AI's Hunger For Memory Chips Could Shrink Smartphone and PC Sales in 2026, IDC Says

The global smartphone and PC markets face potential contractions of up to 5.2% and 8.9% respectively in 2026, according to downside risk scenarios from IDC that trace the problem to memory chip manufacturers shifting production capacity away from consumer electronics toward AI data centers. Samsung Electronics, SK Hynix and Micron Technology have pivoted their limited cleanroom space toward high-bandwidth memory for AI servers, restricting supply of the conventional DRAM and NAND used in phones and laptops.

IDC expects 2026 DRAM supply growth to hit 16% year-on-year, below historical norms. The smartphone industry's decade-long trend of bringing flagship features to affordable devices is reversing. Memory represents 15-20% of the bill of materials for mid-range phones, and thin-margin vendors like Xiaomi, Realme and Transsion will bear the brunt. Apple and Samsung have long-term supply agreements securing components up to 24 months ahead. PC vendors including Lenovo, Dell, HP, Acer and ASUS have warned clients of 15-20% price increases heading into the second half of 2026.

Will somebody with connections please

[Tablizer]

...expose and pop this damned AI bubble!? Jeez

Re: Will somebody with connections please

[Mr. Dollar Ton]

No way, we're all in.

Re:Will somebody with connections please

[gtall]

I think we need to wait until either (1) Wall Street gets spooked about companies' return on their AI investment, and/or (2) major investment funds lose a big chunk of money to AI. According to an op-ed in TheRegister, https://www.theregister.com/20... [theregister.com], (and I hope I'm paraphrasing this correctly), most companies promoting their AI investment are merely buying some AI and then telling their proles to use it figuring they can then cut their workforce. However, that only makes the proles more productive, it does not eliminate roles. To do that, you must redo your internal processes to fit AI. Very few companies do that because that would be expensive. They want AI on the cheap, and it sounds good on quarterlies and to their boards.

To change the processes, it helps to start at the bottom and let it filter up, not establish some company-wide mandate to use it (or else).

That article also mentions something which I think is a killer. AI doesn't learn on its own. Let's presume this is true. To get around it, you must be constantly updating the model. To do that, you must constantly extract new information from your organization. But if your organization does not change, then your AI will not significantly change. And this is a recurring cost, not buy once and flog it forever to your investors, board, customers, etc.

Re: Will somebody with connections please

[rayzat]

I personally struggle to see the value. I use some AI tools and it maybe saves me 2 hours a week, but, some many other people I work with are seeing 20-30% time savings. AI might be overinflated and needs to deflate a bit but I'm slowly moving to the this is the new normal stance.

I regards to AI impacting PC and phone. That's almost a given at this point. I would not be shocked if many second tier, in terms of volume not quality, laptop, phone, and tablet manufactures go bankrupt this year. I have one of

Re:

[Tablizer]

It's not disputed that AI has some use, what's disputed is if all this AI infrastructure being built would be profitable without the market-share-grabbing subsidies in place now. Can it carry itself.

The early web likewise showed promise, but the first batch of dot-coms couldn't run a profitable business. It took market trial and error to solve this. And the economy got a bubble poppage while market Darwinian forces battled it out.

Re:

[unixisc]

Fully agree! Also interesting that all the activists who used to protest the building of power plants for people seem to have gone quiet when it comes to unlimited power for AI datacenters

Re:

[mysidia]

...expose and pop this damned AI bubble!? Jeez

The major problem here is Microsoft and Google, and the fact that we have given these two companies so much money by using their shit too much, so they have trillions in spare cash to spend on AI shit.

The AI bubble is Not a bubble as far as the chip manufacturers are concerned, so long as Microsoft and Google keep spending insane amounts of money on datacenters - these companies have an opportunity to make massive bank.

The chip manufacturers simply have dollar

Re:

[Tablizer]

> [tech oligopolies] have trillions in spare cash to spend on AI shit.

True, but buying market-share usually eventually peters out, based on history.

> The chip manufacturers simply have dollar signs in their eyes, and they are obligated to do what's best to their investors. Which is screw consumers, so they can sell all their capacity to Google.

But if the AI bubble pops, they'll have to go back to consumer sales to survive. AI server farm construction would grind to a halt, meaning no more big chip pur

Re:

[tlhIngan]

The chip manufacturers simply have dollar signs in their eyes, and they are obligated to do what's best to their investors. Which is screw consumers, so they can sell all their capacity to Google.

Except the memory makers *aren't* increasing production. They were burned several times by memory shortages, and when they increase production, the shortage ends and now they have a surplus they have to sell cheap.

Seeing another bubble, they're not scaling the changes to meet the demand - SK Hynix is to bring in a

Re:

[mysidia]

Except the memory makers *aren't* increasing production.

They are actually CUTTING production to create an artificial shortage in the consumer chips.

Micron who owns Crucial is literally shutting down consumer RAM chip production in order to repurpose facilities to make HBM memory for AI companies, since they didn't have the facilities, and the consumer RAM chips are not what the AI companies are interested in buying, either.

Re:

[Tony Isaac]

Where's Dr. Pimple Popper when you need her?
https://www.youtube.com/channe... [youtube.com]

AI reaching natural limits

[gurps_npc]

People think AI is a super-power, but most of it's gains are simply caused by people using better hardware.

The increasing cost of Memory may be the first of many limits that reduces AI's growth. This may cause some disillusion in the fantasy that AI will become more than a College level intern.

The cycle repeats

[computer_tot]

This tightening of computing software was one of the aspects of the rise of smartphones I enjoyed 17 years ago. Desktop software and websites had been growing at obscene rates for years since hardware capabilities tended to double every few years. Then smartphones came on the scene with their single-core ARM CPUs and limited memory and suddenly developers had to scale back, make things more efficient, and create websites that would load over a mobile 3G network connection. It was nice seeing massive Flash sites get replaced with HTML and huge applications get scaled back to work on phones.

Of course, eventually, smartphones got more and more powered and the pendulum began to swing back toward bloat. Limits on RAM sales might help us swing toward efficiency again, at least for a few years.

I was looking at smartphones and thinking

[FudRucker]

Oh boy! New hardware with the same old google owned and controlled androidOS on it, I already have a good but older phone that works fine as a hotspot, so instead of a new phone why not get a blackberry form factor Linux handheld computer (cyberdeck), like that clockwork uConsole, it has a durable metal chassis and the parts inside are easily replaced or upgraded or modified, there are a few others like the Mecha Comet too, so why buy a new smartphone when a cyberdeck would be more suitable for what I want,

Re:

[unixisc]

I currently have a Pixel 3a. I'm thinking that for the next phone, I'd get something like a Pixel 7, which I think was the last one that was not AI powered. I'll avoid the AI featured smartphones like the plague. The storage is good, except that Google's 15GB that one gets w/ every account is what the user effectively has in terms of user data, and anything else is taken by the OS and the apps. That incidentally, is what drove me off Apple, where 5GB was too small given the amount of WhatsApp photos and

We have a solution

[xack]

Luckily Linux works on "end of support" computers and "low ram" ones so you can keep using existing computers until the AI fad blows over.

Re:

[unixisc]

I doubt that will be the case once Linux completely drops 32-bit support, which is what all those retro-computers would need

Re:

[Targon]

64 bit processors have been around in x86 land since the Athlon 64 first came out in 2003, so and there will continue to be 32 bit distributions of Linux for at least another ten years.

Re:

[unixisc]

Except that 3 months ago, they already announced plans to phase out 32-bit support for Linux. They're doing that, and FreeBSD has dropped 32-bit support starting version 15. In fact, the fact that 64-bit versions of the OS have been around since the mid 00s has given them even more reason to end 32-bit support

Re:

[ukoda]

I have been migrating stuff away from PCs to Raspberry Pi as they are so much simpler to set up and easier to keep running. For demanding roles I have been working towards using the RPi5 variant with the maximum RAM but suspect they will fall victim to the price of RAM too.

Yet another way

[0xG]

That the populace is being subordinated by AI companies.

Was going build a 256GB rig for CFD/FEA

[BrendaEM]

I regret not making my purchase last fall. I live on a fixed income. I've had found myself at the edge of 2 problems:
1.) I need more memory for my OpenFoam Computational fluid dynamics tests on a vehicle body. I've done 12 generations of changes on a vehicle body, but at 64GB, I can only resolve to 10mm, a bit less than 1/2", which uses 50GB of memory--even when using fewer cores.
2.) I cannot afford a laptop workstation/gaming-machine for reverse engineering and coffee shop CAD and a desktop for heavy-lif