ADATA believes the memory shortage will last another 10 years and rules out discussing an AI bubble before 2030
ADATA President Chen Li-bai believes that the global memory shortage could last for another ten years, driven by the rapid growth of artificial intelligence in both data centers and local AI devices. The manager has also rejected warnings about a possible investment bubble and believes it is still too early to debate whether the AI market is overheated.
His statements come after Taiwanese stocks fell despite record results presented by TSMC. That move once again raised questions about the pace at which large cloud service providers are allocating capital to artificial intelligence infrastructure and whether real demand will be able to absorb all the computing power deployed.
Chen questions that interpretation and maintains that analyzing the market solely based on the capital expenditure of a few companies or the occasional use of their servers offers too limited a vision. In his view, AI applications will eventually spread across enterprise, government, and consumer markets, steadily increasing needs for computing power, storage, and memory.
The president of ADATA went so far as to say that the industry should wait until after 2030 to start discussing whether a possible AI bubble could occur in 2040 or 2050. He also identified electricity, especially from renewable sources, and memory as the two scarcest resources of the next decade.
The expansion of Samsung, SK hynix and Micron would not be enough to balance the market
The industry is increasing capacity, but ADATA does not expect these projects to quickly solve the imbalance between supply and demand. Chen believes that Samsung, SK hynix and Micron have learned from previous cycles of excess production and will maintain more rational and prudent expansion strategies to avoid new sharp price drops.
Chinese manufacturers are also expanding their production, although they continue to be limited by access restrictions to certain semiconductor manufacturing equipment. Added to this are the long lead times required to build wafer factories, prepare clean rooms and install machinery, so the new capacity would have a limited impact on global supply in the short term.
On this basis, Chen considers the forecasts that point to an upcoming relaxation of the market to be unfounded. ADATA expects DRAM contract prices to rise another 20% or 30% during the third quarter of 2026, while NAND memory could become between 35% and 40% more expensive in the same period.
The company also starts from a favorable position regarding these increases. ADATA had accumulated chip inventory worth more than 30 billion Taiwan dollars at the end of February and participated, along with other local module manufacturers, in debt operations and capital increases aimed at continuing to purchase components. Rising prices raise the value of those stocks and can sustain their profits through the second half of the year.
Demand will not only come from large data centers. Chen also points to robots, autonomous vehicles, automated factories and shops, smart homes and low-orbit satellites as future consumers of memory. The total number of connected devices in these markets could reach tens of billions of units, combining local processing with services run in the cloud.
Apparently, for ADATA, this expansion will cause the current shortage to stop behaving like a traditional short-term cycle and become a structural trend. Even with new factories from the big three suppliers and Chinese companies, Chen believes the industry will not be able to fully meet the demand generated by the extension of AI to the physical world over the next decade.
