As global tech giants frantically hoard computing power to train large models, a "battle" for memory chips is quietly unfolding: on one side is the almost insatiable demand for high-bandwidth memory from AI servers, and on the other, ordinary consumers are noticing the visibly rising prices of smartphones, graphics cards, and laptops.
Since 2026, memory chips have entered a price hike cycle for three consecutive quarters. Samsung Electronics has delivered its best-ever results with an operating profit surging 1,814% year-on-year, and global monthly memory chip sales have hit an industry record of $74.6 billion. The cycle of "as AI heats up, consumer electronics get pricier" is becoming a reality. The essence of this battle is the reallocation of limited memory capacity between the AI computing power industry and the traditional consumer electronics industry—and it has just reached its halfway point.
AI Computing Power Surges, Memory Chips See Volume and Price Growth
Memory chips are experiencing a rare cycle of simultaneous volume and price growth. According to the financial report disclosed by Samsung Electronics at the end of July, the company achieved consolidated revenue of 171.5 trillion KRW in the second quarter, a year-on-year increase of 130%; operating profit reached 89.5 trillion KRW, a massive year-on-year surge of 1,814%, both hitting record highs, with the semiconductor division contributing 89.2 trillion KRW to the operating profit. Samsung Electronics explicitly stated in the report that the growth rate of demand for server DRAM, enterprise SSDs, and HBM is expected to accelerate further, and the tight supply situation in the memory chip market is expected to continue.
Industry data corroborates this judgment. The "Monthly Memory Chip Report" released by UBS shows that in July 2026, global memory chip sales reached $74.6 billion, setting the highest monthly record in the industry's history, a substantial month-on-month increase of 31.7%. Among them, sales of the core category DRAM were approximately $48 billion, a month-on-month growth of 27.7%. UBS expects that DRAM contract prices will rise by 32% quarter-on-quarter in the third quarter of 2026, and continue to rise by 18% in the fourth quarter. At least until the second quarter of 2028, the supply of DRAM will remain in a state of structural shortage.
The core driver of the price hike stems from the explosive growth of AI computing power. The memory consumption of a single AI server is 8 to 10 times that of ordinary equipment, and the demand for memory bandwidth and capacity for large model training and inference is climbing exponentially. Microsoft's second-quarter financial report shows that the Azure cloud business grew by 43%, exceeding market expectations, serving as a strong testament to the commercialization of AI; according to public information, Meta has locked in nearly $700 billion in medium- to long-term computing power and cloud infrastructure expenditures, and Anthropic plans to build a new 1.6-gigawatt computing power park. The pace of global AI computing power expansion has not slowed down. As the "ammunition" for computing systems, the certainty of demand growth for memory chips continues to strengthen.
HBM Becomes a "Capacity Black Hole", Supply Shortage Persists for Years
Within memory chips, a more drastic structural adjustment is taking place. HBM (High Bandwidth Memory), as the core computing power foundation for AI servers, is devouring capacity at an astonishing rate. According to TrendForce's analysis, the wafer consumption of HBM is about 4 to 5 times that of DDR5. HBM shipments are expected to grow by 60% year-on-year in 2026, and another 60% in 2027. Feng Li, President of SEMI China, pointed out that the HBM market size is expected to grow by 58% to $54.6 billion in 2026, accounting for about one-tenth of the DRAM market output value.
Currently, the three major memory manufacturers—Samsung Electronics, SK Hynix, and Micron Technology—have allocated 70% of their new capacity to HBM, yet the HBM capacity gap remains as high as 50% to 60%. As of the first quarter of 2026, the HBM capacity of the three major manufacturers has been completely sold out, with some core customers even locking in capacity until 2028. NVIDIA CEO Jensen Huang explicitly stated that the global HBM supply shortage is "not a short-term market fluctuation at all, but a structural industry dilemma that will last for years." Samsung Electronics has signed long-term supply agreements with five major data centers and warned that the supply shortage will continue until 2028.
This landscape has profoundly changed the profitability structure of the memory industry. The profit margin of HBM is 3 to 5 times that of consumer-grade memory products, and manufacturers such as NVIDIA, Microsoft, and Google have locked in long-term orders up to 2027 in advance. Driven by profits, leading manufacturers continue to divert capacity from consumer-grade DRAM and NAND fields, such as smartphones and PCs, to HBM and enterprise storage, forming a reinforced cycle of "as AI heats up, consumer electronics get pricier"—which is precisely the industrial root cause of this round of consumer electronics price hikes.
Price Hike Transmission Across the Entire Chain: From Servers to Huaqiangbei
The wave of memory chip price hikes is transmitting layer by layer along the industrial chain to consumer terminals. Huaqiangbei was the first to feel the change. According to CCTV Finance, affected by a new round of price hike notices issued by major global graphics card manufacturers, major graphics card brand factories have fully implemented warehouse closure policies, and the spot supply in the entire circulation link is rapidly tightening. The retail price of the RTX 5070 graphics card has risen from 4,500 to 5,000 RMB last year to the current 6,000 to 6,500 RMB; the price of the RTX 5080 has climbed from 8,000 RMB to nearly 12,000 RMB, basically "changing price every day." NVIDIA has issued a GPU bundle price hike notice to graphics card manufacturers, with the price increase involving both GDDR7 and the previous generation GDDR6 memory.
The storage costs for smartphones and computers have also risen accordingly. According to TrendForce data, the cost of 12GB smartphone memory has risen from 200 RMB to 600 RMB, and the unit price of 1TB flash memory has simultaneously tripled. The topic of "models under 5,000 RMB have basically disappeared" in the laptop market has sparked widespread discussion. The price of popular gaming laptops has increased by more than 5,000 RMB in half a year. Even with subsidies and discounts, the prices of high-end laptops have generally broken through the 10,000 RMB mark. The threshold for ordinary consumers to change their phones has been significantly raised, and dealers generally expect this round of price hikes to last for at least one year.
The other side of the price hike transmission is the comprehensive recovery of profitability in the industrial chain. The profit elasticity of memory manufacturers has been fully released in this cycle, and upstream sectors such as memory design, module packaging, and materials and equipment have benefited simultaneously. As Yuan Yu, Deputy General Manager of the Electronic and Information Industry Research Center at CCID Consulting, said, the profit margin of HBM is 3 to 5 times that of consumer-grade memory. The severe supply shortage of HBM attracts manufacturers to continuously divert consumer-grade capacity, pushing the overall memory industry into a boom cycle of simultaneous volume and price growth.
Super Cycle Has Not Yet Peaked, Industrial Value Accelerates Reassessment
The sustainability of this round of memory super cycle is being supported by an increasing amount of data. In May this year, TrendForce significantly raised its forecast for global memory output value: the global memory output value in 2026 was raised from the previous $551.6 billion to $889.3 billion, and is expected to exceed $1.28 trillion in 2027, with an annual growth rate of about 44%. Among them, the DRAM output value is expected to reach $618.7 billion in 2026, a year-on-year increase of as high as 303%; the market size of NAND flash memory in 2026 and 2027 is expected to be $270.6 billion and $379.4 billion, respectively.
Structural changes on the demand side provide long-term support for growth. As AI applications shift from model training to Agentic AI centered on inference, the capacity demand for KV Cache expands proportionally with the context window, directly driving up the demand for HBM and DRAM; the capital expenditures of the world's top nine cloud service providers are expected to grow by 79% in 2026, and the investment logic has shifted from "expanding on demand" to large-scale deployment of AI infrastructure.
TrendForce data shows that in the current DRAM market, server-related applications already account for 60% to 65%, and this proportion will climb to over 65% in 2027, moving towards 70% in 2028. The binding between memory chips and AI computing power will become increasingly tight.
In this "century-defining battle," China's memory industry chain is also ushering in a window for accelerated growth. The capacity of leading domestic memory chip companies continues to be released, manufacturing yield is steadily optimized, and the supporting system in various links such as design, manufacturing, equipment, and materials is accelerating its formation. The voice of domestic memory in the global supply pattern is steadily increasing.
From the perspective of the industry cycle, the superposition of three factors—AI-driven demand, supply rigidity, and long-term order lock-ins—makes the upward logic of memory chip prosperity still clear. Of course, the sustainability of high-price operations and the realization pace of downstream demand will remain variables that the industry needs to continuously observe in the future—this "century-defining battle" revolving around memory has just written its prologue.
- End -