Why are GPU prices going up in 2026?

Because AI data centers absorbed the memory supply graphics cards rely on. Memory makers moved wafer capacity to high-bandwidth memory and server DRAM, GDDR spot prices roughly tripled since autumn 2025, and the fixed-price contracts that shielded card makers through 2025 expired. Nvidia and AMD have both raised the GPU-and-memory kit prices they charge board partners as a result.

How much have Nvidia GPU prices increased in 2026?

There have been three increases: roughly 10% to 15% in January (15% to 20% on 16GB-plus cards), a narrower one in May limited to the RTX 5090 line, and 20% to 30% in late July across all GeForce RTX models. Compounded, a mid-range card is roughly 32% to 50% more expensive than in January. An Asus TUF RTX 5080 went from $1,199 to $1,595, or 33%.

Did Nvidia officially raise MSRP?

No. The reported increases are to the GPU-and-memory kits Nvidia sells board partners, not to official list prices. The RTX 5090 still carries a $1,999 MSRP from its January 2025 launch while street prices run past $4,000, which is why sticker prices look stable while what you actually pay does not.

Should I buy a GPU now or wait for prices to drop?

Buy now if you need a card, particularly at the entry tier where prices sit closest to MSRP. Waiting only pays if you can skip the purchase entirely. Retail has not yet passed through the full July kit increase, memory contract prices are still rising, and no credible forecast has meaningful relief before 2027 to 2028.

Why is VRAM so expensive in 2026?

GDDR competes for the same wafer capacity as high-bandwidth memory for AI accelerators, which earns several times more per wafer. Consumer GDDR6 spot prices went from about $2.50 to about $7.50 per GB since autumn 2025. Denser parts are worse: a 2GB GDDR7 module runs about $20 while a 3GB module costs $60 to $70, so 50% more capacity costs over three times as much.

Are AMD and Intel raising GPU prices too?

Yes. AMD notified board partners including Sapphire, ASUS, XFX, and Vastarmor of a roughly 10% increase on GPU-and-GDDR kits effective July 2026, its second in six months, and its RX 9000 series rose 10% to 18% in Europe and China in January. Intel Arc uses GDDR6 and is exposed to the same shortage. There is no vendor that escapes it.

When will GPU prices come down?

Not before 2027, and 2028 is the more common industry estimate. Relief requires new memory fab capacity, which takes 18 to 24 months to build, or AI memory demand cooling enough to free existing capacity. Prices are also sticky downward, so the eventual floor is likely above the pre-AI level rather than back at it.

GPU Prices in 2026: Why They Rise, and Whether to Buy

Graphics card prices in 2026 are not rising because of a shortage of graphics chips. They are rising because the memory soldered next to the chip got repriced by the AI buildout, and Nvidia has now passed that cost down to its board partners three separate times this year. The third pass, reported in late July at 20% to 30%, is the one that reaches every GeForce RTX model rather than just the flagship. Stack it on January’s increase and a mid-range card is 32% to 50% more expensive than it was in January, before any retailer adds a cent.

Key takeaways

The 2026 increases are a memory pass-through, not a GPU shortage. Spot GDDR6 went from about $2.50 per GB to about $7.50 per GB since autumn 2025, and the fixed-price memory contracts that held card prices steady through 2025 expired.
What Nvidia repriced is the GPU-plus-memory kit it sells board partners, not the official MSRP. That distinction is why sticker prices look stable while street prices do not.
The hikes compound. January was reported at 10% to 15% (15% to 20% on 16GB-plus cards) and July at 20% to 30%, which is roughly +32% to +50% versus January on a card that was not part of the May flagship-only increase.
Retail has absorbed less than the kit hikes imply so far. An Asus TUF RTX 5080 went from $1,199 to $1,595, or +33%, which sits at the bottom of the compounded band. The unabsorbed remainder is the risk to a buyer who waits.
Denser memory is punitively priced. A 2GB GDDR7 chip runs about $20 and a 3GB chip $60 to $70, so 50% more capacity costs more than three times as much. That is why more VRAM is not coming to rescue you.
Relief is a 2027 to 2028 story at the earliest, because it depends on fab capacity that takes 18 to 24 months to build.

Why are GPU prices rising in 2026?

GPU prices are rising in 2026 because AI data centers absorbed the memory supply that graphics cards depend on. Memory makers shifted wafer capacity to high-bandwidth memory and server DRAM, GDDR spot prices tripled, and Nvidia and AMD raised the price of the GPU-and-memory kits they sell to card makers. The silicon is not scarce. The memory is.

The mechanism matters, because it tells you what would have to change for prices to fall. Every graphics card ships with GDDR memory: GDDR7 on the RTX 50 series, GDDR6 on the previous generation and on most of AMD’s and Intel’s lineups. That memory comes off the same wafer capacity that produces the HBM stacks bolted to AI accelerators, and HBM earns several times more per wafer. I walked through why the memory makers made that choice, and why they are in no hurry to unmake it, in the 2026 memory shortage explained.

Through 2025 that squeeze did not reach gamers, for a boring contractual reason: card makers were buying memory on fixed-price agreements struck before the AI boom. Those agreements expired. TrendForce reports consumer-grade GDDR6 spot prices moving from roughly $2.50 per GB to roughly $7.50 per GB since autumn 2025, a threefold increase, and that is the number that started landing in retail once the contracts rolled off.

20-30%

July 2026 kit price hike

Economic Daily News; all GeForce RTX

GDDR6 spot, per GB

about $2.50 to about $7.50 since autumn 2025

+33%

Asus TUF RTX 5080, street

$1,199 winter 2025-26 to $1,595 July 2026

>80%

memory share of GPU bill of materials

TrendForce News, January 2026

The three 2026 hikes, and why they compound

Reporting on each hike has treated it as a standalone event, which understates the damage. The increases are multiplicative, and they land on the same cards.

January came first. TrendForce reported Taiwanese card makers raising RTX 50 prices by 10% to 15% depending on memory configuration, with 16GB-and-above models at the steeper 15% to 20% end, and AMD’s RX 9000 series up 10% to 18% in Europe and China. May brought a narrower increase aimed at the RTX 5090 line. July is the broad one.

Compounded effect of the three 2026 Nvidia GPU kit price hikes on a 16GB RTX 5080
Stage	Reported hike	Compounded low	Compounded high	Change vs baseline
Winter 2025-26	baseline	$1,199	$1,199	baseline
January hike	+10% to +15%	$1,319	$1,379	+10% to +15%
May hike	RTX 5090 line only	$1,319	$1,379	+10% to +15%
July hike	+20% to +30%	$1,583	$1,793	+32% to +50%
Reported July street price	observed street price	$1,595	$1,595	+33%

Modelled street price of a 16GB RTX 5080 through the three 2026 kit-price hikes. Each stage is a band rather than a line because every hike was reported as a range, and the ranges multiply. The gold rule is the reported street price of an Asus TUF card, which lands at the bottom of the compounded band. The price axis starts at $1,000, not zero.Source: Hike ranges per TrendForce (January) and Economic Daily News via Notebookcheck (July); street prices as reported

Read the chart the other way round and it becomes a forecast. The compounded floor for this card is about $1,583. It is selling at $1,595. Every dollar of the remaining band, up to about $1,793 on the reported high end, is pass-through that board partners and retailers have not charged yet. Using the 16GB-specific January figure of 15% to 20% instead of the broad range pushes the modelled band to +38% to +56%, which puts the observed price below the floor rather than at it.

That is the single most useful fact in this whole story for anyone deciding what to do. The retail price you see today is not the top of the current cost stack. It trails it.

What Nvidia actually repriced

Precision matters here, because most coverage says “Nvidia raised GPU prices” and that is not quite what happened.

Nvidia does not set retail prices. It sells board partners such as ASUS, MSI, and Gigabyte a bundle of the GPU die plus the GDDR memory that goes with it, commonly called a kit. The July increase is a 20% to 30% rise in the price of that kit, and Taiwanese reporting describes it plainly as Nvidia declining to absorb GDDR7 inflation and pushing it downstream instead. AMD did the same thing on a smaller scale, notifying Sapphire, ASUS, XFX, and Vastarmor of a roughly 10% kit increase effective July 2026, its second in six months.

Three consequences follow from the kit structure, and they explain the odd shape of this market:

Official MSRPs stay put while street prices move. The RTX 5090 still carries a $1,999 list price from its January 2025 launch, alongside $999 for the RTX 5080, $749 for the RTX 5070 Ti and $549 for the RTX 5070. Those numbers now describe a card nobody can buy at that price. MSRP has become a marketing artifact.
Small board partners get squeezed hardest. A kit price is take-it-or-leave-it. Partners with thin margins and no memory-buying leverage of their own have the least room to absorb an increase, which is how a hike to the channel becomes a hike at the shelf within weeks.
Older cards are not a refuge. The July increase covers GDDR6 kits as well as GDDR7, so the previous generation gets repriced alongside the current one. There is no last-gen escape hatch of the kind that usually exists in a hardware price spike.

How much of a graphics card is now memory

TrendForce News, reporting on Nvidia’s shift toward 8GB configurations in January, put memory at more than 80% of a GPU’s bill-of-materials cost. That figure is worth interrogating rather than repeating, because the published chip prices do not obviously support it.

Supply-chain reporting carried alongside the July hike puts a 2GB GDDR7 module at about $20 and a 3GB module at $60 to $70. Take the 2GB figure, count the modules a card needs, and the memory line looks like this:

Card	VRAM	2GB GDDR7 modules	Memory cost at ~$20 per module	Official MSRP	Memory as share of MSRP
RTX 5070	12GB	6	~$120	$549	~22%
RTX 5070 Ti	16GB	8	~$160	$749	~21%
RTX 5080	16GB	8	~$160	$999	~16%
RTX 5090	32GB	16	~$320	$1,999	~16%

Memory at roughly $10 per GB is real money, and it is roughly triple what the same capacity cost in autumn 2025, so the increase alone is about $100 to $215 per card depending on capacity. On a $999 list price that is ten to twelve points of gross margin, which comfortably justifies a 10% to 15% increase. It does not by itself justify 20% to 30%.

So either the modelled chip price is conservative, or part of the hike is margin recapture, or the expensive 3GB modules are doing more work in the mix than the arithmetic above assumes. Probably some of each. What the 80% claim most plausibly describes is memory’s share of the incremental bill of materials, not the total, since the GPU die itself remains the single largest line item.

The 3GB module price is the detail worth staring at. Fifty percent more capacity for more than three times the price is not a supply curve, it is rationing. It also kills the obvious consumer wish, which is more VRAM per dollar: a vendor that doubled memory capacity at current module prices would add more cost than the entire margin on a mid-range card. Nvidia’s answer has gone the other way, pivoting to 8GB versions of the RTX 5060 and 5060 Ti because four modules are cheaper to source than eight. Buyers get the same sticker and less memory, the same trick PC makers have been pulling on RAM and storage, which I covered in what Apple’s price hike means for local AI.

If you are buying a card specifically to run models locally, that scarcity is aimed directly at you, because VRAM capacity is the binding constraint. Check what you actually need first with how much RAM it takes to run a local LLM and the can I run this LLM checker, because overbuying capacity is expensive in a way it was not two years ago.

Buy now or wait?

The honest answer depends on which of these you are, and on one asymmetry: the downside of waiting is priced (you pay more later), while the upside is speculative (relief depends on fab capacity that does not exist yet).

Should you buy a GPU now or wait, by buyer situation in mid-2026
Dimension	Buy now	Wait for Q4 2026	Buy used or last-gen
Card died, need one now	Only real option	Months with no GPU	Fine without warranty
Working card, want an upgrade	Only near MSRP	Probably dearer	Best value on offer
Budget build, entry tier	Tier closest to MSRP	Most hike still to come	Last-gen is the floor
Need 24GB-plus for local AI	Worst premiums	3GB modules stay dear	Cheapest route per GB
Current card is adequate	Peak price, no need	Avoids the spend	Cheap incremental step

Buy-now versus wait, by buyer situation. Green favours the buyer, amber is a trade-off, red is a drawback. The pattern is consistent: waiting is only the right call when the spend can be avoided entirely, not when it is merely deferred.Source: Capital & Compute analysis

One counterweight deserves stating, because the sensible version of “wait” rests on it. Retail premiums have not been rising in a straight line. TechSpot’s quarterly survey of ten regions found the market well above MSRP but no longer deteriorating each month, with the entry tier close to list price and the steepest premiums concentrated on the flagship. TrendForce likewise expects memory price gains to moderate through the third quarter as consumer demand weakens against a high base, even while AI server demand keeps the floor up. Demand destruction is real: at some price, buyers simply stop, and consumer hardware is closer to that point than the data-center market is.

That is a case for the increases slowing. It is not a case for prices falling.

What would actually bring GPU prices down

Four things would have to happen, in roughly this order of plausibility.

Contract prices stop climbing. TrendForce’s July guidance has server DRAM contract prices rising 13% to 18% quarter over quarter in the third quarter, with long-term agreements shielding some large cloud buyers and everyone else paying the increase. Samsung has reportedly been seeking up to 20% higher DRAM prices for the quarter, per ZDNet Korea. Those are server figures rather than graphics figures, but they set the direction of the market that GDDR competes inside.

New wafer capacity arrives. A memory fab takes 18 to 24 months to build, which puts anything decided this year into 2027 at the earliest. China’s CXMT is the most-watched swing factor on the supply side; I sized up whether it can actually move the market in what CXMT is.

AI memory demand cools. This is the real switch, and nothing in current capital-expenditure guidance suggests it flips soon.

Consumers refuse to pay. Already happening at the margin, and it is the fastest-acting of the four, but it works by shrinking the market rather than lowering the cost of a card.

For the fuller timeline on the memory side of this, including who gets relief first, the forecast in when RAM prices will drop applies directly to graphics memory too, and the current chip and retail readings live on the memory price tracker. The HBM tracker covers the product that started all of this. Current MSRP against reported street price, model by model, is on the GPU price tracker.

Bottom line

The GPU market in 2026 is not broken by scarcity of graphics silicon. It is being repriced by the cost of the memory that sits beside it, and that cost is set in a market where a gamer bids against a hyperscaler and loses. Three kit-price increases in seven months compound to something between a third and a half more expensive than January, retail has not finished passing it through, and the denser memory that would make a card worth the money is priced to discourage exactly that.

If you need a card, the arithmetic says buy the cheapest one that does the job, now, at the tier closest to list price. If you do not need one, the same arithmetic says the best available return is not buying at all.

Frequently asked questions

Why are GPU prices going up in 2026?: Because AI data centers absorbed the memory supply graphics cards rely on. Memory makers moved wafer capacity to high-bandwidth memory and server DRAM, GDDR spot prices roughly tripled since autumn 2025, and the fixed-price contracts that shielded card makers through 2025 expired. Nvidia and AMD have both raised the GPU-and-memory kit prices they charge board partners as a result.
How much have Nvidia GPU prices increased in 2026?: There have been three increases: roughly 10% to 15% in January (15% to 20% on 16GB-plus cards), a narrower one in May limited to the RTX 5090 line, and 20% to 30% in late July across all GeForce RTX models. Compounded, a mid-range card is roughly 32% to 50% more expensive than in January. An Asus TUF RTX 5080 went from $1,199 to $1,595, or 33%.
Did Nvidia officially raise MSRP?: No. The reported increases are to the GPU-and-memory kits Nvidia sells board partners, not to official list prices. The RTX 5090 still carries a $1,999 MSRP from its January 2025 launch while street prices run past $4,000, which is why sticker prices look stable while what you actually pay does not.
Should I buy a GPU now or wait for prices to drop?: Buy now if you need a card, particularly at the entry tier where prices sit closest to MSRP. Waiting only pays if you can skip the purchase entirely. Retail has not yet passed through the full July kit increase, memory contract prices are still rising, and no credible forecast has meaningful relief before 2027 to 2028.
Why is VRAM so expensive in 2026?: GDDR competes for the same wafer capacity as high-bandwidth memory for AI accelerators, which earns several times more per wafer. Consumer GDDR6 spot prices went from about $2.50 to about $7.50 per GB since autumn 2025. Denser parts are worse: a 2GB GDDR7 module runs about $20 while a 3GB module costs $60 to $70, so 50% more capacity costs over three times as much.
Are AMD and Intel raising GPU prices too?: Yes. AMD notified board partners including Sapphire, ASUS, XFX, and Vastarmor of a roughly 10% increase on GPU-and-GDDR kits effective July 2026, its second in six months, and its RX 9000 series rose 10% to 18% in Europe and China in January. Intel Arc uses GDDR6 and is exposed to the same shortage. There is no vendor that escapes it.
When will GPU prices come down?: Not before 2027, and 2028 is the more common industry estimate. Relief requires new memory fab capacity, which takes 18 to 24 months to build, or AI memory demand cooling enough to free existing capacity. Prices are also sticky downward, so the eventual floor is likely above the pre-AI level rather than back at it.

Sources

TrendForce News (2026). VRAM Shortage Reportedly Drives NVIDIA’s RTX 50 Price Hikes at Taiwan’s GPU Card Makers This Month (January 20 2026). Secondary, citing Commercial Times; January hike of 10-15%, and 15-20% on 16GB-plus cards. Verified 2026-07-30. trendforce.com
TrendForce News (2026). NVIDIA Reportedly Pivots to 8GB RTX 5060 amid Memory Crunch, Older GPUs Back in Play (January 14 2026). Secondary; 8GB pivot, module counts, memory as over 80% of GPU bill of materials. Verified 2026-07-30. trendforce.com
TrendForce News (2026). AMD Reportedly Raises GPU-GDDR Kit Prices by ~10% for AIB Partners, Effective Jul. 2026 (July 3 2026). Secondary, citing MyDrivers and Board Channels; AMD kit hike, named partners, GDDR6 spot at about $2.50 to $7.50 per GB. Verified 2026-07-30. trendforce.com
TrendForce (2026). Long-Term Agreements Cap Price Increases; Server DRAM Contract Prices Expected to Rise 13-18% QoQ in 3Q26 (press release, July 9 2026). Primary research; server DRAM contract forecast and the long-term-agreement split. Verified 2026-07-30. trendforce.com
TrendForce (2026). AI Server Demand Continues to Support Memory Prices in 3Q26, but Gains Moderate as Consumer Demand Weakens and High Base Effects Take Hold (press release, July 3 2026). Primary research; moderating consumer demand. Verified 2026-07-30. trendforce.com
TrendForce News (2026). Samsung Reportedly Seeks Up to 20% 3Q26 DRAM Price Increase; LPDDR Hikes May Exceed 20% (July 3 2026). Secondary, citing ZDNet Korea; Samsung is seeking, not agreed. Verified 2026-07-30. trendforce.com
Notebookcheck (2026). Nvidia is expected to raise GeForce RTX GPU prices again by up to 30% (July 28 2026). Secondary, reporting Taiwan’s Economic Daily News; the 20-30% third hike, the Asus TUF RTX 5080 street prices. Verified 2026-07-30. notebookcheck.net
T客邦 / TechBang (2026). GDDR7 shortage: NVIDIA passes costs to board partners. Secondary; the kit mechanism, RTX 5090 MSRP unchanged at $1,999 against $4,000-plus street prices. Verified 2026-07-30. techbang.com
Tom’s Hardware (2025). Nvidia announces RTX 5090 for $1,999, 5070 for $549. Secondary; official RTX 50 launch MSRPs. Verified 2026-07-30. tomshardware.com
TechSpot (2026). GPU Pricing Remains Broken, Even if It Has Stopped Getting Worse (Q2 2026 pricing survey, ten regions). Independent tracking; retail premiums over MSRP by tier. Verified 2026-07-30. techspot.com