Why memory is key element to the AI revolution

Nat King Cole had a memorable year in 1954.

That was when the legendary singer's "Unforgettable" album went platinum, meaning it sold one million copies. The title song was later inducted into the Grammy Hall of Fame in 2000 for its cultural impact.

And memory — particularly the long-term kind — is a key element of AI growth, as it enables systems to recognize patterns, learn from historical data, and make behavior-based predictions.

There's also a lot of money in memory. The global memory market hit record revenue of $170 billion last year, according to market-research firm Yole Group.

DRAM, or Dynamic Random-Access Memory, the most common computer memory, led the way at about $97 billion, while NAND, a non-volatile flash memory for long-term data storage, pulled in roughly $68 billion.

After a historic downturn in 2022–2023, the memory industry is beginning a strong recovery, fueled by AI-training requirements at data centers, with HBM playing a pivotal role due to its premium pricing and performance advantages, the research firm said, (HBM is a form of DRAM used to make AI chipsets.)

That number is projected to reach $200 billion this year, with DRAM and NAND coming in at about $129 billion and $65 billion respectively.

This year is pivotal "for the memory industry: The sector is delivering unprecedented revenue highs despite an environment of growing complexity and competition," Yole said.

"AI continues to serve as the central growth engine, while strategic shifts in supply chains and technology roadmaps redefine the competitive landscape."

Chinese players like ChangXin Memory Technologies and Yangtze Memory Technologies intensify global competition amid export controls and localization efforts, Yole said. Trade tensions with China have injected volatility into global supply chains, prompting users to pull their demand for chips forward.

YMTC is planning to expand into manufacturing DRAM chips, including advanced versions that are used to make artificial intelligence chipsets, Reuters reported on Sept. 25.

The move by the state-backed chipmaker underscores China's growing urgency to boost its capability to produce advanced chips after the U.S. expanded export controls in December. The goal of the restrictions was to restrict Beijing's access to high-bandwidth memory.

The biggest companies in the AI memory space are the HBM manufacturers SK Hynix of South Korea with 62% market share; Micron (MU) , Boise, Idaho, with 21%, and Samsung, also South Korea, which recently slipped to third place with 17%, according to data compiled by Counterpoint Research.

HBM is vital for performance-intensive applications like AI, high-performance computing, and graphics processing.

Analysts expect HBM revenue to double to roughly $34 billion this year from about $17 billion in 2024, as its role in AI infrastructure becomes increasingly critical.

Micron CEO: Trillions to be invested in AI

SK Hynix said in August that it expected the market for HBM chips used in AI to grow 30% annually until 2030, as cloud computing giants keep increasing their demand for AI.

The company said that Amazon (AMZN) , Microsoft (MSFT) , and Google (GOOGL) would likely spend even more on AI in the coming years, which will help boost demand for HBM chips.