To The Moon

Memory stocks are surging while the broader market struggles.

The obvious explanation is simple:

AI demand is strong.

HBM is scarce.

Memory prices are rising.

Micron, SK Hynix and Samsung benefit.

I think that explanation stops one level too early.

Because something much larger is happening underneath the memory rally.

AI data centres are not simply creating another source of semiconductor demand.

They are competing with the rest of the technology industry for a limited manufacturing resource.

And increasingly, the rest of technology is losing.

The result could change much more than memory-company earnings.

It could change how much laptops cost.

How much RAM goes into smartphones.

Which consumer brands survive.

How quickly people replace devices.

Whether affordable phones can run AI locally.

And ultimately, where artificial intelligence itself gets processed.

The memory boom may therefore be less about another semiconductor cycle and more about a massive transfer of pricing power across the technology economy.

And the bill may eventually arrive in your pocket.

🧠 THE MARKET IS WATCHING MEMORY PRICES. I AM WATCHING WHO LOSES THE CAPACITY.

The usual memory thesis focuses on demand.

AI servers require enormous amounts of DRAM.

Accelerators require HBM.

Inference requires more memory as models and context windows grow.

Agentic AI potentially increases memory requirements again.

All true.

But memory manufacturing has an uncomfortable physical constraint.

You cannot simply press a button and produce unlimited additional bits.

Factories take years and billions of dollars to construct.

Advanced memory requires sophisticated manufacturing.

HBM adds complex stacking, through-silicon vias and advanced packaging.

And crucially, manufacturing one kind of memory can consume resources that might otherwise have served another market.

That is where things get interesting.

Micron has publicly explained that HBM requires roughly three times the wafer area of DDR5 for equivalent capacity, with that silicon penalty potentially becoming worse as HBM generations advance.

Think about what that means economically.

If a manufacturer shifts wafer capacity toward HBM because AI customers will pay dramatically higher prices, it does not merely produce a more profitable product.

It also removes a disproportionate amount of potential conventional memory supply from somewhere else.

That somewhere else could be:

A laptop.

A smartphone.

A gaming PC.

A console.

An automobile.

A consumer SSD.

A low-cost electronic device.

So when HBM demand grows, the effect can travel far beyond the data centre.

AI does not have to directly consume your laptop’s RAM to make your laptop’s RAM more expensive.

It only needs to make another use of the same manufacturing capacity more profitable.

💰 THIS IS CAPITALISM DOING EXACTLY WHAT CAPITALISM DOES

Imagine you own a memory factory.

Customer A builds inexpensive smartphones.

Customer B is building a multibillion-dollar AI cluster and desperately needs high-performance memory.

Customer B is willing to sign long contracts.

Customer B wants enormous volumes.

Customer B has extraordinarily deep pockets.

Customer B cares more about securing supply than saving a few dollars.

And Customer B is buying a product that generates much better economics for you.

Who gets priority?

Exactly.

This is not some mysterious market failure.

It is rational capital allocation.

Memory manufacturers are moving production toward the products and customers offering the highest return.

TrendForce says suppliers are currently prioritising server applications, while consumer DRAM supply has been significantly curtailed. Supplier inventories are at historically low levels and additional supply is being directed primarily toward servers.

Micron has gone even further.

The company disclosed around US$22 billion in commitments from 16 strategic customers designed to secure future memory supply. Management has also indicated that tight conditions could persist beyond 2027.

That is not what ordinary commodity procurement looks like.

Companies are increasingly treating memory as a strategic resource.

And whenever a resource becomes strategic, the bargaining power moves toward whoever controls it.

📱 HERE IS WHERE THE STORY BECOMES MUCH BIGGER THAN MICRON

Look at smartphones.

Counterpoint estimates global smartphone shipments fell 11% year over year in Q2 2026, reaching the weakest second quarter since 2013, with the memory shortage identified as the dominant drag on the market.

But shipment numbers only show the surface.

Underneath, the economics of building a smartphone are changing.

Counterpoint estimates smartphone memory prices rose more than 80% quarter over quarter in Q2 2026.

DRAM has now overtaken the system-on-chip to become the most expensive individual component in a smartphone.

For low-end devices, component costs were estimated to have risen around 70% year over year, with almost the entire increase attributable to memory.

That is extraordinary.

For years, smartphone companies were trained by Moore’s Law and semiconductor scaling to expect more capability for roughly similar money.

More storage.

More RAM.

Better cameras.

Better processors.

Better displays.

Every generation could offer a bit more.

The memory shortage turns that model backwards.

Now manufacturers increasingly face three choices:

Raise the price.

Accept lower margins.

Reduce the specification.

And at the cheap end of the market, there is a fourth option:

Stop making the product.

⚠️ THE LOW-END SMARTPHONE MAY BECOME COLLATERAL DAMAGE

This is one of the most underappreciated consequences of the AI infrastructure boom.

A US$1,500 flagship smartphone has room to absorb higher memory costs.

A US$100 smartphone does not.

An extra $20 or $30 of component cost is irritating on a premium device.

On an entry-level phone, it can destroy the economics entirely.

Counterpoint expects the memory crisis to hit lower-priced smartphone makers particularly hard. Its research suggests supply of older LPDDR4 memory could fall more than 40% during 2026, with the sub-US$150 segment facing severe structural pressure. Apple and Samsung are relatively insulated because their premium-heavy portfolios provide greater pricing power.

That creates another economic transfer.

The AI boom does not simply move profits toward memory manufacturers.

It could also move consumer-electronics market share toward companies whose customers can afford higher prices.

Premiumisation therefore becomes partly defensive.

Not necessarily because every consumer suddenly wants a flagship device.

But because cheap hardware becomes increasingly difficult to manufacture profitably.

That matters enormously in emerging markets where inexpensive smartphones are the primary gateway to the internet.

The AI infrastructure boom could therefore have an unintuitive side effect:

The technology designed to make computing more powerful may simultaneously make basic computing less affordable.

💻 NOW LOOK AT LAPTOPS. THE NUMBERS GET EVEN MORE ABSURD.

TrendForce recently modelled the changing cost structure of a mainstream notebook.

In early 2025, the CPU, DRAM and SSD together represented roughly 45% of the machine’s bill of materials.

By the third quarter of 2026, that share had increased to approximately 68%.

TrendForce estimates that a comparable notebook would need a retail price increase of roughly 80% from its early-2025 baseline to preserve the same gross-margin structure.

Read that again.

Not 8%.

80%.

That does not mean every laptop will suddenly double in price.

Manufacturers can sacrifice margin.

Retailers can sacrifice margin.

Configurations can change.

Companies can use lower-cost inventory.

Consumers can trade down.

Brands can reduce RAM or storage.

But someone has to absorb the difference.

And eventually, the supply chain runs out of people willing to absorb it.

That is when prices rise.

Or specifications fall.

Or demand disappears.

TrendForce now expects global notebook shipments to decline around 9.4% in 2026, even after improving its forecast as CPU availability recovered. Memory and SSD costs remain a major burden.

So while investors celebrate stronger memory pricing, downstream hardware companies are staring at something very different.

Their supplier’s pricing power is their margin problem.

🔄 THIS IS THE PROFIT TRANSFER I THINK THE MARKET IS UNDERESTIMATING

Consider the technology supply chain as one giant profit pool.

Money enters from consumers and businesses.

Then everyone fights over who captures it.

Chip designers.

Memory manufacturers.

Foundries.

Device manufacturers.

Cloud providers.

Software companies.

Retailers.

Telecom companies.

For years, memory often behaved like a commodity.

When supply became excessive, prices collapsed and buyers gained power.

When supply tightened, prices recovered.

The cycle eventually corrected itself.

But AI is temporarily changing who sits at the negotiating table.

The biggest companies on Earth are now competing for memory.

Microsoft.

Google.

Meta.

Amazon.

NVIDIA and its ecosystem.

Huge AI infrastructure operators.

These companies do not behave like a budget smartphone manufacturer trying to shave $3 from a bill of materials.

If securing memory means an AI cluster comes online six months earlier, paying more may be economically rational.

That gives suppliers extraordinary leverage.

Meanwhile the consumer-electronics manufacturer faces exactly the opposite problem.

Consumers are price-sensitive.

Upgrade cycles are already long.

Demand for PCs and smartphones is mature.

A smartphone brand cannot casually increase the price of a US$200 phone to US$400 and expect demand to remain unchanged.

So the same memory chip can exist inside two completely different economic realities.

For the AI buyer:

Memory enables revenue growth.

For the consumer-device company:

Memory is a rising input cost.

Guess which customer suppliers prefer?

🤖 AND THEN I FOUND THE PART THAT MAKES THIS THESIS REALLY INTERESTING

The memory shortage could actually influence where AI runs.

For years, the technology industry has talked about moving more AI from data centres onto devices.

Edge AI.

AI PCs.

AI smartphones.

Local inference.

Private on-device assistants.

Models that run without sending every request into the cloud.

That requires capable processors.

But it also requires memory.

Counterpoint says the current memory crisis is already encouraging smartphone manufacturers to use older processors, downgrade specifications and delay some technology transitions in mainstream devices. It specifically expects the shortage to slow the availability of on-device generative AI in cheaper smartphones.

Now think through the feedback loop.

Cloud AI creates huge memory demand.

Memory suppliers prioritise cloud and server products.

Consumer memory becomes scarcer and more expensive.

Phone and PC manufacturers reduce specifications or raise prices.

Affordable devices become less capable of running sophisticated AI locally.

More AI processing therefore remains in the cloud.

Cloud infrastructure requires more servers.

Those servers require more high-value memory.

And the loop begins again.

That’s fascinating.

☁️ CLOUD AI COULD BE STARVING EDGE AI

I don’t think this is guaranteed to happen indefinitely.

Technology will improve.

New factories will open.

Compression techniques will improve.

Models will become more efficient.

Device makers will optimise.

But for the next several years, there is at least the possibility of a self-reinforcing cycle:

Cloud AI demand → memory scarcity → expensive consumer hardware → slower edge-AI adoption → greater cloud dependence → more cloud AI demand.

That could have enormous strategic consequences.

Because where AI runs determines who controls the economics.

If more inference happens locally:

Apple, Samsung, Qualcomm, MediaTek and device ecosystems may capture more value.

If more inference remains in hyperscale data centres:

Cloud platforms and data-centre infrastructure suppliers retain more of the economic power.

Memory allocation therefore becomes more than a semiconductor supply issue.

It can influence the architecture of the AI economy itself.

That is why I’m increasingly uncomfortable calling this simply a “memory cycle.”

🐂 THE MEMORY BULL CASE IS STRONGER THAN JUST HIGH PRICES

For Micron, SK Hynix and Samsung, the attractive part is not simply that prices are high today.

It is the possibility that customer behaviour itself is changing.

Micron’s multibillion-dollar customer commitments are one example.

Samsung has also discussed multiyear binding supply contracts as buyers try to secure capacity.

SK Hynix’s CEO has gone even further, saying the company expects 2027 to be the industry’s worst year ever from a supply perspective, with demand potentially exceeding the company’s production capacity beyond 2030.

If true, memory begins behaving less like an interchangeable commodity bought every quarter and more like capacity that customers reserve strategically.

That matters because long-term agreements can improve:

Revenue visibility.

Pricing discipline.

Capacity planning.

Customer stickiness.

Return on capital.

And potentially the durability of margins.

That is what memory bulls should actually be watching.

Not simply whether Micron goes up another 6% tomorrow.

🐻 BUT WE CANNOT FORGET WHAT INDUSTRY THIS IS

There is one phrase that should terrify anyone who has followed memory stocks for long enough:

This time is different.

Memory cycles have repeatedly convinced investors that structural demand has eliminated cyclicality.

Then supply arrives.

Prices fall.

Margins compress.

Capital expenditure gets cut.

The cycle resets.

AI does not repeal economics.

High profits create an incentive to build capacity.

Samsung is expanding.

SK Hynix is expanding.

Micron is expanding.

Technology transitions increase output.

Customers redesign products.

Alternative suppliers improve.

China continues attempting to develop domestic memory capability.

Demand can also disappoint.

TrendForce already expects different outcomes across memory categories in 2027.

DRAM could remain tight because of HBM allocation and AI-server demand, while NAND supply may become easier in the second half of 2027 as new capacity arrives.

That distinction is important.

There is no single “memory market.”

HBM is different from commodity DRAM.

Server DRAM is different from PC memory.

NAND has different supply economics again.

A broad memory rally can hide very different fundamentals underneath.

So the investor question is not merely:

Is memory scarce?

It is:

Which memory remains scarce, for how long, and who actually captures the economics?

📊 THERE MAY ALSO BE A DEMAND-DESTRUCTION PROBLEM

There is another uncomfortable point for memory bulls.

Pricing power eventually creates its own enemy.

Customers have limits.

TrendForce says PC and smartphone customers are already reaching their ability to absorb further increases, even as AI server demand keeps the overall DRAM market tight.

That means a supplier can simultaneously experience:

Excellent pricing.

Huge margins.

Strong AI demand.

And collapsing demand elsewhere.

Initially, that can still be fantastic for profits because high-value server sales replace lower-value consumer sales.

But the farther the imbalance moves, the more complicated it becomes.

If smartphones become too expensive, fewer get sold.

If laptops become too expensive, replacement cycles lengthen.

If consumer SSDs become too expensive, buyers choose lower capacities.

Eventually demand destruction begins offsetting pricing gains.

The memory industry’s dream scenario is therefore not infinite prices.

It is:

High prices without killing the customer.

That’s a very different optimisation problem.

🌎 THIS COULD BECOME AN AI INFLATION STORY

There’s an even broader implication.

We normally discuss AI and inflation in terms of electricity.

Data centres use huge amounts of power.

Grid investment raises costs.

Copper demand rises.

Construction becomes more expensive.

But memory gives AI another channel into everyday inflation.

A cloud company spends billions on AI.

That raises demand for HBM and server DRAM.

Memory producers redirect capacity.

Consumer memory becomes more expensive.

Laptop manufacturers raise prices.

Smartphone manufacturers raise prices.

Automakers and electronics companies pay more.

Retailers pass those costs through.

The customer ultimately pays.

Industry groups representing automakers, retailers and electronics companies have already warned U.S. authorities that the memory imbalance could produce sustained price increases and disrupt supply chains.

UK electronics retailer Currys has similarly warned that the shortage is likely to push smartphone and laptop prices higher.

So perhaps AI inflation is not simply about electricity bills.

It may increasingly appear in:

Phones.

Computers.

Storage.

Cars.

Gaming hardware.

Networking equipment.

Almost anything containing memory.

🎯 MY TAKE: THE MEMORY RALLY IS SHOWING US WHERE THE ECONOMIC POWER IS MOVING

I started with the same question as everyone else:

Why are memory stocks rallying while indexes fall?

The easy answer is:

Strong pricing.

AI demand.

Supply shortages.

HBM growth.

But after digging through the supply chain, I think there is a much more important story.

AI is changing who memory manufacturers choose to serve.

And that changes who gets pricing power.

The memory companies gain it.

AI infrastructure customers can often afford it.

Premium device manufacturers may survive it.

Lower-margin consumer businesses struggle with it.

Budget products get squeezed.

Specifications get reduced.

Consumers delay upgrades.

And potentially, edge AI itself develops more slowly because the memory required to democratise local AI has become more valuable inside enormous cloud systems.

That creates one of the strangest paradoxes of the AI boom:

AI infrastructure could make artificial intelligence more powerful while simultaneously making the hardware needed to access that intelligence more expensive.

And if affordable devices become less capable of running AI locally, the AI industry may become even more dependent on centralised cloud infrastructure.

Which demands even more memory.

Which reinforces the shortage.

Which strengthens supplier pricing.

And the circle continues.

So I don’t think the biggest question for memory investors is:

How high can Micron go?

I think the bigger question is:

WHO PAYS FOR AI’S MEMORY HUNGER?

Because the answer increasingly looks like:

Everyone else using memory.

And that is a much bigger market story than Friday’s stock rally.

👇 COMMUNITY QUESTION

Here’s the question I would put to investors:

If AI infrastructure keeps absorbing the highest-value memory capacity, who ultimately feels the greatest impact?

Memory suppliers benefit from scarcity.

Hyperscalers pay more but need the capacity.

Premium device brands may have the pricing power to survive.

Lower-margin consumer electronics companies may not.

And consumers eventually see the result in prices and specifications.

But the angle I find most interesting is this:

Could cloud AI’s demand for memory actually slow the adoption of AI on consumer devices, reinforcing the cloud AI boom that created the shortage in the first place?

That is what I will be watching.

$MU $NVDA $AAPL $GOOGL $MSFT

Personal market analysis only. Not financial advice. Always do your own research.