Why Memory Chips Are Getting So Expensive

AI data centers are driving a surge in demand for HBM (high-bandwidth memory). As memory makers shift more production toward HBM, the supply of conventional memory used in phones and computers is tight.

On June 25, Apple raised prices across its entire Mac and iPad lineup, by about $250 on average, and pointed to one major reason: memory. Memory prices have roughly quadrupled over the past year. This memory shortage helped push Micron, SK Hynix and Samsung past $1 trillion in market value within just three weeks. This article explains what caused the memory shortage, why it matters even outside the tech industry, and what may happen next.

The Main Types of Memory

Before getting into the shortage, it helps to understand the main types of memory used in modern devices and AI systems.

DRAM, or dynamic random-access memory, is the working memory in computers, servers, and phones. It holds what you are using right now, such as open apps and browser tabs, and clears when the device is turned off.

HBM is an advanced form of DRAM designed for speed. Eight or more DRAM chips are stacked together and placed next to AI accelerators, including GPUs and custom AI chips. This allows them to move huge amounts of data quickly.

NAND is used for long-term storage. It keeps photos, files, and software saved after a device shuts down. In computers and servers, NAND chips are grouped together inside SSDs. Phones also use NAND, though it is usually built directly into the device.

SRAM, or static random-access memory, is used as ultra-fast cache inside processors. It is more expensive than DRAM and used in much smaller quantities.

For the current AI boom, the main focus is on HBM, DRAM, and NAND. AI systems need them to train models, run inference, and store enormous amounts of data.

Why the Memory Shortage Happened

The memory shortage comes down to three things: strong AI demand, limited existing capacity, and a long wait for new supply.

1. Strong AI demand

AI data centers need huge amounts of HBM. Samsung, SK Hynix, and Micron make about 90 percent of the world's conventional DRAM, and they are also the main suppliers of HBM.

As demand for AI chips grew, these companies shifted more production toward HBM. That put pressure on the supply of ordinary DRAM used in phones, laptops, and servers.

HBM also takes more capacity to produce. For the same amount of memory, it can use roughly three times as much silicon as conventional DRAM. This means that even a modest increase in HBM production can remove a much larger amount of capacity from the rest of the market.

2. Limited existing capacity

The timing made the problem worse. The memory industry went through a severe downturn in 2023 after the pandemic-era electronics boom faded. Demand for PCs, phones, and other devices slowed, while manufacturers were left with too much inventory or manufacturing capacity. Memory prices fell sharply, profits collapsed, and major chipmakers responded by cutting production and delaying factory expansions.

When AI demand accelerated, there was very little spare capacity available.

NAND faced a similar problem. Manufacturers had reduced production during the downturn, and demand later returned from data centers and consumer devices.

3. New supply takes years

Memory companies cannot add large amounts of production quickly. A new factory can cost tens of billions of dollars and take around 18 months to build. After that, manufacturers may need another year to install equipment, improve yields, and reach meaningful output.

The market was therefore hit by rising AI demand just as existing supply was tight, while new capacity was still years away.

What the Shortage Looks Like Today

By mid-2026, the shortage was showing up in four ways.

1. Memory Prices Jumped

Contract prices, which are the prices large buyers pay under supply agreements, rose sharply between 2025 Q3 and 2026 Q2. Conventional DRAM prices nearly quadrupled, while NAND prices nearly tripled. Prices are expected to keep rising in the third quarter, although at a slower pace.

Retail prices moved just as sharply. A 32GB RAM kit for a PC cost about $80 in mid-2025. By early 2026, it was selling for around $432.

Those increases flow through to every company that builds phones, PCs, servers, and storage products.

2. Consumer Devices Became More Expensive

Apple raised prices across its Mac and iPad lineup on June 25. The MacBook Air went from $1,099 to $1,299. The 14-inch MacBook Pro rose from $1,699 to $1,999. The iPad Air increased from $599 to $749.

Microsoft raised Xbox prices by $100 to $150. Samsung increased the price of the Galaxy S26 by $100.

The iPhone is another product to watch. It also uses DRAM and NAND, especially in models with more storage. Higher memory costs could show up in future iPhone prices.

Lower-priced devices face even more pressure because memory makes up a larger share of their total cost. According to HP, memory used to account for about 15 to 18 percent of the cost of building a PC. It now accounts for around 35 percent.

Device makers have a few choices. They can raise prices, reduce the amount of memory or storage, cut production, or accept lower profits. Gartner, an industry research company, expects the squeeze to reduce both PC and smartphone shipments this year.

3. Memory Stocks Soared

Higher prices have transformed the earnings of memory companies. Samsung, SK Hynix, and Micron all crossed $1 trillion in market value. SanDisk stock was up 40 times in the last 12 months.

Most of the profit growth came from higher selling prices. Micron shipped only a few percent more chips in its latest quarter, while its average selling price rose about 60 percent.

Memory stocks have had some sharp swings as investors questioned whether AI companies were building more computing capacity than they needed. Reports that Meta might sell some of its capacity added to those concerns. Even so, memory stocks are still trading near record highs after a huge run. Investors agree that memory remains scarce. The main question is how long the shortage will last.

4. Memory Is Adding to Inflation

Gartner, an industry research company, calls this effect "memflation", short for memory inflation. Higher memory costs raise the price of computers, phones, game consoles, storage devices, and data-center equipment.

For the broader economy, the effect is likely to remain modest. Households and businesses will still feel it when they replace phones, computers, and storage equipment. Some may delay upgrades and keep older devices for longer.

Why the Shortage Could Continue

Below are several factors that point to high prices through 2027.

1. Much of the Supply Is Already Booked

SK Hynix has sold its entire 2026 production of DRAM, NAND, and HBM. It is already booking customers for 2027. Micron says it can fill only half to two-thirds of the orders it receives.

Large buyers are also signing long-term contracts to secure supply. Some of Micron's agreements require customers to pay for committed capacity even if they later need fewer chips. This keeps more supply tied up for longer.

2. New Factories Take Time

A new cleanroom, the controlled space where chips are made, may take about 18 months to build. Reaching stable production can take another year.

Most of the large projects from Samsung, SK Hynix, and Micron will begin adding meaningful supply in 2028 or 2029. Investment during 2026 can improve existing plants. It cannot create a large amount of new factory space right away.

3. Memory Makers Are Expanding Carefully

The major suppliers remember the losses of 2023. They have strong reasons to avoid building too much capacity too quickly.

High prices and full factories are producing record profits. The companies can protect those profits by expanding at a measured pace.

4. AI Demand Remains Very Strong

Amazon, Microsoft, Google, and Meta are expected to spend about $700 billion on data centers in 2026. That is roughly 70 percent more than the year before. Some analysts expect their combined spending to exceed $1 trillion in 2027.

Each new AI cluster needs processors, HBM, regular DRAM, and storage. A slower rate of growth could still keep total memory demand high because the starting base is already so large.

Together, these factors are likely to keep memory prices high through 2027, with HBM remaining in short supply for even longer.

Why the Shortage Could Ease Sooner

There are also signs that the market may turn earlier than expected.

1. AI Spending Could Slow

The current shortage assumes that AI infrastructure spending will remain extremely strong for several years.

Cloud companies may become more careful if they have unused computing capacity. They may also improve the efficiency of their AI systems, get more work from existing hardware, or delay some data-center projects.

A modest slowdown would matter because so much future memory supply has been planned around continued AI growth.

2. Buyers Are Reaching Their Limit

Price growth is already slowing. TrendForce (a market research company) expects conventional DRAM prices to rise 13 to 18 percent in 2026 Q3, compared with about 60 percent in 2026 Q2.

Customers are struggling to absorb more increases. Phone and PC makers are cutting production plans. When they build fewer devices, they also order less memory.

3. New Supply Is Coming

China's CXMT is expanding DRAM production. Samsung, SK Hynix, and Micron also have major new facilities scheduled to ramp in 2028 and 2029.

Memory markets can change quickly when several factories begin producing at the same time. If new supply arrives while AI demand is slowing, the shortage could turn into oversupply.

Ordinary DRAM and NAND would probably become cheaper first. HBM is harder to make and package, so it is likely to remain tight for longer.

What May Happen Next

1. Memory prices are likely to stay high through 2027

AI data centers are still driving strong demand for HBM, while new factories take years to build and ramp up. That will likely keep DRAM and NAND expensive through 2027. HBM could remain tight for even longer.

2. 2028 may bring the first real relief

New capacity from Samsung, SK Hynix, Micron, and CXMT is expected to come online in 2028 and 2029. If AI spending slows at the same time, supply could finally catch up with demand and prices could begin to ease.

The timing is still uncertain. New factories can take longer than expected to reach full output, and continued growth in AI infrastructure could absorb much of the added supply.

3. What this means for you

If you already plan to buy a laptop, phone, console, or SSD, waiting may not save you money in the near term. Some products may become more expensive, while others may come with less RAM or storage at the same price.

For devices that cannot be upgraded later, it may be worth choosing the memory and storage you expect to need from the start. Many consumers may also decide to keep their current devices longer as replacement costs rise.

The memory shortage began because AI demand grew faster than the chipmakers could respond. The next phase will depend on which happens first: new supply catches up, or demand begins to slow. Until then, memory will remain one of the most important parts of the AI economy.