Why GPUs are so expensive in 2026 - and whether to buy now or wait

If you've priced a graphics card lately and flinched, you're not imagining it. In 2026 the whole market is running above sticker — from around 5% over MSRP on budget cards to roughly 77% on the flagship — and the reason isn't the usual crypto-boom story. It's memory. This page explains exactly why prices climbed, how bad it is by model, whether it gets better in 2026 or 2027, whether waiting for the next generation makes sense, and — with a clear rule — when to just buy. If you already know you're buying, skip to the value picks.

How bad are GPU prices right now?

Here's the reality by model. The percentages are how far above MSRP each card was selling in spring 2026, from TechSpot's first-party tracking across ten countries; the street prices are approximate mid-2026 US figures and move week to week.

Two patterns matter. First, where you live changes the price — the global average was around 16% over MSRP, but the US ran closer to 22%. Second, the markup scales with memory: the flagship 32 GB RTX 5090 is worst-hit, while AMD's RX 9070 XT — which uses older, less contested GDDR6 — barely moved. Prices also stopped rising through spring 2026 as demand softened, but "stopped getting worse" is not "getting better."

Why GPUs got so expensive: the AI memory shortage

The cause is a single chain: AI datacenters need enormous amounts of memory, that memory pays better than the memory on a gaming card, so the memory makers build the profitable kind first — and gamers lose the bidding war.

A modern GPU is mostly memory

On a high-end graphics card, VRAM is reportedly more than 80% of the bill of materials — by some industry estimates around $820 of an RTX 5090's cost is memory alone. The price of 16 GB of GDDR7 reportedly jumped from roughly $65-80 in mid-2025 to over $200 by the end of the year once long-term supply contracts expired and cards were exposed to spot pricing. When the most expensive component doubles, the card doubles down with it.

AI is outbidding gamers for DRAM and HBM

The three companies that make almost all the world's memory — Samsung, SK Hynix and Micron — are steering capacity toward high-bandwidth memory (HBM) and server DRAM for AI, which carry far higher margins than consumer GDDR7. Analyst firm TrendForce describes a memory "supercycle," with DRAM prices projected to rise more than 70% in 2026 and contract prices already up an estimated 90-95% quarter-on-quarter in early 2026 — roughly a doubling. IDC has estimated AI datacenters could absorb around 70% of global memory output in 2026 (reportedly, up from 20-30% a few years ago), while total DRAM supply grows only about 16% for the year. HBM also consumes several times the wafer area per gigabyte of standard memory, so even moderate AI demand claims an outsized share of the fabs.

One footnote worth flagging honestly: in June 2026 a class-action antitrust lawsuit was filed in US federal court alleging the three memory makers coordinated to restrict supply. Those allegations are unproven, and the companies haven't been found liable — but it's a real filing, and a sign of how far prices have moved.

NVIDIA's response: fewer cards, no refresh

Rather than sell gaming GPUs at thin margins during a memory crunch, NVIDIA has prioritised AI silicon. Reports put gaming-GPU production cuts at 30-40% in 2026, and NVIDIA has effectively confirmed a GeForce supply shortage running through its 2027 fiscal year — unsurprising when its datacenter business now dwarfs gaming. The rumored RTX 50 "SUPER" refresh, which would have added VRAM to the mid-range, has been repeatedly reported as delayed or cancelled, because it needs the 3 GB GDDR7 modules that are the most constrained memory part of all. The fix for the VRAM squeeze is stuck behind the VRAM squeeze.

Why this isn't 2021 (and why that matters)

The last GPU price crisis, in 2020-2021, was driven by crypto mining demand stacked on a pandemic supply crunch — and when mining profitability collapsed, prices fell fast. This time is different in a way that matters for your wallet: the demand is AI, and it's structural. Hyperscalers are signing multi-year, multi-billion-dollar memory and compute commitments; that demand won't evaporate the way a mining bubble did. So don't plan around a sudden 2021-style correction — the more realistic path is a slow easing as new fab capacity comes online, not a crash.

It's not just GPUs: consoles, handhelds, RAM and SSDs

Because the shortage is about memory, it reaches everything that uses memory. Reported 2026 moves include the Nintendo Switch 2 rising about $50 to $499.99, the PS5 climbing toward $649, and the Steam Deck OLED up roughly 44% to around $789. System DDR5 RAM and SSDs have climbed too, since they draw on the same DRAM and NAND fabs. If you're building or upgrading a whole PC, budget for the memory and storage to cost more than they did a year ago — not just the GPU.

Will GPU prices drop in 2026 or 2027?

The honest forecast: not much in 2026, with real relief unlikely before 2027-2028. TrendForce and IDC both tie any loosening to new memory-fab capacity that isn't expected at volume until 2027, and Intel has pointed to a similar 2027-2028 window. There was one hopeful signal — DRAM price increases were reported to be decelerating by the third quarter of 2026 (still rising, just more slowly) — but decelerating increases are not decreases. Planning a purchase around a near-term price collapse is planning around something no forecaster is currently predicting.

Should you wait for the next generation (RTX 60 / RDNA 5)?

This is the question the rumor mill muddies, so here's what's actually known. NVIDIA has effectively confirmed no new gaming GPUs launch in 2026. For the RTX 60 "Rubin" generation, reporting genuinely splits: TechRadar has floated a 2027 launch, while Tom's Hardware, citing The Information, reports 2028. AMD's RDNA 5 is rumored for late 2027 to early 2028. We track the details on our RTX 60 series and AMD RDNA 5 pages.

But for a buying decision the disagreement barely matters, because both answers lead to the same place: the next generation is roughly 12 to 30 months away, with no confirmed price — and it will launch into the same tight memory market, so its prices will be shortage-inflated too. Waiting a year or two to pay elevated prices for an unconfirmed card is not a savings strategy. The next generation is a reason to wait only if you don't need a card at all right now.

So — buy now or wait? The decision rule

Strip away the noise and it comes down to a short checklist.

The rule of thumb behind the checklist: under ~20% over MSRP is a reasonable deal in this market; 50%+ means you're paying the full AI tax. And because a used last-generation card sidesteps the current-gen memory premium entirely, the used market (a clean RTX 4070 Ti Super, RTX 4070, or RX 7800 XT from a reputable seller) is unusually attractive this cycle.

If you're buying anyway: the shortage-smart value picks

If you've decided to buy, the move is to dodge the memory premium where you can. These three current cards give the most performance per dollar in this market; for the full ranked field with specs and pricing, see our best GPUs of 2026, and for local-AI builds our best GPU for local AI.

Prices move — the links go to the live Amazon listing so you see current pricing and stock. The RX 9070 XT leads because its GDDR6 memory is the least exposed to the shortage; the Arc B580 gets you 12 GB on a budget without the 8 GB trap; the RTX 5070 Ti is the NVIDIA feature-set value pick once it's near MSRP. Whatever you choose, aim for 12 GB of VRAM as a floor and skip the flagship at today's premium. Torn between the top two? See our RTX 5070 Ti vs RX 9070 XT head-to-head.

Frequently asked questions

Why are GPU prices going up in 2026?

A global memory shortage. AI datacenters are buying up DRAM and high-bandwidth memory (HBM) at prices consumer graphics cards can't match, so the three big memory makers (Samsung, SK Hynix, Micron) are steering capacity toward server memory. On a modern high-end GPU, memory is reportedly more than 80% of the bill of materials, so when memory prices roughly doubled in early 2026, card prices followed. NVIDIA has also cut gaming-GPU production to prioritise far more profitable AI chips.

How much have RTX 50 and RX 9000 prices increased?

As of spring 2026, graphics cards averaged around 16% over MSRP globally (worst in the US at roughly 22%), per TechSpot's multi-country tracking. The spread is wide: the RTX 5090 has run about 77% over its $1,999 MSRP (street $3,500+), while the RTX 5080 and 5070 Ti sat around 23% over, the RTX 5070 about 11%, and AMD's RX 9070 XT only about 8% over. AMD's RDNA 4 cards use GDDR6 rather than the scarcer GDDR7, which is why they held closest to MSRP. All figures are approximate and move week to week.

Will GPU prices drop in 2026 or 2027?

Probably not meaningfully in 2026. Analyst firms TrendForce and IDC, and comments from Intel, all point to 2027 at the earliest for memory supply to loosen, with genuine relief closer to 2028 - because new memory-fab capacity isn't expected at volume until 2027 and AI demand is structural, not a passing spike. Prices stopped getting worse through spring 2026 (demand softened), but 'stopped rising' is not the same as 'falling.'

Should I buy a GPU now or wait?

If your current card can't run what you play and the model you want is close to MSRP, buy now - waiting is unlikely to save money before 2027-2028, and the next generation won't launch cheaper. If your current card is adequate, or the only options are 16 GB+ cards at 30%+ over MSRP, waiting (or buying last-gen/used) is defensible. The one clear 'don't buy' is a flagship at its shortage peak: an RTX 5090 at 77% over MSRP is the weakest-value moment in the stack.

Should I wait for the RTX 60 series or next-gen GPUs?

Not as a way to save money. NVIDIA has effectively confirmed no new gaming GPUs launch in 2026, and reports on the RTX 60 'Rubin' generation split between 2027 (optimistic) and 2028 (the better-sourced read, from The Information). AMD's RDNA 5 is rumored for late 2027 to early 2028. So the next generation is roughly 12-30 months away with no confirmed price - and it will launch into the same tight memory market, so its prices will also be shortage-inflated. Waiting for next-gen is not a price strategy.

Why does AI cause a GPU shortage for gamers?

Graphics cards and AI accelerators compete for the same memory. AI datacenters need enormous quantities of DRAM and HBM, and that memory carries much higher margins than the GDDR7 on a gaming card, so memory makers allocate wafers to the AI side first. IDC has estimated AI datacenters could absorb around 70% of the world's memory output in 2026 (up from 20-30% a few years ago). Since memory is the biggest single cost on a high-end GPU, gamers effectively bid against hyperscalers for the same chips - and lose.

Is 8 GB of VRAM enough in 2026?

It's increasingly a real limitation. 8 GB cards still run many games at 1080p, but modern titles at higher settings, ray tracing, and texture packs now spill past 8 GB, causing stutter or forced-lower textures. For a card you want to keep a few years, 12 GB is the sensible floor and 16 GB the target - which is why value picks like the 12 GB Arc B580 and 16 GB RX 9070 XT look better than an 8 GB card at a similar price.

Are consoles and handhelds affected too?

Yes - the same memory shortage touches everything with RAM. Reported 2026 moves include the Nintendo Switch 2 rising about $50 to $499.99, the PS5 climbing toward $649, and the Steam Deck OLED up roughly 44% to around $789. System RAM (DDR5) and SSDs have also risen, because they draw on the same DRAM and NAND supply. There is no fully insulated corner of gaming hardware right now.

Is there going to be an RTX 50 Super refresh?

It's uncertain and, if it happens, delayed. NVIDIA never officially confirmed a SUPER refresh, and reporting through 2026 has swung between 'cancelled' and 'delayed,' with CES 2027 floated as the earliest realistic window. The reason is almost poetic: the SUPER cards were rumored to use 3 GB GDDR7 modules - the single most supply-constrained memory part - so the product meant to fix the VRAM squeeze needs the exact component the shortage made scarce.

What's the best value GPU to buy right now?

For most buyers, AMD's RX 9070 XT is the standout - it uses GDDR6 (less exposed to the shortage), so it has held closest to MSRP while offering 16 GB and strong 1440p/4K performance. On a budget, the 12 GB Intel Arc B580 sidesteps the 8 GB ceiling for around $260-300. If you want NVIDIA's feature set (DLSS, CUDA for creator work), the RTX 5070 Ti is the value pick once it's near MSRP. The RTX 5090 is the one to avoid as a pure gaming buy at current prices.

Is it worth buying a used GPU in 2026?

It's a genuinely good strategy this cycle, because the shortage has kept last-generation cards desirable. A used RTX 4070 Ti Super, RTX 4070, or RX 7800 XT can deliver strong value without paying the current-gen memory premium - just buy from a reputable seller with return protection, and prefer cards that weren't obviously run hard. Last-gen VRAM tiers (12-16 GB) still cover the vast majority of games.

The bottom line

GPU prices in 2026 are high for a structural reason — AI has made memory scarce — and that reason won't resolve on the timeline of a crypto crash. Prices most likely hold elevated into 2027-2028, the next generation is a year or two out and won't launch cheaper, and the nearest relief valve (the RTX 50 SUPER refresh) is stuck behind the same shortage. So don't wait for a rescue that forecasters aren't predicting: if you need a card, buy the best value near MSRP now — favour GDDR6 and 12-16 GB parts, lean on the used market, and skip the flagship at its peak. If your current card is fine, this is a perfectly good year to sit tight.

Prices and availability cited here are approximate and current as of July 2026; graphics-card street prices move week to week, so treat every figure as a snapshot and check the live listing before buying. Market data draws on TrendForce, IDC and TechSpot's price tracking; next-generation timelines are rumor-tagged and unconfirmed.

Related buyer's guides

All reviews →

Buyer's guide / PC builds

Best gaming PC under $1,500 in 2026

The honest $1,500 build: a 16GB RX 9060 XT, Ryzen 5 and 32GB DDR5 for strong 1080p-ultra and entry-1440p gaming — with the real shortage-adjusted total, where every dollar goes, and whether a prebuilt beats it this year.

See the build →

Buyer's guide / PC builds

Best gaming PC under $2,000 in 2026

The value sweet spot: near-flagship 1440p gaming built around the RX 9070 XT and an X3D CPU — with the honest 2026 total, budget allocation, and exactly how to duck under a hard $2,000.

See the build →

Buyer's guide / PC builds

Best gaming PC under $3,000 in 2026

Where 4K becomes real: the fastest gaming CPU paired with an RTX 5080 for 1440p-max and 4K-high with DLSS 4 — the honest shortage-adjusted total, where every dollar goes, and a matched monitor.

See the build →

Buyer's guide / PC builds

Best gaming PC under $5,000 in 2026

The honest high-end build: a fully-maxed RTX 5080 creator-gaming machine with a 16-core 9950X3D, 64GB and 4TB — plus the truth about why $5,000 does not reliably buy a 5090 at 2026 street prices.

See the build →

Buyer's guide / AI PCs

LLM VRAM calculator

Enter a model size, quantization and context length to estimate the VRAM or unified memory an LLM needs to run locally — and instantly see which 2026 machines can hold it.

Use the calculator →

Buyer's guide / AI PCs

How to run LLMs locally: the 2026 hardware guide

How much VRAM or unified memory you need to run 7B-120B models locally, the capacity-vs-bandwidth tradeoff, and the 2026 hardware that hits each point on the curve — GB10, Ryzen AI Max, Apple Silicon and RTX GPUs.

Read the guide →