Why AI PCs and graphics cards cost so much in 2026, and what to do about it
Quick answer
Memory. AI data centers buy huge amounts of high-bandwidth memory, and memory makers moved factory capacity to it and to server memory, leaving less for PCs. Memory contract prices rose 58% to 63% in a single quarter in spring 2026, and are forecast to rise another 13% to 18% in the third quarter, according to TrendForce. AI PCs are hit hardest because their selling point is lots of fast memory: in September 2026 we found graphics cards at 31% to 119% above their launch prices and the RTX 5090 at more than three times its list price.
If you have priced a computer for AI in 2026, you will have noticed: everything with a lot of fast memory costs far more than it did a year earlier, and graphics cards sell well above their launch prices. The reason is the same in every case, and it shapes how to buy.
The cause: memory went to AI data centers
AI data centers need enormous amounts of memory, above all high-bandwidth memory (HBM) stacked next to their AI chips. Memory makers earn far more per chip on HBM and server memory than on the memory in PCs and phones, so they moved factory capacity there. What remained for everyone else got scarcer and more expensive.
The numbers are steep. The market research firm TrendForce reported that contract prices for conventional memory, what PC and phone makers pay, were expected to rise 58% to 63% in the second quarter of 2026 alone, with flash storage up 70% to 75%. Memory makers also cut shipments to PC makers, pushing their prices up further. For the third quarter, TrendForce expected smaller but still sizable rises of 13% to 18%, and said retail laptop prices were expected to rise across the board.
Why AI PCs are hit hardest
A computer for running AI models locally is, above all, a lot of fast memory: the models live in it, and its speed sets how fast they write (see our bandwidth study). So the machines that make the best AI PCs are the ones with the most memory cost to pass on.
What we saw when we checked prices in September 2026:
- Graphics cards: well above their launch prices. The RTX 5060 Ti 16 GB sold for nearly double its $429 launch price, AMD’s RX 9060 XT 16 GB for 58% more, and the Radeon AI PRO R9700 for 31% more. The cheapest RTX 5090 we found was more than three times its $1,999 list price, and NVIDIA’s planned refresh with more memory was reported to be on hold over memory costs.
- Compact AI computers: NVIDIA raised the DGX Spark’s list price from $3,999 to $4,699 in February 2026.
- Memory upgrades from Apple: going from 64 to 128 GB in a Mac Studio with M5 Max costs $1,600, and from 48 to 128 GB in a 16-inch MacBook Pro $2,000.
- Mini PC sales: the makers of 128 GB Ryzen AI Max+ 395 mini PCs were selling them 7% to 20% below their own list prices.
Our deals page tracks the latest price we recorded for every machine we cover.
How to buy well anyway
- Buy the memory size you need, once. In almost every fast AI machine the memory is soldered, and upgrades at purchase cost far more than they used to. Our memory guide and calculator show what each size runs.
- Compare against real prices, not “was” prices. A discount from an inflated list price may still be above what the machine has been selling for.
- Check the maker’s own store. Brand stores ran the deepest mini PC sales we saw.
- Consider used graphics cards. A used RTX 3090, with 24 GB and fast memory, remains one of the best local AI cards per dollar; buy from a seller with a return policy.
- Use what you have first. Many PCs can already run small models; see our guide to running AI without a graphics card.
- Don’t wait for a crash. Forecasts in mid-2026 still pointed to rising, if slower-rising, memory prices. If you need a machine, the best time to buy is when you find a price below the ones we have recorded.
Questions people ask
Why is RAM so expensive in 2026?
Memory makers are putting their factory capacity into high-bandwidth memory and server memory for AI data centers, which earn them far more per chip. That leaves less for PCs and phones. TrendForce reported conventional memory contract prices rising 58% to 63% in the second quarter of 2026 alone.
Will AI PC prices come down?
Not soon, by current forecasts. TrendForce expected memory contract prices to keep rising in the third quarter of 2026, though more slowly, by 13% to 18%. Individual sales still happen, especially at brand stores, so compare prices rather than waiting for a general drop.
Should I wait to buy an AI PC?
If you need one now, buy the memory size you need at the best price you can find; waiting has mostly meant paying more in 2026. If you can wait, watch for sales against the prices we record, such as during Black Friday, and consider a used graphics card.
Why did graphics cards get more expensive than mini PCs?
Both rose. Graphics cards use GDDR memory, which is also in short supply; NVIDIA's planned refresh with more of it was reported to be on hold over its cost. In September 2026 we found the RTX 5060 Ti 16 GB at nearly double its launch price, while 128 GB mini PCs were on sale 7% to 20% below their makers' list prices, which had themselves risen.
Sources
- TrendForce: AI server demand to drive memory contract price increases in 2Q26 (March 31, 2026), checked Sep 26, 2026
- TrendForce: AI server demand continues to support memory prices in 3Q26, but gains moderate (July 3, 2026), checked Sep 26, 2026
- VideoCardz: RTX 50 SUPER launch on hold over 3 GB GDDR7 pricing, checked Sep 26, 2026
- TechSpot: graphics card prices, Q3 2026, checked Sep 26, 2026
- NVIDIA Developer Forums: DGX Spark price change announcement (February 23, 2026), checked Sep 25, 2026
What changed
- Sep 26, 2026: First published.