Desktop graphics card shipments climbed to 12.5 million units in the second quarter of 2026, the highest quarterly total since the first quarter of 2022, according to data from Jon Peddie Research reported by Tom’s Hardware on September 11, 2026. The number lands at an odd moment for the PC market: street prices for add-in boards are elevated, memory is scarce, and shoppers are buying anyway. Nvidia captured roughly 90% of that desktop discrete-GPU volume, leaving AMD with about 8% and Intel’s Arc lineup with around 2%, based on the same shipment tracking.
The shipment record is not a sign that the GPU shortage 2026 story is easing. It is closer to the opposite: buyers are rushing to lock in graphics cards before the next round of price hikes lands, and the memory squeeze behind those hikes is getting worse, not better. Intel CEO Lip-Bu Tan told an audience at Splunk’s .conf26 broadcast that memory prices have already risen five to sevenfold, and that Intel itself can currently meet only about 50% of customer CPU demand because of the same supply crunch. That is the backdrop against which gamers are still finding a way to buy GPUs in record numbers.
What the Q2 2026 shipment numbers actually show
Jon Peddie Research’s quarterly add-in-board tracking puts desktop discrete GPU shipments at 12.5 million units for Q2 2026, up roughly 5.9% from the prior quarter and about 7.8% year over year. That is the strongest reading in four years. Widen the lens to include notebook GPUs and integrated graphics, and total PC graphics processor shipments reached 75.5 million units in the same quarter, a 10.4% sequential gain though only a modest 1.1% increase from a year earlier. Discrete GPU shipments specifically, across desktop and notebook combined, rose 12.2% quarter over quarter and 14.1% year over year.
Those figures matter because they cut against the usual narrative that shortages suppress volume. Instead, the 2026 pattern looks like buyers pulling purchases forward. Tom’s Hardware framed it plainly: shipments are rising despite record-high prices, with the increase tied to shoppers trying to beat looming price spikes rather than a sign that supply constraints have lifted. TweakTown’s coverage of the same JPR data pegged first-half 2026 shipments at 24.3 million units combined, with Nvidia supplying an estimated 11.25 million of the 12.5 million Q2 desktop units.
The gap between the two readings on the exact multiplier (some coverage cites a 5.9% sequential gain, other outlets frame it closer to 10%) reflects the usual noise between different research cuts of the same underlying JPR dataset. The consistent thread across every version is a four-year high in unit volume paired with a near-record concentration of that volume in Nvidia’s column.
Nvidia’s 90% share and what it means for AMD and Intel
A 90% share of desktop discrete GPU shipments is close to a ceiling for a two-and-a-half-player market. AMD’s roughly 8% and Intel Arc’s roughly 2% leave very little room for a shortage to knock Nvidia off its perch. That said, the picture isn’t static month to month. Tech Insider reported that during the first week of September 2026, AMD briefly outsold Nvidia at Mindfactory, one of Europe’s largest PC parts retailers, taking around 56% of unit sales there against Nvidia’s roughly 40%. That single-retailer, single-week swing shows how price-driven the current buying pattern is: when Nvidia’s flagship cards spike in price, budget-conscious buyers pivot hard to Radeon, at least until the next restock.
AMD’s near-term position isn’t purely a pricing opportunity, though. PC Guide reported in mid-September that AMD’s graphics card partners could face roughly 10% higher chip costs by the end of 2026, attributed to rising TSMC wafer pricing, with the likely outcome being another Radeon MSRP increase in the fourth quarter. That would remove some of the price-gap advantage AMD has been using to pull share away from Nvidia in retail snapshots like the Mindfactory data point.
Data table: Q2 2026 desktop discrete GPU market at a glance
| Metric | Q2 2026 figure | Comparison |
|---|---|---|
| Desktop discrete GPU shipments | 12.5 million units | Highest since Q1 2022 |
| Sequential shipment growth | ~5.9% | vs. Q1 2026 |
| Year-over-year shipment growth | ~7.8% | vs. Q2 2025 |
| Total PC GPU shipments (desktop + notebook + integrated) | 75.5 million units | +10.4% QoQ, +1.1% YoY |
| Nvidia desktop discrete share | ~90% | ~11.25 million units |
| AMD desktop discrete share | ~8% | Radeon RX 9000 series |
| Intel Arc desktop discrete share | ~2% | Smallest of the three |
| H1 2026 combined desktop shipments | 24.3 million units | Q1 + Q2 2026 |
Source: Jon Peddie Research data as reported by Tom’s Hardware.
The memory shortage driving the price spikes
None of this shipment data happens in a vacuum. The root cause sits one layer down in the supply chain: GDDR7 and DRAM output. Samsung, SK Hynix, and Micron have reportedly redirected wafer capacity toward AI-datacenter HBM production, specifically the ramp of HBM4, rather than the GDDR7 modules gaming GPUs need. A semiconductor industry briefing from Distill Intelligence, dated September 11, 2026, put Samsung’s and SK Hynix’s memory inventories below 10 days of supply as that shift accelerated, a level that industry watchers describe as an extreme-pricing trigger point.
Intel’s Tan gave the clearest public numbers on how far prices have already moved. Speaking at Splunk’s .conf26 Global Broadcast, he said memory prices had risen roughly five to sevenfold, and that many hardware projects are being delayed industry-wide because companies simply cannot secure enough memory at any price, according to Sedaily’s report on the September 16, 2026 briefing. He added that Intel is currently able to meet only about 50% of customer CPU demand, a figure TrendForce also captured in its own writeup of the same remarks.
That reallocation of fab capacity toward HBM for AI accelerators is the same dynamic shattered.io covered when Nvidia’s Rubin Ultra platform lost a third of its planned memory allocation to the HBM shortage, and it is the same dynamic behind RAM now accounting for as much as 60% of the bill of materials on budget phones and laptops. Consumer GPUs, phones, and laptops are all drawing from the same shrinking pool of DRAM and GDDR wafer capacity that data-center AI buildouts are consuming at scale.
Nvidia’s RTX 5090 as the flagship case study
Nvidia’s GeForce RTX 5090 is the clearest single example of the price distortion. The card carries a listed price of $1,999 and ships with 32GB of GDDR7 memory. Street prices for the RTX 5090 reportedly climbed to roughly $3,500 to $4,000 or more earlier in 2026 as GDDR7 supply tightened, and by the time it disappeared from most US retail listings later in the year, tracked prices had pushed toward the $9,500 mark, as shattered.io reported when the card effectively vanished from US shelves. A flagship card selling for four to five times its MSRP, while shipment volumes simultaneously set a four-year record, is the paradox at the center of the current GPU shortage 2026 story: scarcity and record unit sales are happening at the same time, because buyers are treating every purchase as a race against the next price increase rather than a discretionary upgrade.
How this compares to the 2020-2022 GPU shortage
Veteran PC builders will recognize the shape of this story, even if the mechanism is different. The 2020-2022 shortage was driven by a mix of pandemic-era demand, crypto mining, and general semiconductor capacity constraints across the whole chip industry, not memory specifically. That earlier crunch also pushed shipment volumes down at times, since fabs simply could not produce enough finished GPUs, not just enough memory chips to pair with them.
The 2026 version is narrower and, in some ways, more mechanical: the GPU dies themselves are not the bottleneck, the memory that ships alongside them is. That is why shipment volume can hit a four-year high (fabs can still build the logic chips) while prices still spike (nobody can source enough GDDR7 or HBM to pair with them). It is also why the fix looks different this time. There is no crypto-demand collapse waiting to free up supply, because the competing demand now comes from AI data centers that are not going away. Acer’s CEO has publicly argued that memory makers are overstating how long the shortage will last to protect margins, projecting PC prices could ease by late 2027 as cheaper Chinese memory capacity comes online, per Tom’s Hardware’s archive coverage. Intel’s Tan, by contrast, has offered no such relief timeline and has said the shortage will get worse before it gets better.
Market impact: what this means for PC builders and retailers
For PC builders, the practical read is uncomfortable. Waiting for prices to normalize has not been a winning strategy in 2026, because every quarter of waiting has coincided with another round of increases rather than relief. That is precisely the behavior the JPR shipment data is capturing: people buying now, at inflated prices, because they expect the alternative (buying later) to cost even more. It is the same logic that has pushed AMD’s Radeon pricing higher and that shattered.io tracked around the launch of Nvidia’s RTX Spark platform earlier this year, where new hardware entered a market already primed for price sensitivity.
Retailers, meanwhile, are sitting on a genuinely unusual inventory dynamic. Sell-through is strong (hence the record shipment numbers) but margins on memory-heavy SKUs like the RTX 5090 are being squeezed from the supplier side even as street prices climb, because the retailers themselves are paying more for GDDR7-equipped boards before marking them up further. Component distributors and system integrators building pre-built gaming PCs face a similar squeeze, and several have already begun re-pricing bill-of-materials contracts quarterly instead of annually to keep pace.
Competitive comparison: Nvidia vs. AMD vs. Intel positioning
| Vendor | Desktop discrete share (Q2 2026) | Near-term pressure | Notable 2026 move |
|---|---|---|---|
| Nvidia | ~90% | GDDR7 scarcity on flagship SKUs (RTX 5090) | RTX Spark launch, DLSS 5 rollout across RTX 50 series |
| AMD | ~8% | Reported ~10% TSMC chip cost increase into Q4 | Briefly outsold Nvidia at Mindfactory in early September |
| Intel | ~2% | Meeting only ~50% of CPU customer demand per CEO Lip-Bu Tan | Arc line remains a distant third in discrete GPUs |
The picture that emerges is one where Nvidia’s dominance in discrete GPUs is not seriously threatened by the shortage, since it has the pricing power and allocation priority to absorb higher memory costs and pass them to buyers who keep purchasing anyway. AMD’s opportunity window (undercutting Nvidia on price to grab share, as it briefly did in Germany) is narrowing as its own component costs rise. Intel’s constraint is less about discrete GPU share, which was always small, and more about the company’s broader CPU supply chain being squeezed by the same memory shortage, a dynamic shattered.io detailed when Tan first warned there would be no relief until 2028.
Why the RTX 60 delay makes this shortage last longer
Adding to the pressure: there is no next-generation consumer GPU on the near horizon to reset pricing expectations. Reporting attributed to hardware leaker Kepler_L2 and covered in Tom’s Hardware’s archive says Nvidia’s next-gen RTX 60 family may not arrive until 2028, with the company reportedly prioritizing data-center Rubin-generation products over a near-term consumer refresh. AMD’s RDNA 5 architecture has reportedly been pushed to a similar 2028 timeframe as well, with one exception, an AMD chip codenamed AT2, which shattered.io covered in its report on the RDNA 5 and Rubin delays.
That timeline matters for the shortage story because a new generation typically resets the price ladder: last-gen cards get discounted to clear inventory, and buyers who were priced out get a fresh option. With RTX 60 and RDNA 5 both pushed out to 2028, the RTX 50 series and RX 9000 series will likely remain the current generation for roughly two more years, meaning today’s inflated prices have no obvious reset point until the memory shortage itself eases rather than a new architecture arriving to reshuffle the market.
AI demand is the structural reason this shortage is different
It’s worth separating cause from symptom here. The symptom is expensive, scarce gaming GPUs. The cause is that memory manufacturers are allocating capacity to HBM for AI accelerators ahead of GDDR7 for gaming cards, because AI data-center customers are willing to pay more per wafer and are signing multi-year supply commitments that consumer GPU makers generally do not match. Nvidia’s own AI chip roadmap, including the Rubin Ultra platform’s reported memory shortfall, competes directly with Nvidia’s own gaming division for the same upstream HBM and DRAM capacity from Samsung, SK Hynix, and Micron.
This is a structurally different shortage than 2020-2022 because it isn’t cyclical demand (crypto booms end, pandemics end) but a standing, multi-year commitment of fab capacity to a different, higher-margin product category. That is why Acer’s more optimistic 2027 timeline and Intel’s more pessimistic no-relief-until-2028 framing both agree on the underlying cause even as they disagree on the end date.
Historical context: how GPU shipment cycles have moved before
Desktop discrete GPU shipments last peaked near current levels in Q1 2022, at the tail end of the pandemic-and-crypto-driven boom, before crashing through 2022 and 2023 as crypto mining collapsed and channel inventory normalized. The 2024-2025 period saw a slower, steadier recovery in unit volumes as gaming demand rebuilt without crypto’s distortion. The 2026 spike back to four-year-high levels is unusual precisely because it isn’t being driven by a speculative secondary market like mining; it’s being driven by consumers pulling forward purchases to avoid future price increases, a behavior pattern more commonly seen in categories like housing or commodities than in consumer electronics.
Predictions: what happens next in the GPU market
- Expect at least one more Radeon MSRP increase in Q4 2026, given the reported ~10% rise in AMD partners’ TSMC chip costs, which would narrow AMD’s price advantage over Nvidia.
- Nvidia’s desktop discrete share is unlikely to drop meaningfully below its current ~90% through the rest of 2026, barring a sustained, multi-week AMD price advantage rather than isolated retailer snapshots like the September Mindfactory data.
- Memory-inventory warnings (Samsung and SK Hynix reportedly below 10 days of supply) suggest GDDR7 and DRAM pricing pressure continues at least through Q1 2027, regardless of whether the broader shortage ends in 2027 or 2028.
- With RTX 60 and RDNA 5 both reportedly delayed to 2028, expect continued high demand for current-gen RTX 50 and RX 9000 cards well into 2027, since there is no next-gen refresh to wait for.
- Watch Chinese memory capacity additions closely: Acer’s 2027 relief timeline depends specifically on cheaper Chinese DRAM output coming online, which is the single variable most likely to move the end date earlier or later.
What buyers should actually do right now
Given the data, timing a GPU purchase around an expected price drop has not worked for most of 2026. Every quarter of delay has coincided with either a new price hike or continued scarcity rather than relief. Buyers who need a card for a specific build timeline are generally better served buying into the current generation now, at current prices, rather than waiting for a reset that multiple industry voices, including Intel’s own CEO, say isn’t coming before 2028 at the earliest. Buyers who can wait indefinitely have more room to sit out the cycle, since the eventual arrival of RTX 60 and RDNA 5, whenever it lands, should still reset pricing on the tier below it the way it always has historically.
Related
- RTX 5090 Vanishes From US Stores, Hits $9,500 [2026]
- RDNA 5, Rubin GPUs Slip to 2028; AT2 Ships 2027 [2026]
- Intel’s Tan: Memory Up 7x, No Relief Until 2028 [2026]
- RAM Now 60% of Phone Cost, Sparks Fake-Chip Checks [2026]
- Rubin Ultra Loses 33% of Its Memory to HBM Shortage [2026]
Frequently asked questions about the 2026 GPU shortage
Why are GPU shipments at a 4-year high if there’s a shortage?
Because the shortage is in memory (GDDR7 and DRAM), not in GPU logic chips. Fabs can still build the processor dies at scale, so shipment volume rises, while prices climb because the memory paired with those chips is scarce and expensive. Buyers are also pulling purchases forward to beat further price increases, which pushes unit volume higher even as affordability worsens.
What percentage of the GPU market does Nvidia control in 2026?
Nvidia held approximately 90% of desktop discrete GPU shipments in Q2 2026, according to Jon Peddie Research data reported by Tom’s Hardware. AMD held roughly 8% and Intel’s Arc lineup held about 2%.
How much has the RTX 5090 price increased in 2026?
The RTX 5090 carries a listed price of $1,999 with 32GB of GDDR7 memory. Street prices reportedly climbed to roughly $3,500 to $4,000 earlier in 2026, and tracked pricing later pushed toward $9,500 as the card grew scarce in US retail.
When will the GPU memory shortage end?
Estimates vary. Acer’s CEO has projected PC prices could ease by late 2027 as cheaper Chinese memory capacity comes online. Intel CEO Lip-Bu Tan has offered a more pessimistic view, saying the shortage will worsen before it improves, with no clear relief signaled before 2028.
Is now a good time to buy a graphics card?
Based on 2026 pricing trends, waiting has generally not resulted in lower prices, since successive quarters have brought further increases rather than relief. Buyers with an immediate build timeline are typically better off buying current-generation hardware now rather than waiting for a price reset that multiple industry executives say is still years away.
Will AMD’s RDNA 5 or Nvidia’s RTX 60 series fix pricing?
Not soon. Reporting attributed to hardware leaker Kepler_L2 indicates both Nvidia’s RTX 60 series and AMD’s RDNA 5 architecture have been pushed back to around 2028, with AMD’s AT2 chip cited as a possible earlier exception. Until a new generation arrives, current RTX 50 and RX 9000 series cards are likely to remain the primary options, with pricing tied to memory costs rather than a generational refresh.
Why is AI demand blamed for gaming GPU prices?
Memory manufacturers including Samsung, SK Hynix, and Micron have reportedly redirected wafer capacity toward HBM production for AI data-center accelerators, which pay a premium over consumer-grade GDDR7. That reallocation reduces the GDDR7 and DRAM supply available for gaming GPUs, phones, and laptops, pushing prices up across all three categories simultaneously.
