As AI computing supply expands, demand also surges “This is different from past investment cycles” Wall Street: “The boom could continue through 2028”

At the 57th annual general meeting of Samsung Electronics held at the Suwon Convention Center in Yeongtong-gu, Suwon, Gyeonggi Province, last March, a shareholder viewed HBM4 and HMB4E memory products. [Yonhap News] 사진 확대 At the 57th annual general meeting of Samsung Electronics held at the Suwon Convention Center in Yeongtong-gu, Suwon, Gyeonggi Province, last March, a shareholder viewed HBM4 and HMB4E memory products. [Yonhap News]

As investment in artificial intelligence (AI) data centers surges, forecasts are mounting that the memory semiconductor boom could last longer than expected. Analysts say a longer “memory supercycle” may emerge as demand that had been concentrated on AI-oriented high-bandwidth memory (HBM) spreads to conventional DRAM and NAND.

According to the Financial Times (FT) on the 7th (local time), Daniel Roberts, co-CEO of an NVIDIA data-center partner, said, “It is really difficult to imagine a situation in which the supply curve overtakes demand.” He assessed that current investment in AI data centers is unfolding differently from past commodity and investment cycles.

Roberts explained that the AI industry is creating a structure in which new demand emerges as computing resources increase. Newly supplied computing capacity does not merely meet existing demand; it “creates several times more additional demand,” he said.

As the use of AI agents expands and data-processing speeds increase, the need for computing resources is also growing. This makes it highly likely that investment in data centers will continue.

The scale of data-center expansion is also expected to grow rapidly. Goldman Sachs forecasts that U.S. data-center capacity will double by the end of 2027 from its 2024 level and reach approximately 125 gigawatts (GW) in 2030, more than triple the 2024 figure. IREN also plans to invest up to $30 billion in AI infrastructure by June next year.

Some analysts say the factors constraining data-center expansion are practical problems arising during construction rather than a lack of demand. Roberts noted that data-center operators are already facing “social, political and physical limits” in securing power. Opposition is also growing in various parts of the United States over the impact of data centers on local electricity and water supplies.

As data-center investment continues, demand for memory semiconductors is also rising rapidly. Demand is spreading beyond HBM used in AI servers to server and low-power DRAM, tightening supply and increasing upward pressure on prices.

Inside an SK hynix plant. [SK hynix] 사진 확대 Inside an SK hynix plant. [SK hynix]

As the expansion of HBM production constrains conventional DRAM production capacity, demand for memory and data storage needed for AI servers is also increasing. Pressure from supply shortages is therefore spreading to the NAND market.

Against this backdrop, Wall Street is focusing on the possibility that the current memory boom could last longer than previous ones. Bank of America forecasts that fundamentals in the memory industry will continue to improve until between mid-2027 and before 2028, when newly added production capacity begins full-scale operations.

The expansion of long-term supply contracts with major customers is also cited as a difference from past cycles. Bank of America analyzed that long-term contracts could account for 50% to 70% of future production capacity, potentially reducing volatility in memory prices and companies’ earnings compared with the past.

There are risks as well. High memory prices are encouraging companies to expand capacity, while Chinese manufacturers are also increasing production capacity. Once new supply enters the market in earnest, price growth could slow. The possibility also remains that investment in AI data centers could decelerate faster than expected.

Roberts nevertheless placed greater weight on the possibility that, in the AI industry, increased supply will not stop at meeting existing demand but will generate new demand again. “Even if it is risky to say that this time is different, I believe there is something fundamentally different from past commodity and investment cycles,” he said.

This article has been translated by GripLabs Mingo AI.