Why RAM Prices Are Rising in 2026

NVIDIA GPUs were hard to get. Then they got expensive. Data centers kept buying more of them, and anyone building a gaming PC or local-AI workstation felt the pain.

Now the problem has moved.

It is memory.

On September 23, TrendForce said DRAM supply is still tight because suppliers are favoring HBM and server memory for cloud companies. A week earlier, Intel CEO Lip-Bu Tan said memory prices had already gone up five to seven times in parts of the market. IDC says DRAM and NAND costs for phone makers are up more than 300% year over year.

The weird part is where those numbers end up.

Not only in an AI server.

They end up in the price of your next laptop, phone, SSD, mini PC, console, and probably the RAM kit you were planning to buy next month.

Accces without medium partner: AI Data Centers Ate the GPUs

I spent some time checking whether this was just another “chip shortage” headline. It isn’t that simple. The shortage is real, but there is also disagreement about how long it lasts. Some memory companies talk about pressure extending toward 2030. Acer CEO Jason Chen says that is overdone and expects prices to start coming down in 2027.

So yes, the story is messy.

But right now the consumer is paying for it.

The number that stopped me was not 300%. It was 65%.

IDC published a brutal number in August.

At the low end of the smartphone market, memory costs now make up more than 65% of the bill of materials for some devices. IDC says those costs are up nearly 300% from a year earlier.

That is absurd when you think about what a cheap phone is supposed to be.

A budget phone still needs a display, battery, camera modules, modem, processor, casing, motherboard, charging hardware, speakers and all the tiny bits nobody talks about.

And yet RAM plus storage can now eat most of the hardware budget.

Fairphone CEO Raymond van Eck described the same problem from the smaller-manufacturer side. He said the AI boom pushed memory suppliers toward high-bandwidth server DRAM, leaving less normal memory for everyday devices. Fairphone still launched its Gen. 6+ with 12GB of DDR5, but his post reads less like a launch announcement and more like somebody describing a supply-chain headache.

Tom’s Hardware reported that memory can now account for up to 60% of the material cost of a roughly $400 device for some smaller manufacturers.

That is where this stops being a data-center story.

A company building a $2,000 laptop has room to absorb a bad component quarter. A company building a $180 phone does not. The cheap end breaks first.

AI did not literally take your laptop RAM

This part needs some care because the simple version sounds better than the real one.

NVIDIA is not buying the same DDR5 SODIMM you put into a laptop.

AI accelerators use HBM, or high-bandwidth memory. Phones use LPDDR. PCs use DDR5. SSDs use NAND. They are different products.

But they compete for factories, cleanroom space, packaging capacity, engineering attention and money.

That is the link.

HBM also uses a lot of silicon area and more complicated packaging than a normal DRAM product. TrendForce says rising HBM demand since the second half of 2025 has crowded out conventional DRAM capacity. Its September 23 market bulletin says suppliers are still favoring HBM and server DRAM, while cloud service providers keep buying for AI.

Micron has been prioritizing higher-priced server DRAM in a capacity-constrained market. SK hynix reported record results in July and said sales of HBM, AI-server DRAM and enterprise SSDs were major drivers.

From the memory maker’s point of view, this is not hard to understand.

Why use scarce production capacity for a low-margin consumer part if Microsoft, Meta, Google, Amazon or an AI server builder will pay much more for memory attached to expensive compute?

The consumer RAM stick loses that argument.

This is why cheap electronics are getting weird

The first thing manufacturers do when one component becomes too expensive is not always raise the sticker price.

Sometimes they quietly change the product.

A laptop that would have shipped with 32GB gets 16GB.

A cheap phone keeps 8GB instead of moving to 12GB.

A company removes a storage tier.

A product launch gets delayed.

A smaller brand starts qualifying memory from a supplier it would not have considered two years ago.

That last part is already happening.

TrendForce says mobile and PC makers are testing alternative memory sources as the big suppliers keep more capacity around server products. Tom’s Hardware reported that some smaller device makers are doing more checks for counterfeit or refurbished memory because supply is so strange.

I hate this part of a shortage because it is hard to see on a spec sheet.

A $999 laptop can still be called the same model even if the base configuration quietly becomes worse value. The sticker price tells only half the story.

Framework’s response tells you how strange this market has become

Framework is almost built for this kind of mess because its laptops use replaceable memory.

When RAM prices jumped, Framework had to react to cost changes like everyone else. But a modular machine gives the buyer another option: buy the laptop with less memory and install your own later, or reuse modules you already have.

That sounds boring.

In 2026, boring is useful.

Most thin laptops moved in the opposite direction. LPDDR is soldered. Apple’s unified memory is on-package. Many mini PCs are now using fixed LPDDR5X because wide memory buses are needed for their integrated GPUs.

Those designs are fast, but they remove the escape hatch.

If memory prices go mad, you cannot wait six months and upgrade when prices calm down.

You bought the memory decision on day one.

This is one reason I think upgradeable RAM is going to become interesting again, at least for normal PCs. Not because SODIMMs are sexy. They are not. But when 32GB can swing by hundreds of dollars across a shortage cycle, a slot starts looking pretty nice.

The smartphone market is already showing the damage

IDC now expects worldwide smartphone shipments to fall 16.7% in 2026, to just over one billion units.

That would be the steepest annual drop it has recorded.

At the same time, IDC expects the average selling price of a smartphone to rise 27.6% to $581.

Fewer phones.

Higher prices.

That is a nasty combination.

Usually a weak market forces manufacturers to discount products. This time their component costs are moving the wrong way at the same time.

IDC says the sub-$100 smartphone category is in the worst position. Around 173 million phones under $100 shipped last year, and the firm now describes that segment as facing an “existential crisis.”

That wording sounds dramatic until you look at the math.

If memory takes 65% of a cheap phone’s bill of materials, there is not much left to cut.

You can downgrade the display. You can reuse an old camera sensor. You can make the casing cheaper. At some point the product becomes bad.

Or the price goes up and it is no longer a $99 phone.

PC buyers are next, and some of them already know it

The PC side is less clean because buyers have more choices.

You can build a desktop.

You can buy used RAM.

You can keep an older SSD.

You can choose a laptop with replaceable memory.

But OEMs are still exposed to the same component market.

TrendForce said on September 16 that AI infrastructure demand is tightening DRAM supply while notebook makers cut output and raise prices. Its newer September 23 update says the market remains undersupplied and that suppliers continue to favor HBM and server DRAM.

That is fresh data, not a forecast from six months ago.

And the pain is not limited to DRAM.

AI servers need huge amounts of NAND for storage as well. Enterprise SSD demand has become another part of the data-center buildout. A local AI cluster needs storage for models. A cloud training system needs much more.

So the pressure can hit both sides of a PC purchase: memory and SSD.

This is where I had to correct my own first thought. I assumed the story was mostly about HBM stealing DRAM capacity. It is bigger than that. AI infrastructure is pulling on the whole memory stack.

Different products, same factories and investment decisions.

Intel’s CEO says memory is now the problem customers call him about

Lip-Bu Tan has been talking about memory pressure for months.

In September he said customers were telling Intel that memory is one of the biggest blockers in AI infrastructure. According to reports from his Splunk .conf26 appearance, some memory prices are already five to seven times higher than earlier levels.

He also said shortages are delaying projects.

That sounds like a data-center problem, and mostly it is.

But every large AI customer that signs a long-term memory deal reduces how much flexible supply is left for somebody else.

TrendForce says long-term agreements are shrinking the amount of DRAM that can be negotiated on the open market.

The big buyer does not wait for Black Friday.

It signs a contract.

The small laptop maker gets what is left.

Memory companies are having a very good crisis

There is another side to this.

For memory manufacturers, 2026 has been an excellent business.

TrendForce says total DRAM industry revenue jumped 59.5% quarter over quarter in Q2 2026, reaching nearly $154.73 billion.

Samsung led the market with 39.4% share in that quarter. SK hynix was second, and Micron continued moving its product mix toward higher-priced server DRAM.

SK hynix reported record quarterly performance in July, driven by AI memory and higher-value products.

None of this means memory companies created a fake shortage.

Demand is real. HBM needs a lot of capacity. Server DRAM demand is high. Building new fabs takes years.

But the incentives are obvious. Expensive server memory is better business than cheap consumer memory.

I do not blame a company for selling the product that makes more money. I also don’t want to pretend consumers are somehow separate from that choice.

They are not.

We are downstream from it.

Acer’s CEO thinks the 2030 shortage talk is too much

This is where the story gets more interesting.

Not everyone agrees the pain lasts for years and years.

Acer CEO Jason Chen said this week that memory suppliers may be overstating fears of a shortage lasting to 2030. He expects more Chinese capacity to come online and thinks PC component pricing can start easing around late 2027.

That is a very different view from the most pessimistic supplier talk.

He may be right.

CXMT and other Chinese memory makers are expanding. More capacity is being built. High prices attract investment because, well, of course they do.

Shortages contain the seeds of their own end.

The annoying part is timing.

A fab announced today does not put cheap RAM in your laptop next Tuesday.

Even Chen’s more optimistic view still leaves buyers dealing with high prices through much of 2027.

So I would not plan a PC purchase around some magical collapse in RAM prices three months from now.

Could it happen? Sure.

I would not bet on it.

This is starting to change how PCs should be reviewed

For years, laptop reviews treated RAM almost like a checkbox.

16GB.

32GB.

Soldered or upgradeable.

Done.

That is not enough anymore.

I want to know how much the 32GB upgrade costs. I want to know whether the SSD is replaceable. I want to know if a 64GB configuration exists, and whether the manufacturer doubled the price to get there. I want to know whether the exact same laptop shipped with a better base memory configuration last year.

This matters even more for local AI.

We have spent 2026 talking about 128GB, 160GB and 192GB mini PCs because model size is becoming a buying decision.

Now memory itself is becoming scarce at the same time.

Kind of perfect timing, right?

The people who suddenly want more RAM are arriving just as the industry decided RAM should cost more.

Local AI makes the problem worse in a funny way

Local AI is supposed to be the escape from cloud AI costs.

Buy hardware once.

Run models locally.

No token bill.

Keep data on your machine.

I still like that idea.

But a serious local-AI machine wants exactly the component that is getting squeezed.

Memory.

A 32GB machine is easy.

A 64GB machine is better.

Then you look at 128GB and 192GB unified-memory systems because you want to run 70B, 120B or larger quantized models locally.

That is why AMD’s Ryzen AI Max+ machines became so interesting this year.

It is also why Apple’s 512GB M5 Ultra Mac Studio exists.

Capacity is now a product feature.

And the market is charging accordingly.

The local-AI crowd is basically walking into a restaurant during a food shortage and asking for the biggest plate.

Bad timing.

I would not panic-buy RAM

This is the part where a lot of tech content goes off the rails.

“Buy now before prices double.”

No.

If you need a machine for work, buy what you need.

If your current PC is fine, I would not hoard five DDR5 kits because somebody on Reddit posted a chart.

Consumer retail pricing does not move in a straight line. Different regions have old inventory. Promotions happen. New supply appears. OEM contracts are not the same as retail kit prices.

And this shortage has a real counter-force: high prices are making customers buy less.

TrendForce already says PC and smartphone buyers have limited ability to absorb more increases.

There is a ceiling.

We just do not know exactly where it is.

The part that worries me is not a $400 RAM kit

Enthusiasts will complain about expensive DDR5.

They will survive.

The uglier effect is at the bottom of the market.

A developer buying a 192GB mini PC can decide to wait.

A student buying a first laptop may not have that choice.

A family replacing a broken phone may get pushed from a $120 device to a $180 device.

A school ordering 2,000 basic laptops feels every extra $30.

That is why the IDC number matters more than an expensive desktop kit.

When memory becomes 65% of a low-end device’s hardware cost, cheap technology stops being cheap.

And once companies discover customers will tolerate the higher price, some of that increase may stick even after supply improves.

We have seen that movie before.

AI did not stop at the GPU

The AI boom was easy to see when the scarce thing had NVIDIA written on it.

Now the cost is spreading into boring parts.

DRAM.

NAND.

Packaging.

Server memory contracts.

Factory allocation.

Most people will never read a TrendForce DRAM bulletin. They will just notice the next laptop costs more than the old one, or the base model has less memory than expected.

That is the part worth watching.

AI data centers are not only creating a market for expensive accelerators. They are changing what memory manufacturers choose to build, who gets first access to it, and what the rest of us pay.

Maybe Acer is right and the worst of this starts easing in 2027.

I hope so.

But on September 24, 2026, the latest data still points the other way.

The GPU shortage was obvious because the shelf was empty.

The memory shortage is sneakier.

The shelf can stay full.

The device just gets more expensive.