If you tried to buy a graphics card this month, you already know: GPU prices in 2026 are the highest they have ever been. The cheapest new RTX 5090 now costs about $6,900, more than three times its $1,999 launch price. An RTX 5080 costs about twice its MSRP. And in the first week of November, board partners announced yet another round of increases. This article explains what is happening, why it is happening, how long it is likely to last, and what you can do about it if you need a GPU for AI or gaming right now.
- November 2026 hike: Chinese board partner Colorful raised cost prices by roughly $330-$550 on the RTX 5090D v2, $380-$450 on the RTX 5080 and $150 on the RTX 5070 Ti (Wccftech, 3 Nov).
- US street prices: RTX 5090 from about $6,900, RTX 5080 about $2,000, RTX 5070 Ti about $1,400, RTX 5060 Ti 16 GB about $770.
- Main cause: a global memory shortage. AI data centres buy HBM and DRAM faster than it can be made, so GDDR7 for gaming cards got scarce and expensive.
- RTX 50 Super: still not launched; the latest rumour puts a 24 GB RTX 5080 in Q1 2027.
- Outlook: memory makers and analysts do not expect real relief before 2027-2028, and some say 2030.
- What to do: buy only the VRAM you need, consider AMD/Intel 32 GB cards or used cards, and rent cloud GPUs for big jobs.
GPU Prices Right Now
These are the lowest new-card prices in the US from Tom’s Hardware’s GPU price tracker (update of 3 November 2026, as reported by Tech Insider), compared with NVIDIA’s launch prices:
| GPU | Launch MSRP | Lowest price, Nov 2026 | Markup |
|---|---|---|---|
| RTX 5090 32 GB | $1,999 | $6,919 | +246% |
| RTX 5080 16 GB | $999 | $1,999 | +100% |
| RTX 5070 Ti 16 GB | $749 | $1,399 | +87% |
| RTX 5070 12 GB | $549 | $899 | +64% |
| RTX 5060 Ti 16 GB | $429 | $769 | +79% |
Many listings are far higher. Wccftech’s 3 November report shows Newegg RTX 5090 listings from $6,919.99 up to $11,896, an ASUS ROG Astral RTX 5090 on Amazon at $7,998 and an MSI RTX 5080 at $2,422.56. AMD is not immune: Wccftech says the RX 9070 XT is approaching $1,000, and Intel’s Arc Pro B70 (launched at $949) has roughly doubled.
The November 2026 Price Increase
On 3 November 2026, Wccftech reported new cost-price increases from Chinese board partners, based on internal sources and a post by leaker Harukaze5719. Colorful’s list:
| Card | Increase (CNY) | Increase (approx. USD) |
|---|---|---|
| RTX 5090D v2 | 2,200-3,300 yuan | $330-$550 |
| RTX 5080 | 2,500-2,800 yuan | $380-$450 |
| RTX 5070 Ti | 1,000 yuan | $150 |
| RTX 5070 | 600-900 yuan | $90-$135 |
| RTX 5060 Ti 16 GB | 250-400 yuan | $37-$60 |
| RTX 5060 Ti 8 GB | 100-150 yuan | $15-$22 |
| RTX 5060 8 GB | 150-200 yuan | $22-$30 |
Galax was also expected to raise RTX 5060-and-above prices by at least 200 yuan, and KFA2 had already added $30 to the RTX 5060. Price increases at board partners in China usually reach US and European shelves within weeks, so expect the table above to look optimistic by December.
Why GPUs Are So Expensive
1. The global memory shortage
This is the big one. Memory makers (Samsung, SK hynix and Micron) have shifted wafer capacity to HBM, the stacked memory used on AI accelerators like NVIDIA’s B200, B300 and Rubin. HBM uses far more wafer area per bit than normal DRAM, so every HBM stack made means less DDR5, LPDDR5X and GDDR7 for everyone else. Demand from AI data centres, including OpenAI’s Stargate project, is enormous. The results:
- DRAM prices reportedly rose 172% during 2025.
- Conventional DRAM contract prices rose about 93-98% quarter-over-quarter in Q1 2026 (Engadget, citing industry data).
- TrendForce forecast on 30 September 2026 that conventional DRAM contract prices would rise another 10-15% in Q4 2026.
- Micron reported record revenue of $54.23 billion for its fiscal Q4 2026 and $133.19 billion for fiscal 2026, up from about $37 billion the year before. Memory makers are making record profits; buyers are paying for them.
Graphics cards are hit twice: GDDR7 itself is more expensive, and the new 3 GB GDDR7 chips that were supposed to give us 24 GB mid-range cards are the most contested of all. I cover the RAM side in detail in why RAM is so expensive in 2026.
2. NVIDIA’s priorities
Every wafer NVIDIA puts into a gaming GPU could instead become a datacenter GPU that sells for many times more. NVIDIA has said it secured adequate memory supply, but its supply goes first to the products with the highest margins. Gaming supply is thin, and thin supply means retailers and scalpers set prices.
3. AI buyers compete with gamers
The RTX 5090 is the cheapest way to get 32 GB of very fast memory on one card, so local AI users, small labs and cloud marketplaces buy them too. The same happened to the RTX 3090 and 4090. A card that earns $0.50-$1.00 an hour on a GPU rental marketplace is worth a lot more to a host than to a gamer.
4. Tariffs and currency
US import tariffs on electronics and component costs add to the final price, and partner price lists in China are in yuan, so currency moves pass straight through.
What Happened to RTX 50 Super?
The RTX 50 Super refresh was meant to fix the VRAM problem by using 3 GB GDDR7 chips: a 24 GB RTX 5080 Super, and 18-24 GB 5070-class cards. According to eTeknix’s timeline, reports in November 2025 said the cards might be cancelled because of memory shortages, January 2026 reports said the launch was delayed indefinitely, a leaker said in June 2026 that they were back on track, and in July 2026 the cards reportedly reached partners but the launch was held over memory pricing. The latest rumour (10 October 2026) puts a 24 GB RTX 5080 in Q1 2027. NVIDIA has not confirmed any of this. Do not plan a purchase around it.
Workstation and AI Boxes Are Up Too
- NVIDIA DGX Spark: launched at $3,999 for 128 GB; ServeTheHome reported in October 2026 that the price is now $6,950, and a new 64 GB model starts at $4,999.
- RTX PRO 6000 Blackwell 96 GB: listed at $8,565 in 2025; Spheron reports about $16,000 retail in October 2026.
- RTX PRO 5000 72 GB: lowest US listing about $9,400 (gpuprix, 4 November 2026).
- Radeon AI PRO R9700 32 GB: launched at $1,299; street price $1,700-$1,850 in late October.
Even the 32 GB AMD and Intel cards are still much cheaper per gigabyte than NVIDIA’s gaming flagship. I compare them in the best workstation GPUs for AI.
The Used Market
High new prices pull used prices up. A used RTX 3090 (a 2020 card) averaged about $1,463 on eBay in September 2026, up 13.8% in 30 days, according to OpenClaw’s analysis of 291 listings. That is close to its original $1,499 launch price. Used cards still make sense for AI because nothing new gives you 24 GB at 936 GB/s for less, but there is no bargain left. My used RTX 3090 guide lists what to check before you buy one.
When Will GPU Prices Drop?
Honest answer: not soon. The memory shortage is the root cause, and new memory fabs take years to build. A Kearney analysis cited on Wikipedia expects the shortage to last until at least 2030, and SK Group’s chairman has said the same. Tech Insider reports that relief for GPUs has been pushed to 2028. Things that could bring prices down sooner:
- A slowdown in AI data centre spending, which would free memory capacity quickly.
- New fabs from Samsung, SK hynix and Micron coming online in 2027-2028.
- NVIDIA launching RTX 50 Super or RTX 60 cards with more VRAM at set prices.
- Software that needs less memory (better quantization, MoE models and memory compression techniques).
RAM kit prices did fall nearly 30% between March and April 2026 after a wave of memory-saving AI research, before rising again, which shows the market can move fast in both directions. But planning on a big drop in the next six months is a gamble.
What to Do If You Need a GPU Now
- Work out the VRAM you actually need. Use my VRAM guide. Many people buy a 5090 when a 16 GB card would run their models fine.
- Buy VRAM, not brand. For LLMs, a 32 GB R9700 at $1,700-$1,850 gives you as much memory as an RTX 5090 at a quarter of the price. See the best GPU for local LLMs.
- Rent for big jobs. An RTX 5090 costs $0.51-$0.89 an hour on Vast.ai and $1.19 on RunPod. At $6,919, buying only pays off after roughly 6,000-13,000 rental hours. Compare options in the cheapest cloud GPUs for AI.
- Consider a Mac or unified-memory box for large models. Apple’s prices have held better than GPU prices. See Mac vs PC for local AI.
- Avoid scalper listings. Use price trackers and buy from retailers with returns.
Rent vs Buy: The Break-Even Maths
Using the lowest RTX 5090 price above ($6,919) and the rental prices from my cloud GPU guide:
| Rental price per hour | Hours to equal the purchase price | At 8 hours a day, that is |
|---|---|---|
| $0.51 (cheap Vast.ai listing) | About 13,570 hours | About 4.6 years |
| $0.89 (typical Vast.ai) | About 7,770 hours | About 2.7 years |
| $1.19 (RunPod) | About 5,810 hours | About 2 years |
This ignores electricity (a 5090 at 575 W uses about 4.6 kWh in an 8-hour day) and resale value, which push in opposite directions. If you use a GPU only a few hours a week, renting is clearly cheaper right now; if it runs most of the day, buying a cheaper 16 to 32 GB card often makes more sense than either.
How to Buy a GPU Safely in a Shortage
- Track prices: use PCPartPicker price history, retailer stock alerts, and Amazon price trackers such as Keepa or camelcamelcamel before you buy.
- Buy from the retailer, not a marketplace seller, where possible. Check “Sold by” and “Ships from” on Amazon and Newegg.
- Use buyer protection: pay by credit card or PayPal Goods and Services, never by bank transfer, gift cards or crypto.
- Beware of too-good prices: a new 5090 listed far below the market is almost always a scam or an empty box.
- Check warranty terms: some brands only honour warranty for the original buyer or for authorised retailers.
- Test immediately: run a stress test and a real AI workload inside the return window. My used RTX 3090 guide has a testing checklist that works for any card.
Frequently Asked Questions
Why are GPU prices going up in November 2026?
Board partners raised prices again because GDDR7 memory costs keep rising. The cause is a global memory shortage driven by AI data centres buying HBM and DRAM.
How much is an RTX 5090 now?
The lowest new US listings were about $6,919 on 3 November 2026, with many listings between $7,000 and $11,900. Its launch MSRP was $1,999.
Will GPU prices go down in 2027?
Some relief is possible if RTX 50 Super cards launch or AI spending slows, but most analysts expect memory to stay tight until 2028 or later.
Should I wait for the RTX 50 Super?
Only if you can wait without a date. It has been delayed several times, and the latest rumour points to Q1 2027 for a 24 GB RTX 5080. NVIDIA has confirmed nothing.
Is it cheaper to rent a GPU than to buy one?
For most people, yes, at today’s prices. Renting an RTX 5090 or H100 by the hour beats buying unless you use the card many hours every day for years.
Is it a good time to buy a GPU in November 2026?
Only if you need one now. Prices are far above launch MSRP, so if you can wait, rent for big jobs and watch price trackers; if you need a card today, buy the VRAM you need rather than the most expensive model.
Final Thoughts
GPU prices in 2026 are not a short spike. They are the consumer side of the biggest memory crunch in the industry’s history, and AI demand is the cause. If you need a card, buy for the VRAM you need, look at AMD and Intel 32 GB cards, and rent for anything bigger. I will update this page as prices move.
Sources
- Wccftech: NVIDIA RTX 50 GPUs see a massive price increase in November (3 Nov 2026)
- Tech Insider: GPU price tracking 2026, citing Tom’s Hardware (5 Nov 2026)
- Tech Insider: GPU relief pushed to 2028
- Wikipedia: 2024-present global memory supply shortage
- Engadget: How much RAM cost in 2020 vs 2026 (2 Nov 2026)
- TrendForce: Q4 2026 DRAM and NAND price forecast (30 Sep 2026)
- Micron fiscal Q4 2026 results press release (SEC filing)
- eTeknix: RTX 50 Super hub (13 Oct 2026)
- ServeTheHome: DGX Spark 64 GB and 128 GB price increase
- OpenClaw: used RTX 3090 eBay price data (24 Sep 2026)
