Why buying your next phone now beats waiting it out

The world is running short of memory chips again, but this time the problem looks very different from the semiconductor crisis that disrupted electronics between 2020 and 2023. Back then, factories were affected by lockdowns, shipping delays, and broken supply chains. In 2026, the shortage is being driven by something more fundamental: the explosive demand for artificial intelligence.

The memory-chip industry is increasingly prioritising the technology needed to power AI data centres. Manufacturers such as Samsung Electronics, SK Hynix, and Micron are shifting a greater share of their production capacity towards high-bandwidth memory, or HBM. Unlike conventional DRAM used in smartphones and computers, HBM is designed to move enormous amounts of data quickly between processors and memory. It has become a critical component in AI accelerators used by companies building large-scale AI systems.

This shift is causing an issue. Even though a lot of money is being invested in making memory, manufacturers cannot just make as much of it as they want. They do not have factory space, and they also do not have unlimited advanced packaging and production equipment. When they use more of their space to make HBM they have space to make the regular memory that people use in their everyday devices.

This is where people started talking about "RAMmageddon" or "RAMpocalypse". These names may sound funny. They show a real problem for companies that make things like smartphones. Some companies that work with AI are willing to pay a lot of money for special memory, so manufacturers would rather sell to them than to the companies that make smartphones.

The biggest buyers of memory are technology companies like Microsoft, Amazon, Google and Meta. These companies are spending a lot of money to make their data centres bigger and to work on AI. Because they are so big they can buy a lot of memory at once. Make deals that last a long time. This means that smaller companies that make electronics have time to get the memory they need and they have to pay more money for it.

The effects are already reaching the smartphone industry. Rising memory costs increase the expense of manufacturing a phone, and companies ultimately have to decide how much of that additional cost they can absorb. Some may raise retail prices, while others could reduce storage options, delay product launches or use cheaper components elsewhere.

Budget smartphone manufacturers are particularly exposed because their customers are highly sensitive to price increases. Companies competing in the lower-cost segment cannot easily increase prices without risking weaker sales. Premium manufacturers, meanwhile, generally have greater margins and more flexibility to absorb higher component costs.

The problem is not restricted to smartphones. Computers, servers, gaming devices and automobiles all depend on memory chips. Modern vehicles increasingly contain sophisticated electronic systems, meaning a disruption in semiconductor supply can affect everything from infotainment systems to advanced driver-assistance technology.

What makes the current situation particularly complicated is the uncertainty over how long it could last. Building new semiconductor capacity is extraordinarily expensive and takes years. Manufacturers also cannot simply construct factories based on temporary demand because memory markets are notoriously cyclical. A company that invests billions during a shortage could find itself facing oversupply if demand falls before the new capacity becomes operational.

The AI boom has therefore created a difficult calculation for memory manufacturers. Investing in HBM can produce attractive returns because demand from AI companies is enormous. But consumer electronics still depend heavily on conventional memory, and starving those markets of supply risks pushing prices higher and reducing device sales.

The situation also reveals an unexpected consequence of the AI revolution. Much of the public discussion around artificial intelligence focuses on software, chatbots and powerful processors. Far less attention is given to the physical infrastructure underneath them. Every AI model ultimately depends on enormous data centres filled with processors, storage systems, networking equipment and memory.

The memory shortage is a reminder that the AI economy does not exist only in the cloud. Its effects can reach an ordinary consumer standing in an electronics store deciding whether to upgrade a phone.

For now, the central question is not simply whether more memory chips will eventually be produced. It is whether manufacturers can expand capacity quickly enough to satisfy two very different markets at once: the rapidly growing AI industry demanding specialised high-bandwidth memory and billions of consumers still needing ordinary memory for phones, laptops, cars, and other devices.

The pandemic-era semiconductor shortage was largely a story of supply chains unable to keep up with demand. This new shortage is different. This time, the supply chain is being reshaped by manufacturers’ view of where the biggest profits and fastest growth will be. And if AI keeps gobbling up a larger and larger chunk of the world's memory, consumers might feel the cost of the AI boom more and more every time they buy a new gadget.

References:

. . .