Apple’s next iPhones are likely to arrive with a higher price tag. That would confirm what companies making memory parts have been pointing to: RAM costs are rising fast, and there is no quick fix in sight.
RAMageddon, Explained
People have started calling it “RAMageddon.” Company transcripts have been full of talk about memory prices and a memory dearth — language about memory prices and tight supply appeared in 473 company transcripts last quarter, according to data from AlphaSense. For decades, memory got cheaper as it got better, letting phones and computers gain power without costing much more. That trend is now running in reverse.
Many assume AI alone is to blame. But the causes of the dearth go back further than that.
Before ChatGPT
The trouble started before ChatGPT existed. In 2021, even while demand for electronics during the pandemic was strong, Micron came to see that improving memory technology alone would not be enough to keep up with lasting demand. Makers of memory would need to process more material and build new facilities to do it.
Then the business fell apart. Buying of phones, tablets and computers during the pandemic had moved demand earlier than it would have come otherwise. When people spent less afterward, makers were left holding too much stock, lost money, and cut back their plans to grow.
By the time the business began to recover, generative AI had created a wave of demand for memory far bigger than anyone had planned for.
Three Companies, Most of the Business
The memory business is held by only a few companies. Three of them control about 90% of it, according to Counterpoint. Samsung led with 39% of the business in the second quarter of 2026, followed by SK Hynix at 26% and Micron at 25%.
That means AI data centers and ordinary devices are competing for supply controlled by just a few makers.
Manish Bhatia, president and COO of Micron, told The Verge the industry needs to build far more capacity to produce memory material. “[It’s] a very different challenge for the industry than it had been for many years before, where technology alone was able to keep up with the demand,” Bhatia said.
Two Kinds of Memory
Memory in a phone or PC generally falls into two kinds. One kind holds the information a device needs while it works, such as an app opening or a page coming up on screen. The other kind keeps information for a longer time, such as pictures and files stored on a device.
Inside AI data centers, a special kind of memory known as HBM does the heavy lifting. HBM is built by joining several memory parts together with advanced connections, letting it move large amounts of information quickly to power AI processors.
David Naranjo, associate director at Counterpoint, told The Verge that the memory used in data centers is not the same as ordinary memory, and that the amount of one being used does not simply explain the amount of the other being used.
HBM is harder to make, but it sells for more. Deep-pocketed AI companies like Nvidia and AMD, along with Meta and Microsoft, want as much of it as they can get and are willing to pay for it.
It also uses far more of a maker’s capacity than the ordinary kind of memory. Because HBM joins larger parts together, it needs much more material for each amount produced. Micron says making a given amount of HBM takes roughly three times as much material as making the same amount of the ordinary kind.
Why Makers Favor AI Buyers
The money points one way. Rather than guessing how many phones or laptops will sell next year, memory makers can sign deals lasting many years with some of the richest companies in the world.
“We have been engaging in discussions with customers, prioritizing those who can guarantee committed future captive demand,” said Jaejune Kim, Samsung’s executive vice president of memory, on the company’s last earnings call.
AI is also pushing up demand for the ordinary kind of memory. Phone and PC makers want to run small AI models directly on devices, and that calls for more advanced memory — and more of it.
The three big memory makers are largely giving out the capacity they have to these companies.
Song Hyun-jong, president of SK Hynix, said on the company’s most recent earnings call that as AI reaches more services, such as search, coding and productivity tools, demand for memory is coming from more directions than before. He said both AI memory and ordinary memory are now growing together, marking a lasting change in how demand is shaped.
Samsung and Micron both say they remain committed to producing the ordinary kind of memory alongside HBM.
What It Means for Buyers
No one is promising a quick end to this.
Below is a rough picture of how the dearth has built over time, based on events described by Micron, SK Hynix and Samsung executives.
| Period | Development |
|---|---|
| 2021 | Micron concludes technology alone can’t keep pace with demand; making enough material becomes the limit |
| During the pandemic | Demand for electronics was strong, then dropped off, leaving makers with too much stock |
| After ChatGPT’s rise | Demand for HBM tied to AI grows fast, straining supply that was already tight |
| Q2 2026 | Samsung holds 39% of the memory business, SK Hynix 26%, Micron 25%, per Counterpoint |
| Last quarter | Language about memory prices and tight supply appeared in 473 company transcripts, per AlphaSense |
For now, the message from the industry’s biggest companies is steady: supply is limited, AI demand keeps growing, and ordinary devices like phones and PCs are competing for the same limited material. That competition is what’s about to show up on iPhone price tags.
Source: theverge.com
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.