The Memory Crisis

If you haven’t looked at the price of computer memory recently, you might want to sit down first.

Something very strange happened to the memory market in 2026.

In February, TrendForce revised its forecast for conventional DRAM contract pricing in the first quarter from an already painful 55–60% quarter-over-quarter increase to an almost unbelievable 90–95%.

PC DRAM was even worse.

TrendForce expected PC DRAM contract prices to increase more than 100% in a single quarter.

Not over five years.

Not since the pandemic.

One quarter.

Counterpoint Research saw essentially the same thing, reporting memory prices rising roughly 80–90% between Q4 2025 and Q1 2026 across DRAM, NAND and HBM.

If you’re a PC builder who remembers when RAM and SSDs seemed to get cheaper every year, welcome to 2026.

The question is:

What happened?

The short answer is AI.

The longer answer is considerably more interesting.

Everyone Wants the Same Silicon

To understand the memory crisis, you have to understand something about semiconductor manufacturing.

Capacity is not infinite.

Memory manufacturers have fabrication plants, wafer capacity, packaging capacity, engineering resources and capital that must be allocated somewhere.

And right now, the most valuable somewhere is increasingly the data center.

AI infrastructure consumes enormous amounts of memory.

GPUs and AI accelerators require High Bandwidth Memory, or HBM. AI servers also need enormous quantities of conventional server DRAM. AI inference is increasing demand for high-capacity RDIMMs. Data centers need enterprise SSDs to feed enormous datasets into these systems.

Suddenly the world’s memory manufacturers have an interesting economic decision to make.

They can allocate resources toward lower-margin commodity products.

Or they can allocate more resources toward the products hyperscale cloud providers and AI companies desperately want.

Guess which customer gets priority.

TrendForce reported in September that growing demand from AI servers continues driving shipments of HBM3e, LPDDR5X and high-capacity RDIMMs while supplier inventories remain at historically low levels. Additional supply has been primarily allocated toward servers.

That last part is important.

This isn’t simply:

AI companies are buying lots of memory.

It’s:

AI infrastructure is changing where the memory industry wants to allocate its limited production capacity.

And consumers are beginning to feel the consequences.

Your Gaming PC Is Competing With an AI Data Center

Obviously, NVIDIA isn’t buying the exact DDR5 DIMM sitting in your Newegg shopping cart and installing it into an AI cluster.

The supply chain is more complicated than that.

Different memory products use different processes, packaging technologies and manufacturing resources.

But they ultimately exist inside the same global memory industry.

And manufacturers follow margins.

When hyperscalers are spending tens of billions of dollars building AI infrastructure, memory manufacturers have a powerful incentive to prioritize the products those customers require.

The numbers tell the story.

Global DRAM revenue reached approximately $97 billion in Q1 2026, according to Counterpoint, an increase of 80% from the previous quarter and 260% year over year. AI data centers and rapidly increasing memory prices were major drivers.

TrendForce reported similarly extraordinary conditions. Conventional DRAM contract pricing increased approximately 93–98% quarter over quarter in Q1, while suppliers entered Q2 with extremely low inventory and prioritized additional supply for high-capacity server RDIMMs.

That leaves PC manufacturers and smartphone manufacturers fighting over a tighter pool of supply.

And eventually that fight reaches you.

SSDs Aren’t Escaping Either

This isn’t just a RAM problem.

NAND flash—the technology underneath SSD storage—is experiencing its own pricing shock.

In its February forecast, TrendForce expected NAND Flash contract prices to increase 55–60% in Q1 2026, while enterprise SSD prices were projected to jump 53–58%. Strong orders from North American cloud service providers were specifically cited as a major driver.

Counterpoint’s measurements were even more dramatic in some categories.

Its March analysis found NAND Flash pricing rising more than 90% quarter over quarter during Q1.

Again, look at where the demand is coming from.

Modern AI isn’t simply a GPU problem.

Training requires data.

Inference requires data.

Retrieval-augmented generation requires data.

Databases require storage.

Checkpoints require storage.

Models require storage.

Logs require storage.

And the massive infrastructure being constructed around AI needs somewhere to put all of it.

The AI boom is therefore becoming a memory and storage boom too.

Q1 Was Insane. Q3 Is Merely Painful.

There is some good news.

Memory prices are not continuing to double every quarter.

There is also bad news.

They’re still going up.

TrendForce’s July forecast projected conventional DRAM contract prices increasing another 13–18% quarter over quarter in Q3, while NAND Flash was expected to rise another 10–15%.

After what happened earlier this year, those numbers almost look reasonable.

They aren’t.

A 15% quarterly increase would normally be significant.

It just looks small when the previous comparison involved prices nearly doubling.

Think of it this way:

If something costs $100 and increases 90%, it becomes $190.

If it then increases another 15%, you’re at $218.50.

The rate of increase fell dramatically.

Your wallet did not recover.

That distinction matters when people start talking about the memory market “cooling.”

Cooling Doesn’t Mean Getting Cheaper

This is perhaps the most important thing consumers should understand.

The memory market is showing signs that price increases can moderate.

But one major reason isn’t suddenly abundant supply.

It’s that customers are reaching the limit of what they’re willing—or able—to pay.

TrendForce explicitly says record-high contract pricing is pushing PC and smartphone customers toward their affordability limit, weakening consumer demand and moderating further increases.

That’s not the same thing as the supply problem being solved.

It means prices got high enough that people started saying no.

Notebook demand has weakened.

PC manufacturers are resisting increases.

Consumers postpone upgrades.

Manufacturers reduce specifications.

Demand destruction begins doing what new supply hasn’t done yet.

The market finds equilibrium.

Just at a much more expensive level.

The Damage Is Already Reaching Finished Products

Memory chips are components.

Consumers don’t normally walk around checking DRAM contract indexes.

They notice the problem when the laptop, smartphone or graphics card they want suddenly costs more—or ships with less memory than expected.

We’re already seeing that effect.

Counterpoint reported that rising memory costs pushed smartphone bill-of-material costs sharply higher in Q1, with manufacturers responding by simplifying product portfolios and adjusting hardware specifications. The firm concluded that higher retail pricing in 2026 was increasingly unavoidable.

In Southeast Asia, the average smartphone selling price reached a record $375 in Q1 2026, up 21% year over year, with Counterpoint identifying memory inflation as an important driver. Shipments of phones below $150 fell 29%.

This is how an infrastructure boom eventually reaches consumers.

You don’t need to own an AI server.

You don’t need a ChatGPT subscription.

You don’t even need to care about artificial intelligence.

You can still end up paying for the AI infrastructure race when you buy your next computer.

Gamers Just Got an Interesting Warning

Then there is Micron.

Reports this month found that Micron’s 2GB—or 16Gb—GDDR7 components used in some NVIDIA GeForce RTX 50-series graphics cards have reached end-of-life status.

Micron’s 3GB GDDR7 products remain.

Samsung and SK hynix can still provide 2GB GDDR7, so this doesn’t mean GeForce cards are about to disappear.

But the development is worth watching because it illustrates what is happening across the memory industry.

During a period of constrained supply, manufacturers naturally favor products and customers offering better economics.

TrendForce’s reporting confirms that Micron’s 28Gbps and 32Gbps 2GB GDDR7 products have been designated EOL, although Micron has not publicly explained the decision.

So I would be careful about claiming that Micron explicitly killed those products because of AI.

We don’t know that.

But the broader direction of the industry is difficult to miss.

Capacity is valuable.

And increasingly, AI infrastructure is where the money is.

Even the Boring Memory Is Getting Hit

Here’s where this story gets more interesting than expensive gaming PCs.

You probably know what DDR5 is.

You probably know what an SSD is.

You may even know HBM.

You probably don’t spend much time thinking about NOR Flash or SLC NAND.

And you shouldn’t have to.

These are the boring memory technologies hiding inside industrial systems, automobiles, networking equipment, medical devices, aerospace systems and countless embedded products.

They’re getting squeezed too.

TrendForce says major memory suppliers are prioritizing higher-value products including HBM and advanced 3D NAND, tightening capacity for mature products such as NOR Flash and SLC NAND.

Contract prices for both categories increased more than 100% during the first half of 2026.

And as of September, TrendForce says high-capacity NOR Flash could increase another 90–110% during the second half of the year.

This matters because shortages in boring components can cause surprisingly large problems.

A $50,000 system can be delayed because someone can’t source a relatively inexpensive chip.

We learned that lesson during the pandemic semiconductor shortage.

You don’t need to run out of processors to disrupt manufacturing.

Sometimes you just need to run out of the little component required to make everything else work.

Memory Has Become Strategic

For years, the technology industry treated memory almost like a commodity.

Processors were exciting.

GPUs were exciting.

Memory was the thing sitting beside them.

AI changed that perception.

A modern AI accelerator is only useful if you can feed it data quickly enough.

That is why HBM became so important.

You can build an extraordinarily powerful processor, but if memory bandwidth cannot keep the compute engines supplied with data, part of that expensive silicon sits waiting.

Suddenly memory isn’t just capacity.

It’s performance.

It’s throughput.

It’s system architecture.

And increasingly, it’s a strategic constraint on AI infrastructure.

That’s why companies are willing to pay for it.

Should You Buy RAM and SSDs Now?

This is where everyone wants the prediction.

Should you buy now?

Or wait?

I don’t think the answer is simply “buy everything immediately before prices get worse.”

It depends on whether you actually need it.

If you’re building a computer right now and know you need 32GB or 64GB of memory, waiting for the market to magically return to 2024 pricing is a gamble.

Current industry forecasts still show upward pressure.

If you’re buying storage for a system you need today, the same logic applies.

But if you’re thinking about buying components simply because you’re afraid prices might increase, remember that technology has another problem:

It ages.

Buying a 4TB SSD you don’t need for another two years just to avoid a potential price increase isn’t necessarily brilliant financial planning.

Buy what you need.

Maybe add reasonable headroom.

Don’t panic-buy computer components.

And don’t assume the old rule that memory will automatically be cheaper six months from now still applies.

At least not in this market.

And About 32GB of RAM…

There’s been discussion online suggesting Microsoft now recommends 32GB of memory for Windows 11.

That’s not quite accurate.

Microsoft’s official Windows 11 minimum remains only 4GB. Its current consumer PC buying guide recommends 8GB to 16GB for general use.

Gaming is different.

Microsoft currently describes 16GB as the baseline for gaming PCs and 32GB as the “no worries” upgrade, particularly when gaming alongside browsers, Discord or streaming applications.

Personally, if I were building a serious new desktop in 2026, I would strongly consider 32GB.

Not because Windows suddenly requires it.

Because software rarely becomes less memory-hungry over time.

But that’s a recommendation, not a Microsoft requirement.

When Does This End?

Eventually.

Memory is famously cyclical.

High prices encourage manufacturers to invest.

New fabs get built.

Processes improve.

Production increases.

Demand changes.

Inventories rebuild.

Prices eventually normalize.

But semiconductor capacity doesn’t appear because somebody changed a setting in a factory.

Fabrication plants take years and billions of dollars to build.

Advanced memory manufacturing is technologically difficult.

And AI infrastructure demand isn’t showing signs of disappearing.

TrendForce’s current data still shows historically low supplier inventories and additional DRAM supply being prioritized toward servers.

So I wouldn’t expect a magical return to cheap memory simply because quarterly price increases begin slowing.

The first stage of recovery may not be falling prices.

It may simply be prices stopping their climb.

That’s an important difference.

This Is the Data Center Boom Showing Up at Best Buy

I’ve written before about the enormous infrastructure buildout happening underneath artificial intelligence.

Data centers.

Power.

Cooling.

Networking.

GPUs.

Storage.

For most consumers, those numbers are abstract.

A company announces another $20 billion data-center campus and it sounds like something happening far away in an industrial park.

But infrastructure markets don’t stay isolated forever.

Hyperscalers buy GPUs.

GPUs consume HBM.

AI servers consume DRAM.

AI clusters consume storage.

Memory manufacturers allocate production toward those markets.

Other products compete for what’s left.

PC manufacturers pay more.

Phone manufacturers pay more.

Hardware vendors adjust configurations.

Eventually you walk into a store and wonder why the computer you planned to buy suddenly costs several hundred dollars more.

That’s the same AI infrastructure boom.

You’re just seeing it from the other end of the supply chain.

AI Has a Physical Economy

One of the biggest misconceptions about artificial intelligence is that it exists somewhere inside “the cloud.”

It doesn’t.

AI is incredibly physical.

It requires silicon.

It requires memory.

It requires storage.

It requires fiber.

It requires transformers.

It requires cooling systems.

It requires enormous amounts of electricity.

It requires buildings large enough to contain all of it.

And when demand for one technology becomes large enough, it starts competing with everything else for those physical resources.

That’s what the memory crisis of 2026 really is.

It isn’t merely a strange year for RAM prices.

It’s one of the first consumer-visible examples of what happens when the AI infrastructure boom begins reorganizing the technology supply chain around itself.

For years we’ve talked about what artificial intelligence might do to jobs, software and the economy.

Now we’re discovering something much simpler.

AI needs hardware.

A lot of it.

And somebody has to compete with the data centers for what’s left.

AI didn’t just change how we use computers.

It made building one more expensive.