AI Computing Power "Eats" Memory: DRAM Prices Soar Over 60%, Domestic Chips Break Through Against the Trend

May 18, 2026

A "supercycle" driven by artificial intelligence (AI) is propelling the global memory industry to unprecedented heights. In the second quarter of 2026, contract prices for mainstream memory products generally surged by 60% to 70%, with some server and mobile memory chips even recording quarterly increases exceeding 150%. While Samsung, SK Hynix, and Micron are shifting over 80% of their production capacity to high-profit AI-specific memory, downstream mobile phone and PC industries are experiencing an unprecedented cost storm. Meanwhile, Changxin Memory, a leading domestic memory manufacturer, delivered an "explosive" performance, successfully becoming the fourth largest player in the global DRAM market.

According to the latest data released by TrendForce, in the second quarter of 2026, DRAM (Dynamic Random Access Memory) contract prices surged by 58%-63% quarter-on-quarter, while NAND Flash (flash memory) prices rose even more dramatically by 70%-75%. This marks the second consecutive quarter of significantly higher-than-expected price jumps.

The price increases for specific product categories are even more astonishing. 64GB DDR5 memory for AI servers saw a 150% increase in a single quarter, 12GB LPDDR5X memory rose by 130%, and some high-specification products even reached 300%. Meanwhile, traditional LPDDR4X contract prices also surged by 70%-75%. Analysts point out that the current price increases and speed of memory have exceeded the industry peak of 2017-2018, a phenomenon rarely seen in the past decade.

“The fundamental reason is simple: AI is consuming all available capacity,” said an analyst at ICwise Research. Samsung, SK Hynix, and Micron collectively control over 90% of the global memory market. In pursuit of the ultra-high profit margins of HBM (High Bandwidth Memory) and enterprise-grade SSDs, they have shifted over 80% of their new and adjustable capacity to these AI-related products. The HBM capacity gap remains as high as 50%-60%, severely squeezing general-purpose DRAM capacity. One HBM chip consumes the equivalent of 3-4 standard DDR5 chips, and this supply-side distortion is transmitted throughout the entire market.

The price surge in memory costs is being borne by end consumers and system integrators.

Driven by AI, the arms race among the three major memory manufacturers has intensified.

Samsung plans to extend its HBM technology to mobile devices, increasing bandwidth in phones and tablets by 30%, and will provide samples of its next-generation CXL 3.1 memory modules to customers in the third quarter. Simultaneously, Samsung has officially discontinued its LPDDR4/4X product line, focusing entirely on the high-profit AI sector.

SK Hynix, leveraging its deep partnership with Nvidia, sold nearly $5.2 billion worth of products to Nvidia in the first quarter, a 62.6% year-on-year increase. The company's market capitalization has surpassed 1 trillion won, making it the second-largest company in South Korea by market capitalization. Even more noteworthy is that, thanks to record profits, its employees' annual performance bonuses are expected to reach an average of approximately 700 million won (about 3.2 million yuan) per person.

Micron, having recovered from massive losses three years ago, is projected to become the sixth most profitable publicly traded company in the United States. The company has launched samples of 256GB DDR5 RDIMM memory modules with speeds reaching 9200 MT/s and has pledged to invest $150 billion to expand its factories in the United States.

Amidst this euphoria, concerns are emerging.

IDC predicts that DRAM market revenue will reach $418.6 billion in 2026, nearly doubling year-on-year. Goldman Sachs believes the supply-demand gap will reach a new high since 2011, and supply tightness may continue until 2027. However, historical experience shows that high profits can trigger aggressive capacity expansion, and the market is wary of a potential oversupply around 2028. Furthermore, the continuous improvement in the efficiency of AI models may reduce the absolute demand for hardware in the future.

“Every chip we use now is fighting for AI, but whether demand will always outpace production capacity, no one dares to bet,” a veteran industry insider frankly stated.