Sofia Lindström
July 6, 2026
16 min read
The graphics card you priced out last winter costs meaningfully more today, and the reason has almost nothing to do with gaming. Through the first half of 2026, an artificial-intelligence buildout of unprecedented scale has swallowed the world’s memory supply, and the fallout has landed squarely on PC gamers. GPU prices are climbing across every tier, flagship cards are selling for roughly 50% over their launch MSRP, and memory chips now account for more than 80% of a GPU’s bill of materials, with RTX 5090 retail listings spiking to ~$4,300–$5,000. This is not a rumor cycle. It is a supply shock that memory makers, board partners, and console manufacturers have all confirmed in writing.
As of early July 2026, the picture is unambiguous: Nvidia and AMD have both pushed through phased price increases, Nintendo has raised the Switch 2, Sony has bumped the PS5 a second time, and Valve has hiked the Steam Deck by nearly 50%. Every one of those moves traces back to the same root cause – AI data centers outbidding consumer hardware for DRAM and graphics memory. This analysis breaks down exactly how much GPU prices have risen in 2026, why the memory crisis is structurally different from the 2021 shortage, what the hard data from TrendForce and IDC says, and when – if ever – relief arrives.
Google · Preferred Sources
Don’t miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions
Why Gaming GPU Prices Are Surging in 2026
The mechanics are simple even if the scale is staggering. A modern graphics card is mostly memory. Every RTX 50 or Radeon RX 9000 board ships with several gigabytes of high-speed GDDR6 or GDDR7, and the price of that memory has exploded. When the cost of the single most expensive component on a board doubles or triples, the finished product gets more expensive – full stop. That is the entire story of 2026 GPU prices in one sentence
What changed is demand. Generative-AI training and inference run on enormous banks of memory, and the hyperscalers building those clusters will pay almost any price to secure supply. Memory fabs are finite, so every wafer allocated to high-margin AI parts is a wafer not allocated to the commodity DRAM and graphics memory that consumer GPUs, consoles, and handhelds depend on. The result is a bidding war that gaming hardware was always going to lose
Board partners started signaling the squeeze early. AMD began adjusting Radeon RX 9000 pricing in January 2026, and Nvidia followed with RTX 50 increases in February. By spring, the phrase “RAMpocalypse” had migrated from enthusiast forums into mainstream tech coverage, and by summer the price tags on retail shelves had caught up with the headlines. For anyone tracking the Steam Hardware Survey, the effect on upgrade behavior is already visible – gamers are holding onto older cards far longer than they did a generation ago.
The AI Memory Shortage, Explained
To understand the GPU price surge, you have to understand the memory market underneath it. There are three dominant DRAM manufacturers on the planet – Samsung, SK Hynix, and Micron – and all three have pivoted hard toward high-bandwidth memory (HBM) and server DRAM for AI accelerators. Those products carry far fatter margins than the GDDR that ships on a gaming GPU, so the incentive to reallocate capacity is overwhelming
The scale of the reallocation is the part that shocks people. Analysts at IDC have forecast that AI data centers could consume as much as 70% of the world’s memory output in 2026, up from roughly 20% to 30% in 2022. When the majority of global memory production is being routed to server rooms, consumer devices are left fighting over a shrinking remainder. That is why this shortage touches everything from graphics cards to game consoles to the SSD in your gaming rig simultaneously – they all draw from the same well.
Supply cannot simply ramp to meet the demand, either. New memory fabrication capacity takes years and billions of dollars to bring online. According to IDC, “IDC expects 2026 DRAM and NAND supply growth to be 16% and 17% year-on-year, below historical norms.” With demand accelerating and supply growth constrained below trend, price is the only release valve – and it is being pulled hard
How Much RTX 50 Prices Have Risen
Nvidia’s GeForce RTX 50 series is the clearest barometer of the crisis because it uses cutting-edge GDDR7, the exact memory class in shortest supply. When Nvidia notified its add-in-card (AIC) partners of increases, the impact concentrated on the cards carrying the most memory. According to TrendForce, RTX 50 models equipped with 16GB or more of VRAM recorded price hikes of roughly 15% to 20%, while broader supply-chain increases across AMD and Nvidia ran 10% to 15% depending on memory configuration.
The flagship tells the starkest story. The RTX 5090 launched at a $1,999 MSRP, and Nvidia passed an estimated $300 increase to its board partners on the 5090 and 5090D V2 SKUs. Street prices moved further still: GPU price trackers logged the RTX 5090 selling around $2,999 in mid-2026 – roughly 50% above launch – with individual retail listings spiking north of $5,000. Leaks circulating since late 2025 have floated the possibility of the RTX 5090 reaching $5,000 by year-end, though that figure remains a projection rather than an official price.
Board partners moved at different speeds. MSI raised its RTX 50 pricing first, with ASUS and Gigabyte following. The table below summarizes where the major gaming GPUs stood as of early July 2026. All launch prices are the official MSRPs; the 2026 figures reflect adjusted pricing and reported street levels drawn from vendor notices and price-tracking data
| GPU model | Launch MSRP | 2026 status | Reported change | Memory |
|---|---|---|---|---|
| Nvidia RTX 5090 | $1,999 | ~$2,999 street; listings to $5,000+ | +~50% street; +$300 to AIC partners | 32GB GDDR7 |
| Nvidia RTX 5080 | $999 | Adjusted upward | +15–20% (16GB tier) | 16GB GDDR7 |
| Nvidia RTX 5070 Ti | $749 | Supply cut, price up | +15–20%; production reduced | 16GB GDDR7 |
| Nvidia RTX 5060 Ti 16GB | $429 | Sharp production cut | +15–20% | 16GB GDDR7 |
| AMD RX 9070 XT | $599 | Two hikes since launch | +10–15%; more planned | 16GB GDDR6 |
| AMD RX 9000 (spot, EU/China) | varies | Spot-market premium | +10–18% | GDDR6 |
AMD Radeon RX 9000 Faces Its Own Hikes
AMD’s Radeon RX 9000 line runs on GDDR6 rather than GDDR7, but that has offered little protection. GDDR6 costs have surged in parallel, and AMD was actually first to move. The company raised RX 9000 pricing by roughly $10 for every 8GB of VRAM in late 2025, then scheduled a second increase for January 2026. That put AMD ahead of Nvidia on the price curve, an unusual position for a company that has spent recent generations competing on value
The pressure has not let up. In the second half of 2026, supply-chain reporting out of Board Channels – a Chinese trade forum with a track record of anticipating AMD moves – pointed to another 10% to 15% RX 9000 price increase, with GDDR6 costs reported to have increased by 15–25% since late 2025. In the European and Chinese spot markets, TrendForce data showed AMD RX 9000 cards commanding 10% to 18% premiums as availability tightened. None of the second-half increases were officially confirmed by AMD at the time of reporting, but the direction of travel is not in question.
The strategic wrinkle for AMD is that its value proposition depends on undercutting Nvidia. When memory eats the same share of both companies’ costs, AMD loses room to maneuver on price – and the mid-range buyers who traditionally flock to Radeon are exactly the ones most sensitive to a $50 or $100 jump. That dynamic connects directly to the broader shift covered in our look at PC gaming versus console economics in 2026
GDDR7 Now Dominates a High-End Card’s Cost
Here is the statistic that reframes the entire discussion: on some high-end 2026 graphics cards, VRAM now accounts for more than 80% of the total bill of materials. The GPU die that Nvidia or AMD designs – the actual silicon everyone thinks of as “the graphics card” – has become a minority of the build cost. The memory strapped around it is the expensive part
GDDR7 pricing tells that story. By the second quarter of 2026, GDDR7 memory prices had risen roughly 15% to 20% price increase reported by PC vendors, and lead times stretched out for weeks as board partners scrambled to secure allocation. Because a flagship card like the RTX 5090 carries 32GB of the stuff, the per-card memory bill scales brutally: double the capacity, and you double your exposure to a memory market that has lost its mind. That is precisely why Nvidia’s 16GB-and-up SKUs took the sharpest increases while its 8GB cards were comparatively insulated.
The relationship between memory cost and card price is close to linear at the high end. A simplified model of how a memory-cost spike flows into a card’s price makes the mechanism obvious:
# Illustrative model: how a GDDR7 price spike flows into a GPU's price
vram_gb = 32 # RTX 5090-class card
base_mem_cost = 12.0 # $ per GB, late-2025 baseline
spike_multiple = 1.40 # +35–47% HBM in AI chips by Q2 2026; other_bom = 260 corrected to memory BOM increase of ~$1.23 per Gb in 2026 (up from $0.52 in 2025), a 138% jump.0 # die, PCB, cooler, VRM, assembly
mem_cost_2025 = vram_gb * base_mem_cost
mem_cost_2026 = mem_cost_2025 * spike_multiple
bom_2025 = mem_cost_2025 + other_bom
bom_2026 = mem_cost_2026 + other_bom
print(f"Memory share of BOM 2026: {mem_cost_2026 / bom_2026:.0%}")
print(f"BOM increase from memory alone: ${bom_2026 - bom_2025:.0f}")
# Values are illustrative, not vendor-published costs.The takeaway is not the exact dollars – vendors do not publish their component costs – but the shape of the curve. When memory is the dominant line item, a memory shock is a card-price shock, and no amount of manufacturing efficiency on the die side can offset it
Nvidia Cuts Supply and Shelves the Refresh
Faced with memory it cannot get at a sane price, Nvidia has done the rational thing for its margins and the frustrating thing for gamers: it has cut supply. Reports through the first half of 2026 described year-over-year production reductions ranging from 20% to as much as 40%, with the highest-VRAM cards hit first. TrendForce specifically flagged the RTX 5060 Ti 16GB and RTX 5070 Ti 16GB as facing the sharpest cuts, while the 8GB RTX 5060 and 5060 Ti received higher allocation to keep mainstream shelves stocked.
The clearest signal of how bad the constraint is: as of late June 2026, there was no confirmed desktop RTX 50 Super refresh. In a normal cycle, Nvidia would juice mid-generation sales with higher-VRAM “Super” variants. In 2026, adding more memory to a card is the last thing a company wants to do when memory is the scarce, expensive input. The refresh that enthusiasts expected has effectively been shelved by the math
This supply discipline has a knock-on effect. With fewer high-end cards in the channel and no refresh to reset expectations, the existing RTX 50 and RX 9000 inventory holds its elevated price. Scarcity plus strong demand equals sticky prices – a pattern that also shaped the debut of Nvidia’s data-center push detailed in our coverage of the Nvidia RTX Spark superchip, where the same AI demand pulling memory away from gaming is the engine of Nvidia’s record data-center revenue
What the Data Says: TrendForce and IDC Numbers
The anecdotes about pricier cards are backed by hard market data from the two firms that track memory and PC economics most closely. The figures are sobering. On the raw memory side, TrendForce documented a violent repricing at the start of the year: “TrendForce indicates average DRAM memory prices will increase by 50% to 55% in Q1 2026 compared to Q4 2025.” That was the conservative read
As demand data firmed up, the projection got worse. In revised numbers, “TrendForce’s revised Q1 2026 projections show conventional DRAM prices up 90 to 95% quarter-over-quarter.” A near-doubling of DRAM prices in a single quarter is the kind of move the memory market normally sees only during severe supply disruptions – except this time the disruption is demand, not a factory fire or an earthquake
The downstream impact on PCs is measurable too. IDC modeled how the memory crunch ripples into finished systems: “IDC estimates the PC market could contract by 4.9% to 8.9% in 2026, with PC average selling prices rising 4 to 8%.” A shrinking market with rising prices is the textbook signature of a supply-driven shock rather than a demand boom. There is a faint glimmer later in the year – Tom’s Hardware reported that “SIH (via TrendForce) projects conventional DRAM contract prices to rise 13% to 18% quarter-over-quarter in Q3 2026” – still an increase, but a decelerating one as consumers hit their affordability ceiling. The table collects the key data points.
| Metric | Figure | Period | Source |
|---|---|---|---|
| DRAM price change (conservative) | +50% to 55% | Q1 2026 vs Q4 2025 | TrendForce |
| DRAM price change (revised) | +90% to 95% QoQ | Q1 2026 | TrendForce |
| DRAM contract price change | +13% to 18% QoQ | Q3 2026 | SIH / TrendForce |
| GDDR7 memory price change | +~40% | By Q2 2026 | Supply-chain reports |
| DRAM / NAND supply growth | +16% / +17% YoY | 2026 | IDC |
| PC market volume | −4.9% to −8.9% | 2026 | IDC |
| PC average selling prices | +4% to 8% | 2026 | IDC |
The Price Shock Hits Consoles and Handhelds Too
This is where the story stops being a PC-only problem. Every modern game platform is a memory-dependent device, and the same DRAM bidding war that inflated GPU prices has forced console and handheld makers to raise prices mid-generation – something that essentially never happens in a healthy market. Hardware traditionally gets cheaper as a generation matures. In 2026, it got more expensive
Nintendo confirmed a $50 US increase on the Switch 2, from $449.99 to $499.99, effective September 1, 2026 (rising to $679.99 in Canada), citing medium-to-long-term market conditions widely understood to mean the DRAM shortage. Nintendo gave four months of notice – more warning than its rivals. Sony raised the PS5 a second time, to $649, in a move attributed to memory costs. Valve was the most aggressive: it lifted Steam Deck pricing by roughly $240 to $300 – close to a 50% jump on some configurations, pushing the OLED model to around $789 – with essentially no notice, explicitly blaming global memory and storage shortages. The parallel to the earlier gaming handheld RAM crisis is exact, only now it spans every platform at once.
| Game platform | Prior price | 2026 price | Increase | Effective |
|---|---|---|---|---|
| Nintendo Switch 2 (US) | $449.99 | $499.99 | +$50 (~11%) | Sep 1, 2026 |
| Nintendo Switch 2 (Canada) | – | $679.99 CAD | Raised | Sep 1, 2026 |
| Sony PlayStation 5 | lower | $649 | Second hike | 2026 |
| Valve Steam Deck OLED | ~$549 | ~$789 | +$240 (~44%) | May 2026 |
| Nvidia RTX 5090 (context) | $1,999 | ~$2,999 street | +~50% | 2026 |
The through-line matters for buyers weighing a platform. When a graphics card, a console, and a handheld all rise in price for the identical reason, there is no “safe” corner of gaming hardware to retreat to. For the numbers behind Nintendo’s momentum despite the hike, see our breakdown of why the Switch 2 topped 19.86 million units even as a PS6 handheld was hinted
2026 vs 2021: Why This Shortage Is Different
Gamers who lived through the 2020–2021 GPU drought are experiencing déjà vu, but the two events are structurally different, and the difference dictates how this one ends. The 2021 shortage was driven by a collision of speculative cryptocurrency mining, pandemic-era demand, and a broad semiconductor supply crunch. Cards sold for two to three times MSRP because scalpers and miners were competing with gamers for the same silicon. Crucially, that demand was fragile: when Ethereum moved off proof-of-work and crypto prices fell, mining demand evaporated almost overnight, and prices normalized through 2022.
The 2026 crisis has no equivalent pressure-release valve. AI data-center demand for memory is not speculative – it is backed by the capital budgets of the largest, most profitable technology companies on earth, and those budgets are growing, not shrinking. There is no “AI crash” analogous to a crypto crash on the near horizon that would suddenly free up memory capacity. The demand is structural, the buyers are price-insensitive, and the supply response is measured in years of fab construction, not months.
That is the uncomfortable conclusion: 2021 was a spike that reverted, while 2026 looks more like a re-leveling. The floor under GPU prices has moved up, and it may not come back down to old levels even after the acute shortage eases. The same forces reshaping component costs are visible in platform-longevity decisions like AMD extending the AM5 socket to 2029, where vendors are stretching hardware lifecycles to protect buyers from constant re-purchasing in a high-cost environment
Nvidia vs AMD vs Intel: Who Absorbs the Memory Bill
The three GPU vendors are navigating the crisis differently, and their choices reveal their competitive positions. Nvidia, sitting on record AI revenue, has the least incentive to protect the gaming market. Every gigabyte of GDDR7 it can secure competes internally with the far more lucrative HBM going into its data-center accelerators. Nvidia’s response – raise prices, cut supply, skip the refresh – is the behavior of a company for whom gaming is now a strategic afterthought relative to AI.
AMD is more exposed. Gaming remains a larger share of its consumer-facing identity, and its brand equity rests on value. AMD has passed costs through but has tried to do so in smaller, incremental steps – the $10-per-8GB structure early on, then percentage bumps. The company is walking a tightrope: raise prices too fast and it forfeits the value argument that is its whole reason for existing in the GPU market; raise too slowly and it sells cards below viable margin. Intel, the smallest player with its Arc line, uses GDDR6 and is equally exposed to the memory squeeze, but its strategy has been to hold the budget end where it can still win share – a segment where even small price increases sting buyers the most.
The net effect is a market where none of the three can offer relief. When the shared input cost dominates the bill of materials, competition on price collapses, because there is nothing left to compete with. That is a fundamentally different landscape from the one that produced the value wars of previous generations
When Will GPU Prices Come Down?
The honest answer is: not soon. The consensus across TrendForce and IDC is that meaningful relief depends on new memory fabrication capacity, and that capacity is not scheduled to reach volume production until 2027 at the earliest. Micron and SK Hynix are building, but fabs are slow. Most analyst forecasts place genuine consumer relief no sooner than 2028, with elevated pricing persisting through late 2027
There are tentative signs of deceleration. The Q3 2026 projection of 13% to 18% quarterly DRAM increases – down from the 50%-plus swings of Q1 – suggests the market is cooling as buyers hit an affordability wall. But “prices rising more slowly” is not “prices falling.” For gamers, the practical horizon is that 2026 and most of 2027 will remain a high-cost environment for any memory-heavy purchase, whether that is a graphics card, a console, or a handheld
Predictions for the Rest of 2026 and Beyond
- Prices stay elevated through 2027. With no new fab volume before 2027 and consumer relief pegged to 2028, expect the current high-cost regime to be the baseline for at least 18 more months.
- Vendors tilt toward low-VRAM SKUs. Expect more 8GB and 12GB cards to receive priority allocation while 16GB-plus models stay scarce and carry premium pricing.
- No desktop RTX 50 Super refresh in 2026. Adding memory to a card is uneconomical while memory is the constraint; the mid-cycle refresh is effectively shelved.
- The used-GPU market heats up. As new cards stay expensive, prior-generation and used GPUs will hold value far better than in a normal cycle, making the secondhand market unusually attractive.
- A new price floor persists. Even after the acute shortage eases, GPU prices are unlikely to fully revert to pre-2026 levels – the structural AI demand has permanently raised the cost of memory.
What PC Gamers Should Do Right Now
The buying calculus has inverted. In a normal year, waiting rewards patience with lower prices and better cards. In 2026, waiting largely means paying more later, because the trend line points up through 2027. That does not mean panic-buying a flagship – it means being realistic about what you actually need and buying it deliberately rather than hoping for a correction that the data says is not coming
Three practical moves make sense. First, if your current card still runs your games, hold it – the upgrade math has rarely been worse, and stretching a GPU another year is a legitimate strategy. Second, if you must buy, favor the sweet-spot SKUs that vendors are still allocating generously (the 8GB and 12GB tiers) rather than chasing scarce, premium-priced 16GB-plus cards. Third, take the used market seriously; prior-generation cards at fair prices are often the smartest value in a high-cost environment. A quick way to sanity-check any deal against current street prices before you commit:
# Sanity-check a GPU deal against its launch MSRP
model="RTX 5070 Ti"
msrp=749
asking=899
premium=$(python3 -c "print(f'{($asking-$msrp)/$msrp:.0%}')")
echo "$model – asking $$asking vs $$msrp MSRP = $premium over launch"
# Rule of thumb in 2026: anything under ~20% over MSRP is a good deal;
# 50%+ over MSRP means you are paying the AI memory tax in full.The broader point is to plan around a high-cost year rather than fight it. The memory crisis is bigger than any one product launch, and no amount of waiting will make an RTX 5090 cheap in 2026
Related Coverage
Frequently Asked Questions
Why are GPU prices going up in 2026?
Because AI data centers are consuming the majority of the world’s memory supply, driving up the cost of the DRAM and GDDR7 that graphics cards depend on. Memory has become the dominant cost in a modern GPU, so a memory price spike is a card price spike. Nvidia and AMD both pushed through phased increases starting in early 2026
How much have RTX 50 and RX 9000 prices increased?
According to TrendForce, RTX 50 cards with 16GB or more of VRAM rose roughly 15% to 20%, and broader AMD/Nvidia increases ran 10% to 15%. The RTX 5090 has sold around $2,999 on the street – about 50% over its $1,999 launch MSRP – with some listings above $5,000. AMD RX 9000 cards saw multiple hikes and 10% to 18% spot-market premiums in Europe and China
When will GPU prices come back down?
Not before 2027, and meaningful relief may not arrive until 2028. New memory fabrication capacity from Micron and SK Hynix will not reach volume production until 2027 at the earliest. Q3 2026 data shows prices rising more slowly (13% to 18% quarterly versus 50%-plus earlier in the year), but slowing increases are not the same as price cuts
Is now a good time to buy a graphics card?
If your current card still handles your games, holding is reasonable – the upgrade value is poor right now. If you must buy, the trend points up, so waiting is unlikely to save money. Favor well-stocked 8GB and 12GB SKUs over scarce, premium 16GB-plus cards, and take the used market seriously as a value option
Why does AI cause a memory shortage for gamers?
There are only three major DRAM makers, and all three prioritize high-margin AI memory (HBM and server DRAM) over the GDDR used in consumer GPUs. IDC has forecast AI data centers could consume as much as 70% of global memory output in 2026. Every wafer routed to AI is one unavailable for gaming hardware, and supply cannot expand fast enough to compensate
Are game consoles and handhelds affected too?
Yes. The Nintendo Switch 2 rose $50 to $499.99 (effective September 1, 2026), the PS5 was raised to $649 in a second hike, and Valve increased the Steam Deck by roughly $240 to $300. All three cited the same memory and storage shortage that is driving GPU prices, because consoles and handhelds depend on the same DRAM supply
Will there be an RTX 50 Super refresh in 2026?
As of late June 2026, no desktop RTX 50 Super refresh was confirmed. A “Super” refresh typically adds VRAM, which is exactly the scarce, expensive component right now – making a higher-memory refresh uneconomical. Nvidia has instead cut production of high-VRAM cards and prioritized lower-memory SKUs to keep mainstream shelves stocked
