Is the AI memory chip boom over?

Everything we know about the semiconductor industry, in one place. Get our report whether you build or invest.

Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us

In our report covering the semiconductor market, we turn what is vague and complex into simple, useful visuals. Market maps like this one show you at a glance where every startup stands.

Why are people asking if the AI memory chip boom is over?

Memory chip stocks fell hard in the AI selloff, even as the companies selling AI memory keep reporting record numbers.

Micron's latest quarterly revenue reached $54.23 billion, compared with $11.32 billion a year earlier, and it guided the next quarter to $61.5 billion plus or minus $1.5 billion. SK hynix reported 104.6 trillion won of quarterly revenue, up 32% from the previous quarter. Samsung said its memory business again set records for quarterly revenue and operating profit. Sandisk's revenue rose 23% sequentially to $11.03 billion.

The forward-looking data is strong too. SK hynix says long-term agreements cover more than 50% of its 2027 and 2028 production capacity. Sandisk is signing multi-year purchase agreements with data-center and edge customers. Samsung is expanding advanced-memory production, while Micron is shipping HBM4 for a lead customer's platform.

So for now the doubt comes from the stock market, not from the order books.

Why did Micron and Sandisk crash if memory demand is still strong?

Because Micron and Sandisk shares are trading on how profitable memory will be in the future, while today's orders and earnings are still near record levels.

On November 5, Sandisk fell 22% and Micron lost more than 17% in a broad semiconductor selloff. MarketWatch reported that the Philadelphia Semiconductor Index fell about 12%, its worst session since March 2020, as investors worried that political pressure over electricity prices, permitting and regulation could slow data-center construction.

The weakness had already started before that session. Bloomberg reported in late October that Sandisk, Micron, Seagate and Western Digital had moved into the bottom 10% of S&P 500 performers during the second half of the year even while memory earnings remained strong. Stocks moving before earnings do is the kind of signal we follow every quarter in our semiconductor industry report.

Memory stocks can fall well before revenue does. Investors only need to believe that shortages will ease, price increases will slow or customers will spend less later.

The table below shows the gap between the market reaction and the latest operating results.

Are memory-chip orders slowing, or are customers still locking in supply?

Customers are still locking in memory supply years ahead, especially for AI and data-center products.

SK hynix says long-term agreements now cover more than 50% of its total production capacity for 2027 and 2028. The company has signed these agreements with 14 major customers, including leading cloud providers, and says the contracts typically run for about five years.

Sandisk gives a similar picture in NAND. Its latest filing describes multi-year agreements with several data-center and edge customers. The agreements set purchase volumes, include pricing formulas and use financial guarantees. Sandisk says this model is expected to become its predominant way of doing business.

Those contracts also change how suppliers plan capacity. SK hynix can expand production with more visibility into future volumes, while Sandisk says the agreements improve production planning and supply assurance. Customers still have business risks of their own, but empty order books are not the problem today.

Our research about semiconductors gets updated several times a month, so you always get the latest view of where things stand.

Is HBM demand still growing as HBM4 arrives?

Yes. HBM demand is still climbing, and the market is already moving into HBM4.

SK hynix said HBM4 sales expanded in its latest quarter and expects HBM4 to become its mainstream high-bandwidth memory product in the fourth quarter. Samsung is already shipping commercial HBM4 and said expanding HBM4 sales helped its semiconductor business reach another record. Micron says its HBM4 is in high-volume shipments for a lead customer's platform and is being qualified by other customers.

The transition is happening while suppliers are still expanding capacity. SK hynix is pushing its M15X facility toward maximum capacity for HBM4, while Samsung and Micron are also increasing advanced-memory production.

Competition between suppliers can still move quickly because every new accelerator platform has to qualify its memory. Demand for the category itself remains strong.

Is ordinary server DRAM still getting an AI boost?

As a matter of fact, yes. AI servers are lifting demand for ordinary high-capacity server DRAM as well as HBM.

Samsung has said server products reached a record share of its memory sales and expects demand for server DRAM, enterprise SSDs and HBM to stay strong together. SK hynix is expanding DDR5 and LPDDR5 SOCAMM sales for AI servers. Micron has completed several customer qualifications for 8,800 MT/s server RDIMMs and started sampling 512GB DDR5 RDIMMs.

That makes sense at the system level. An AI server needs HBM next to accelerators, but it also needs large pools of system memory and storage around them. More inference and more complex serving workloads increase the amount of memory outside the accelerator package too.

The current semiconductor memory boom is broader than HBM alone.

Is NAND flash booming too, or only data-center flash?

Data-center NAND is booming, while consumer flash is moving in the opposite direction.

Sandisk's latest results make the split clear. Data-center revenue reached $4.80 billion, up 61% from the previous quarter. Edge revenue rose 7% to $5.83 billion. Consumer revenue fell 29% to $397 million. Total revenue still climbed 23% to $11.03 billion.

Sandisk then guided the next quarter to $12.2 billion to $12.6 billion of revenue. The midpoint is about 12% above the latest quarter even with consumer demand much weaker than the AI and enterprise side of the business. Data-center flash is booming while consumer flash is shrinking, which is why we look at who earns the money in our report on the semiconductor industry.

The table below shows how far Sandisk's end markets have moved apart.

In our market report about semiconductors, we look at what the winners do differently, so you can learn from the best instead of guessing.

Are memory prices starting to fall?

Not broadly. Memory pricing is still very strong, although investors are clearly expecting the pace of increases to cool.

Sandisk reported an 85.0% GAAP gross margin in its latest quarter, up from 29.8% a year earlier. Micron reported an 80.7% GAAP gross margin for fiscal 2026 and guided the next quarter to roughly 86%. Samsung said higher memory prices helped drive its record results.

Daily spot prices can weaken before contract prices, and memory categories do not all move together. The large suppliers are still reporting pricing strong enough to support record revenue and exceptional margins.

A slower rate of price increases would still hurt memory stocks because investors price the next part of the cycle before it appears in reported earnings.

How extreme are today's memory-chip profits?

Extremely. The latest memory-chip margins are far above normal cyclical levels, which makes even a modest slowdown a big deal for the stocks.

Micron's fiscal 2026 revenue reached $133.19 billion, up from $37.38 billion the year before, while its GAAP gross margin reached 80.7%. Sandisk's latest quarterly gross margin was 85.0%, compared with 29.8% a year earlier. SK hynix reported an 80% operating margin. Samsung's quarterly operating profit reached 107.4 trillion won as its memory business set another record.

Those profits come from a market where usable capacity has lagged demand badly enough for customers to pay up for supply. They also leave a lot of room for earnings to fall even if revenue keeps growing.

The table below compares the latest profitability numbers from the major suppliers.

Are Amazon, Google, Meta and Microsoft still spending enough to support memory demand?

Absolutely. The biggest cloud companies are still committing enormous amounts of money to AI infrastructure.

Amazon still expects about $220 billion of cash capital expenditures this year and says the increase in property and equipment purchases mainly reflects AI investment. Alphabet kept its full-year capex range at $195 billion to $205 billion and expects another significant increase next year. Meta expects $135 billion to $145 billion this year and says it plans to keep investing aggressively in AI. Microsoft added $39.37 billion of property and equipment in its latest quarter.

The demand behind that spending is still visible too. AWS revenue grew 39% year over year. Azure revenue grew 48%. Alphabet says it remains in a supply-constrained cloud environment and has started using more third-party compute capacity while it builds its own.

Not every dollar of capex becomes memory revenue, but these budgets fund the servers, accelerators, storage and data centers that consume HBM, DRAM and NAND.

The table below shows the current scale of hyperscaler infrastructure spending.

In our document detailing everything you need to know about semiconductor, you will always find fresh funding data, taken straight from our own verified funding database.

Could tighter data-center rules slow AI memory demand?

Yes, but the risk is still mostly ahead of the order book today.

The November semiconductor selloff accelerated as investors worried that a tougher US political environment around electricity prices, permitting and regulation could slow data-center construction. MarketWatch tied those concerns directly to the sharp drop in chip and AI-related shares.

A delayed data-center campus can push back server deliveries, which then pushes back demand for GPUs, HBM, DRAM and SSDs. Shares react to that possibility immediately because investors do not need to wait for a cancellation to appear in quarterly results. In our report covering the semiconductor market, we rate risks like data-center delays by probability.

The largest cloud companies are still expanding their infrastructure budgets, so there is no broad cancellation wave in the current data.

Could all the new memory capacity create another glut?

Quite possibly. The biggest fundamental risk to this memory boom is that supply catches up just as AI infrastructure growth slows.

Micron spent $27.37 billion on capital expenditures in fiscal 2026. SK hynix is expanding HBM production and expects investment to increase significantly as new capacity comes online. Samsung is lifting semiconductor investment. Kioxia and Sandisk plan more than $31 billion of investment in Japan through 2032, including more 3D flash capacity.

Most of that supply does not arrive instantly. New fabs, cleanrooms and advanced packaging lines take time to qualify, and HBM capacity cannot be swapped freely with every other type of memory. That delay is one reason shortages can survive even while suppliers spend heavily.

The danger comes later. Memory companies have repeatedly added capacity when prices were high, only to face weaker pricing once those projects started producing. Today's investment plans assume AI demand keeps absorbing much more memory for years.

What would tell us the AI memory boom is ending?

The clearest warning would be weakness spreading from share prices into orders, contract pricing, factory plans and cloud spending at the same time.

We would watch for customers cutting committed volumes, not just negotiating harder. Falling prices across HBM, server DRAM and enterprise NAND would be another warning. Supplier decisions would matter too: lower utilization, delayed fabs, weaker revenue guidance or cancelled capacity would be much harder to dismiss than a volatile stock chart.

The customer side is equally important. A real downturn would become much more convincing if Amazon, Alphabet, Meta and Microsoft started cutting infrastructure budgets or delaying large parts of their buildouts.

Those conditions are not broad enough today. Revenue guidance is still rising at Micron and Sandisk, SK hynix is expanding capacity under long-term agreements, Samsung expects strong AI-memory demand and the hyperscalers are still spending heavily.

Learning from others' mistakes is the cheapest lesson there is. In our report covering the semiconductor market, we look at what broke startups before you, so you don't repeat it.

Is the AI memory chip boom over, or did memory stocks just get ahead of themselves?

The AI memory chip boom is still alive, while the stocks have already moved on to worrying about the next downturn.

Today's business data remains strong across HBM, server DRAM and data-center NAND. Micron, Samsung, SK hynix and Sandisk are reporting record or near-record results, customers are signing longer agreements, and cloud companies are still pouring money into AI infrastructure.

The bearish case is about what happens after today's shortage. Suppliers are spending tens of billions of dollars on new capacity, data-center projects face more political scrutiny, and today's margins leave plenty of room to fall without turning the companies into weak businesses.

So the selloff deserves attention, but it is early evidence of a possible cycle turn, not confirmation that AI memory demand has already rolled over. We would call the boom over only when orders, pricing, capacity plans and cloud spending start weakening together.

Your time is limited. In our research about semiconductors, we put numbers on each part of the market, so you can see where it makes sense to focus.