climbed 55% to 60% in the same period, according to TrendForce’s January 2026 outlook. That pace barely slowed in Q2, with DRAM up another 58% to 63% and NAND up 70% to 75%, per TrendForce data cited by Tom’s Hardware. Even the Q3 2026 “moderation” TrendForce flagged, DRAM up 13% to 18% and NAND up 10% to 15%, still represents ongoing inflation stacked on top of two record-setting quarters.
GPU prices in September 2026: where things stand
Every current-generation GPU tier has moved up in price since the start of 2026, and the increases are no longer limited to the flagship RTX 5090. BigGo Finance’s tracking shows Nvidia GeForce RTX pricing rising by an average of roughly 15% in a single month during one of its 2026 hikes, with a subsequent round pushing select RTX models up 20% to 30%. AMD followed with its own August 2026 increase of at least 10% across select Radeon GPUs and GDDR6 memory kits, a move BigGo Finance attributed directly to the same AI-driven memory cost surge Nvidia cited for its own hikes.
Workstation and professional cards have not been spared either. Nvidia’s RTX Pro 6000 Blackwell, a card built for AI and content-creation workloads rather than gaming, has reportedly doubled from its launch price to around $16,000, according to BigGo Finance’s coverage of the memory crunch. The following table summarizes approximate current street pricing against original launch MSRP for the cards most affected.
| GPU | Launch MSRP | Approx. price, Sept. 2026 | Approx. increase |
|---|---|---|---|
| Nvidia RTX 5090 | $1,999 | $5,000+ | 150%+ |
| Nvidia RTX 5080 | $1,199 | ~$1,595 | ~33% |
| Nvidia RTX Pro 6000 Blackwell | ~$8,000 | ~$16,000 | ~100% |
| AMD RX 9070 XT | ~$599 (original) | $1,037+ | ~20% since Q3 hike alone |
| AMD Radeon GDDR6 kits (select) | N/A | +10% Aug. 2026 hike | ~10% |
Source: BigGo Finance pricing coverage, September 2026. Figures reflect reported retail and MSRP data points rather than a single fixed price, since street pricing fluctuates by retailer and region.
For readers tracking individual models more closely, tech-insider.org has separately covered the RTX 5090’s climb past $4,300 and the broader GPU market share data showing Nvidia holding roughly 90% of shipments even as prices rise.
Why next-gen GPUs are slipping to 2028
The GPU price surge would be a story on its own, but the deeper issue is what it is doing to product roadmaps. Multiple outlets, including WCCFTech and eTeknix, report that Nvidia and AMD are each delaying their next major gaming GPU architectures, Nvidia’s Rubin-based RTX 60 series and AMD’s RDNA 5 lineup, from a 2027 target to 2028. The reasoning cited across these reports is consistent: severe memory supply constraints and a decision by both companies to prioritize memory allocation for AI accelerators over consumer graphics memory.
According to WCCFTech’s reporting, silicon manufacturers are dedicating production lines to high-bandwidth memory destined for AI servers rather than the standard GDDR memory that gaming GPUs require. eTeknix frames the delay as a direct consequence of that prioritization: when a fab has to choose between HBM4 for a data-center customer paying tens of thousands of dollars per chip and GDDR7 for a gaming card selling for a few hundred dollars, the data-center order wins the capacity.
That timeline overlap is itself notable. Because both companies are reportedly citing the same memory constraints, RDNA 5 and Rubin-based gaming cards are expected to launch within months of each other in 2028 rather than staggered a full generation apart, a shift from the historical pattern where Nvidia and AMD have often launched 6-12 months out of sync.
The AT2 exception: AMD’s lone 2027 release
Not every next-gen chip is being pushed to 2028. Reporting indicates AMD is preparing one exception: a mid-range GPU codenamed AT2, expected to arrive in 2027 ahead of the rest of the RDNA 5 lineup. Specs discussed in that coverage point to a chip with up to 40 compute units (roughly 5,120 shaders), a memory bus between 192-bit and 384-bit, and 12GB to 24GB of graphics memory, positioning it as a mainstream part rather than a flagship.
The existence of a single earlier release suggests the delay is not absolute across every SKU, but rather a capacity-allocation decision: AMD can secure enough GDDR memory for one mid-tier chip in 2027, but not the volume needed to launch a full stack, including the halo card gamers are waiting for, until fab and memory supply loosens in 2028.
Memory is the real bottleneck
Every layer of this story traces back to the same root cause: a memory chip market that cannot keep up with simultaneous demand from AI data centers, smartphones, PCs, and now gaming GPUs. TrendForce’s quarterly memory pricing data shows just how compressed the timeline has been.
| Quarter | DRAM contract price change (QoQ) | NAND flash contract price change (QoQ) | Source |
|---|---|---|---|
| Q1 2026 | +90% to +95% | +55% to +60% | TrendForce |
| Q2 2026 | +58% to +63% | +70% to +75% | TrendForce / Tom’s Hardware |
| Q3 2026 | +13% to +18% | +10% to +15% | TrendForce |
TrendForce’s own read on the Q3 slowdown is not that supply has improved. It is that buyers are increasingly unwilling to accept every further price increase, since memory is already so expensive that PC, smartphone, and consumer-device makers are pushing back on new contract terms, according to TrendForce’s Q3 2026 press release. The firm has also said new fab capacity is unlikely to come online in meaningful volume before late 2027 or 2028, which lines up almost exactly with the delayed GPU launch window Nvidia and AMD are reportedly targeting.
This is not the first time tech-insider.org has flagged the inventory side of this story. Coverage of the HBM4 shortage and falling DRAM inventory levels and the ripple effect on DDR5 system memory pricing both point to the same structural cause: AI infrastructure spending is absorbing capacity that used to serve consumer electronics.
Samsung, SK Hynix, and the AI memory land grab
A 20% DRAM price increase from Samsung has been cited as a direct root cause of the latest round of GPU price hikes, according to reporting tracked by BigGo Finance. Samsung, SK Hynix, and Micron collectively control the overwhelming majority of global DRAM and NAND production, and all three have prioritized high-margin, high-volume orders from AI infrastructure buyers, hyperscale cloud operators building out data centers packed with AI accelerators that each require enormous amounts of high-bandwidth memory.
SK Hynix’s own leadership has acknowledged how tight the outlook remains. Chief executive Kwak Noh-jung told Reuters that 2027 is shaping up to be an especially difficult year for the memory industry from a supply standpoint, a comment that underscores why analysts do not expect meaningful relief before 2028 even as current pricing already sits at multi-year highs.
That production math also explains why gaming GPUs, which use GDDR6 and GDDR7 rather than the HBM4 stacks found in AI accelerators, are getting squeezed indirectly. GDDR and HBM are produced on overlapping fab lines and compete for the same wafer starts and packaging capacity. When a fab operator can allocate a production run to HBM4 destined for a data-center customer at a dramatically higher margin, GDDR7 orders for consumer graphics cards move down the priority list.
Stock market reaction: memory makers ride the wave
Wall Street has treated the memory shortage as a windfall for chipmakers even as it squeezes consumer electronics buyers. Micron, SK Hynix, and Samsung have all posted substantial gains through 2026 on the back of the shortage-driven pricing environment, with Micron in particular drawing a wave of price-target upgrades from analysts, according to coverage from The Motley Fool.
That rally has not been in a straight line. Memory stocks have shown real volatility around any news that could soften AI-driven demand, including a sharp dip when a Chinese AI lab claimed a new model required significantly less memory than prior versions. The pattern illustrates how tightly memory-stock valuations, and by extension GPU component costs, are now tied to the AI infrastructure buildout rather than to traditional PC or gaming demand cycles.
Competitive comparison: how Nvidia and AMD are each responding
Nvidia and AMD are running the same playbook with different emphasis. Nvidia has moved first and more aggressively on pricing, implementing three separate rounds of GeForce RTX price increases in 2026, while also raising prices on AI server products by more than 15%, according to BigGo Finance. That reflects Nvidia’s position as the dominant AI accelerator supplier: with roughly 90% of GPU shipments already flowing through Nvidia’s channel, the company has more room to pass memory costs on to buyers who have few alternatives.
AMD has followed roughly a month behind on each occasion, applying smaller percentage increases (around 10% in August 2026) before catching up with its own 20% RX 9000-series hike in Q3. AMD’s decision to greenlight the AT2 chip for a 2027 release, even as the rest of RDNA 5 slips to 2028, also signals a different strategic calculation: AMD appears more willing to ship a smaller, mid-range part early to defend its price-conscious market segment, whereas Nvidia has not signaled any equivalent early release for a scaled-down Rubin variant.
Historical context: how 2026 compares to the 2021 mining boom
Anyone who bought a GPU during the 2021 cryptocurrency mining boom has seen this movie before, but the plot is different this time. During the 2021 shortage, GPU component prices rose as much as 70%, with actual street prices for popular cards reaching 2.5x to 4x their MSRP at the worst of the shortage. That crunch was driven by a combination of pandemic-era supply chain disruption and a speculative wave of cryptocurrency mining demand that could, and eventually did, evaporate almost overnight once mining profitability collapsed and the Ethereum network moved off proof-of-work.
The 2026 shortage is structurally different. It is not being driven by a speculative bubble that can pop; it is being driven by sustained, multi-year capital investment in AI data centers by the largest technology companies in the world. Where the 2021 shortage resolved within roughly 12-18 months once mining demand fell away, the 2026 shortage is tied to a buildout that hyperscalers have publicly committed to sustaining through the end of the decade. That is the core reason analysts and outlets are pointing to 2028, not a shorter horizon, before consumer GPU supply and pricing normalize.
Market impact: gamers, system builders, and data centers
The practical effect for PC gamers is straightforward: budgets that used to buy a high-end card now buy a mid-range one, and mid-range budgets are increasingly landing on last-generation hardware. System builders assembling pre-built gaming PCs are absorbing higher GPU and memory costs directly into their bill of materials, and those increases are showing up in finished-system pricing across the market rather than staying isolated to boutique or enthusiast-tier builds.
Data center operators face a parallel but distinct pressure. Nvidia’s own AI server pricing is rising alongside its consumer GPU pricing, driven by the same memory input costs, which means the price of AI compute itself is climbing even as demand for that compute continues to grow. That dynamic reinforces why memory producers have little incentive to redirect capacity back toward consumer GDDR: AI server customers are absorbing cost increases that consumer GPU buyers are increasingly resistant to.
The data-center squeeze: AI servers vs. gaming cards
It is worth being precise about why gaming GPUs are losing this allocation fight. An AI accelerator built around HBM4 memory can sell for tens of thousands of dollars per unit to a hyperscaler with a multi-year purchase commitment. A gaming GPU using GDDR7 sells for a few hundred to a few thousand dollars to a consumer who can delay the purchase entirely. From a fab operator’s perspective, every wafer allocated to HBM4 production generates dramatically more revenue than the same wafer allocated to GDDR7, and that math does not change until either AI demand cools or new fab capacity comes online in volume.
TSMC’s own results have reflected this dynamic, with the company posting record revenue in 2026 driven heavily by AI chip demand even as the broader semiconductor supply chain, including NAND, tightened. Readers can find more detail in tech-insider.org’s earlier coverage of TSMC’s record revenue and the NAND shortage feeding into it.
What buyers can do right now
Buyers weighing a GPU purchase in late 2026 face an unusual calculus: waiting for a price drop is unlikely to pay off before 2028, but buying now means paying a price that has already risen sharply from launch MSRP. A few practical takeaways follow directly from the reporting above. First, current-generation cards are not going to get meaningfully cheaper before new architectures ship, since the memory shortage driving prices up is structural rather than cyclical. Second, mid-range cards have room to become relatively more attractive as flagship pricing outpaces mid-tier pricing on a percentage basis. Third, anyone planning a build around a specific budget should expect that budget to buy less GPU in 2027 than it does today, since further memory-driven price increases remain more likely than decreases while TrendForce’s own data shows contract prices still rising, just more slowly, through Q3 2026.
Predictions: what happens between now and 2028
Based on the pricing trajectory, memory supply data, and roadmap reporting gathered here, several outcomes look likely over the next two years.
- GPU prices will likely see at least one more round of increases in late 2026 or early 2027, following the same pattern of Nvidia moving first and AMD following within weeks.
- AMD’s AT2 chip will arrive in 2027 as a mid-range option, giving budget-conscious buyers a genuine upgrade path even as flagship RDNA 5 and Rubin cards wait for 2028.
- Memory contract price growth will likely continue moderating quarter over quarter through 2027, per TrendForce’s own trajectory, but moderation means smaller increases, not price cuts, until new fab capacity actually ships.
- Nvidia and AMD’s 2028 flagship launches will likely arrive close together, within the same few months, rather than staggered as in past generations, since both are gated by the same memory supply timeline.
- Data-center AI GPU pricing will keep climbing in parallel with consumer GPU pricing, since both draw from the same constrained memory supply chain, meaning relief for gamers is directly tied to when AI infrastructure demand growth finally slows.
Frequently asked questions
Why are GPU prices so high in September 2026?
A historic DRAM and NAND memory shortage, driven primarily by AI data-center demand, has pushed component costs sharply higher through 2026. Nvidia and AMD have each raised GPU prices multiple times this year in response, according to TrendForce and BigGo Finance reporting.
When will next-generation GPUs like the RTX 60 series launch?
Multiple outlets report that Nvidia’s Rubin-based RTX 60 series and AMD’s RDNA 5 lineup are being pushed to 2028, with AMD’s mid-range AT2 chip as the sole exception arriving in 2027.
Will GPU prices drop before 2028?
Reporting and memory-market data suggest a meaningful price correction is unlikely before 2028. TrendForce’s own data shows memory contract prices still rising, just at a slower pace, through Q3 2026, and new fab capacity is not expected in volume before late 2027 or 2028.
Is this shortage the same as the 2021 crypto mining GPU shortage?
No. The 2021 shortage was driven largely by speculative cryptocurrency mining demand that collapsed once mining profitability fell, resolving within roughly a year to 18 months. The 2026 shortage is driven by sustained AI data-center memory demand tied to multi-year infrastructure spending, which is why analysts expect it to persist longer, into 2028.
Why does AI demand affect gaming GPU prices?
Gaming GPUs and AI accelerators draw on overlapping memory production capacity. Fab operators can earn far more revenue allocating wafers to HBM4 memory for AI accelerators than to GDDR memory for gaming cards, so memory makers prioritize AI orders, tightening supply and raising costs for consumer GPUs.
Which GPUs have seen the biggest price increases in 2026?
Nvidia’s RTX 5090 has seen the steepest increase, rising from a $1,999 launch price to more than $5,000 at retail by September 2026. AMD’s RX 9070 XT has climbed from roughly $830 to over $1,037 following its Q3 2026 MSRP hike, according to BigGo Finance.
Should I buy a GPU now or wait?
If a purchase is needed for current use, waiting is unlikely to result in lower prices before 2028 based on available memory-market data. Buyers with flexibility may find mid-range cards a comparatively better value than flagship models, since flagship pricing has risen faster on a percentage basis.