AI is eating the world's memory chips, and Micron says the shortage lasts until 2028

If you have tried to buy a laptop, a graphics card or a new phone this year, you have probably noticed that everything with memory inside got more expensive. Most people blamed inflation or greedy manufacturers. The real reason is less obvious and much bigger: artificial intelligence is swallowing the world's memory chips, and the companies that make them now say the squeeze will not end soon. In fact, they say it will get worse.

This week two stories landed that belong together. Micron, one of only three big memory makers on the planet, told investors that supply will be even tighter in 2027 and 2028 than it is today. A few days later, Nikkei Asia reported that Apple has told suppliers to cut component orders for the brand new iPhone 18 Pro and Pro Max, because soaring memory costs pushed prices up and buyers held back. One is the cause, the other is the first big visible effect.

What Micron actually said

Micron CEO Sanjay Mehrotra used his fiscal fourth quarter earnings call at the end of September to deliver an unusually blunt outlook. "We expect memory and storage supply demand conditions to be much tighter in calendar 2027 and 2028 than they were in 2026," he said in prepared remarks reported by TechRepublic. On the call itself he added a short sentence that sums it up: "Overall, supply demand environment is only getting tighter," according to Ars Technica.

The most striking number is about time. More than 75 percent of Micron's planned output for fiscal 2027 is already committed to customers. Most new sales conversations are no longer about next year at all. They are about 2028. For high bandwidth memory in particular, Mehrotra said a large part of 2027 volume is sold out, and "the prices are much higher than 2026 prices."

Think about what that means for a moment. A company that sells a commodity, something that used to swing between glut and shortage every couple of years, is telling the market that next year is basically gone and the year after is filling up. That is not how a normal memory cycle looks. It is how a market looks when one type of buyer suddenly needs far more than the industry can build.

Why AI needs so much memory

That buyer is the AI data center. Modern AI accelerators from Nvidia, AMD and the custom chip makers do not work alone. Each one sits next to stacks of HBM, high bandwidth memory, which feeds the chip with data fast enough to keep it busy. The bigger the model, the longer the context window, and the more users and agents run at the same time, the more memory every accelerator needs.

Mehrotra listed exactly these drivers: "larger models, growing context, more concurrency, greater agents across enterprise, and consumer." Every trend that makes AI more useful also makes it hungrier for memory. Long conversations, AI agents that keep working for hours, and video models all push memory demand up, not down.

The catch is that HBM is expensive to make in more than just money. It uses much more wafer capacity per bit than ordinary DRAM, the memory inside your PC or phone. Micron calls this the "trade ratio," and it gets worse with each new generation. Moving from HBM3E to HBM4 and HBM4E means every gigabyte of AI memory costs even more factory space than before.

Samsung sees the same picture. Its executive vice president Kim Taewoo said this week that HBM will take almost 30 percent of DRAM makers' wafer capacity in 2027, up from about 20 percent this year, as Reuters reported and Ars Technica summarized. Every wafer that becomes AI memory is a wafer that does not become memory for laptops, phones, game consoles or streaming sticks.

New factories will not save us quickly

The obvious answer would be to build more factories. Micron is doing that. It plans new clean rooms that come online around 2028. But Mehrotra warned that a new building does not mean new chips the next day. "Even after they are built, even after first wafer output, production ramps up only gradually in the clean rooms," he said.

On top of that, shrinking memory cells to the next production node gives less extra output per wafer than it used to. So the industry is fighting on two fronts: it needs much more capacity, and each upgrade delivers less of a boost than in the past.

Micron still expects industry DRAM bit shipments to grow by roughly 20 percent per year in 2027 and 2028, and NAND flash by around 25 percent. That sounds like a lot. The problem is that AI demand is growing faster, and AI customers pay premium prices. When a hyperscaler signs a multi year contract for HBM, it gets served first. Consumer devices get what is left.

The consumer side: from Crucial to the iPhone

This is no longer a theoretical concern. Micron itself already left the consumer market. Its Crucial brand stopped selling RAM to consumers this year, a move announced in December 2025 so the company could focus on "larger, strategic customers in faster growing segments." In plain language: data centers pay better.

PC makers have reacted by shipping laptops with less RAM at higher prices. Cloud providers have, according to TrendForce data cited by TechRepublic, switched some servers to smaller memory modules because the big ones got too expensive. Prices for phones, consoles and other gadgets are up as well.

And now the most famous consumer product of all is feeling it. According to Nikkei Asia, as reported by CNA and Reuters, Apple has told some suppliers to cut October component orders for the iPhone 18 Pro and Pro Max by at least 15 percent compared with what it originally requested. Nikkei cites several people familiar with the matter. Apple did not comment, and Reuters could not independently verify the report.

The reason given is simple. Memory chip costs went up, Apple passed part of that to buyers, and buyers became more careful. The iPhone 18 Pro starts at 1,199 US dollars and the 18 Pro Max at 1,299 US dollars, both 100 dollars more than the iPhone 17 Pro models a year ago. Apple had already raised iPad and MacBook prices in June, saying it could no longer shield customers from memory and storage costs driven by the AI data center buildout.

To be fair, Nikkei also notes that softer demand from late August into October could partly come from Apple changing its launch schedule. One report about order cuts is not proof of a collapse. But it fits a wider pattern that analysts have warned about for months: memory makers prioritize AI, device makers pay more, and in the end someone has to decide whether a phone is still worth the price.

The other side of the coin: chip makers are booming

While device makers struggle, the companies building AI hardware are having record months. TSMC, which manufactures most of the world's advanced AI chips, reported September revenue of about 511.9 billion Taiwan dollars, roughly 16 billion US dollars, an increase of 54.6 percent on a year earlier, CNBC reported. Its third quarter revenue rose about 50 percent and beat analyst estimates.

Put the numbers side by side and the picture is clear. Money is flowing into AI infrastructure at a pace the supply chain cannot match. Logic chips, memory, power and buildings are all in short supply at once. Memory is simply the bottleneck where ordinary consumers feel it first, because the same factories make chips for both worlds.

What this means for you

If you are planning to buy hardware, a few practical points follow from Micron's outlook.

• Do not wait for prices to fall. Memory makers themselves see no balance before 2028 at the earliest. If you need a new laptop or phone, waiting a year is unlikely to make it cheaper.
• Check RAM configurations carefully. Some manufacturers ship lower memory versions at the old price points. A cheaper looking model may simply have less RAM inside.
• Consider extending the life of devices. For many users, upgrading the storage or memory of an existing PC, where possible, beats buying a new machine in a seller's market.
• For businesses, plan earlier. TechRepublic suggests treating memory as a component to budget for early in hardware refresh plans, rather than as a predictable commodity purchase.

There is also a bigger lesson. For two years, the AI boom mostly showed up in stock prices and chatbot updates. Now it shows up in the price tag of the phone in your pocket. AI is not only software in the cloud. It is a physical industry that competes with the rest of the economy for chips, electricity and factory space.

Can anything change the outlook?

A few things could ease the pressure. If AI spending slows, for example because companies struggle to earn money from their models, demand for HBM would cool and capacity could shift back to regular DRAM. Chinese memory makers are expanding as well and could add supply for consumer products. And new memory technologies that use capacity more efficiently might help, although none are close to mass production.

Right now, though, every signal points the other way. Hyperscalers keep raising their capital budgets, AI labs sign multi year compute deals worth tens of billions of dollars, and AI agents that run for hours need ever more memory per user. Micron's own words are probably the best summary: it does not know when supply will catch up with demand.

For anyone following AI, this is one of the most important stories of the year, even if it lacks the glamour of a new model launch. The future of AI is not only decided by who has the smartest model. It is also decided by who can get enough memory to run it, and right now the answer is: the biggest buyers first, everyone else later.

Sources: [Ars Technica](https://arstechnica.com/information-technology/2026/10/memory-supplies-are-only-getting-tighter-micron-ceo-says/), [TechRepublic](https://www.techrepublic.com/article/news-micron-memory-shortage-2028-ai-demand/), [Nikkei Asia](https://asia.nikkei.com/business/technology/exclusive-apple-cuts-iphone-18-pro-orders-due-to-soft-demand), [CNA / Reuters](https://www.channelnewsasia.com/business/apple-cuts-iphone-18-pro-orders-due-soft-demand-nikkei-asia-reports-6445331), [CNBC](https://www.cnbc.com/2026/10/08/tsmc-september-sales-hit-another-record-as-ai-boom-rolls-on.html).