With talk of an AI investment bubble spreading, the head of memory chipmaker ADATA forecast that memory shortages will persist for a long period. Demand is expanding beyond AI data centres to robots, autonomous vehicles and smart homes, making it hard for memory supply to keep up over the next 10 years, the report said.

TechRadar, an IT media outlet, reported on July 23 that ADATA Chairman Chen Li-bai (천리바이) dismissed market talk of an AI bubble that emerged after TSMC's earnings announcement. "To talk about a real bubble, it would have to be after 2030," he said, calling current concerns premature.

Chen said demand for AI computing capability, memory chips and power is still exceeding market forecasts. Meta's decision to lease out surplus AI computing resources has raised the possibility of overinvestment by cloud providers, but he drew a line against interpreting it as a sign of a broader slowdown in the AI market.

He said judging the entire AI industry based only on the size of some companies' capital spending or factory utilisation rates is like "looking at the sky through a straw" and overlooks a long-term expansion in demand. He said AI use will spread at the same time across companies, governments and consumer services, as well as across various industrial ecosystems.

He also forecast that memory supply shortages will not be resolved easily. While major players such as Samsung Electronics, SK Hynix and Micron are moving to expand production capacity, Chen cited eco-friendly power and memory as the scarcest resources over the next 10 years. He expects AI demand growth to outpace the speed of supply expansion.

This outlook also aligns with a recent rise in memory prices. According to market research firm 3D Center, DDR5 memory kit prices in the German market rose 7 percent in July alone to a record high. Average DDR5 prices jumped 448 percent from the same period last year, rising more than fourfold in a year.

Chen said new memory demand will rise rapidly not only from AI data centres but also from robots, autonomous cars, unmanned factories, unmanned stores, smart homes, low-Earth-orbit satellites and related ground infrastructure. He forecast that even combining the capacity of major memory makers with China's production capacity would make it difficult to meet such demand over the next 10 years.

As a result, memory makers' expansion strategies are expected to differ from the past. Rather than expanding large-scale capacity all at once as before, a step-by-step investment approach that watches demand is likely to take hold, the report said.

The industry is focusing on the possibility that as AI demand spreads beyond central data centres to a variety of physical devices and infrastructure, memory shortages could become a structural phenomenon rather than a temporary business cycle.

The burden on consumers and PC makers is also expected to continue for the time being. With investment in AI infrastructure and demand for new devices rising at the same time, memory prices are not expected to stabilise quickly. Memory supply and demand is expected to become a key variable in the semiconductor market going forward.