As global tech giants Crazy hoarding (madly stockpile) computing power to train large-scale AI models, a silent 'battle' is unfolding over memory chips: On one side, AI servers exhibit an insatiable demand for high-bandwidth memory, while on the other, ordinary consumers find the prices of smartphones, graphics cards, and laptops visibly rising.

Since 2026, memory chips have entered a three-quarter streak of price increases. Samsung Electronics delivered its best-ever quarterly results, with operating profit surging 1,814% year-on-year. Global monthly memory chip sales hit an industry record of $74.6 billion, solidifying the cycle where 'the hotter AI gets, the pricier consumer electronics become.' At its core, this battle represents a reallocation of limited memory production capacity between the AI computing power industry and traditional consumer electronics—and it has only just entered its mid-game phase.

AI Computing Power Surges, Driving Simultaneous Rise in Memory Chip Volume and Price

Memory chips are undergoing a rare market rally of rising both volume and price. According to Samsung Electronics' Q2 financial report disclosed in late July, consolidated revenue reached KRW 171.5 trillion, up 130% year-on-year, while operating profit soared 1,814% to KRW 89.5 trillion—both record highs—with the semiconductor division contributing KRW 89.2 trillion in operating profit. Samsung explicitly stated in its earnings report that demand growth for server DRAM, enterprise-grade SSDs, and HBM is expected to accelerate further, with supply tightness in the memory chip market projected to persist.

Industry data supports this assessment. UBS's Monthly Memory Chip Report revealed global memory chip sales reached $74.6 billion in July 2026, shattering the industry's monthly record with a 31.7% month-on-month surge. Core DRAM sales accounted for approximately $48 billion, up 27.7% sequentially. UBS forecasts DRAM contract prices will rise 32% quarter-on-quarter in Q3 2026 and another 18% in Q4. Structural DRAM shortages are expected to persist until at least Q2 2028.

The primary driver of these price hikes is the explosive growth of AI computing power. A single AI server consumes 8-10 times more memory than standard devices, with model training and inference driving exponential demand for memory bandwidth and capacity. Microsoft's Q2 earnings showed Azure cloud revenue growing 43% above expectations, validating AI's commercialization. Public data reveals Meta has committed nearly $700 billion in medium-to-long-term spending on computing power and cloud infrastructure, while Anthropic plans to build a 1.6-gigawatt computing park. With global AI computing expansion showing no signs of slowing, memory chips—the 'logistical backbone' of computing systems—see strengthening demand certainty.

HBM Becomes a 'Capacity Black Hole,' with Supply Shortages Spanning Years

A more dramatic structural shift is occurring within the memory chip sector. HBM (High Bandwidth Memory), the core computing foundation for AI servers, is consuming production capacity at an astonishing rate. TrendForce estimates HBM wafer consumption is 4-5 times that of DDR5, with 2026 HBM shipments projected to grow 60% year-on-year and another 60% in 2027. SEMI China President Feng Li noted the 2026 HBM market is expected to expand 58% to $54.6 billion, accounting for roughly one-tenth of DRAM market value.

Currently, Samsung, SK Hynix, and Micron have allocated 70% of their new capacity to HBM, yet a 50-60% capacity gap remains. As of Q1 2026, all HBM production from the three major manufacturers is fully booked, with some key clients securing capacity through 2028. NVIDIA CEO Jensen Huang stated the global HBM shortage represents 'not a short-term market fluctuation but a structural industry challenge lasting years.' Samsung has signed long-term supply agreements with five major data centers and warns shortages will persist until 2028.

This dynamic has profoundly reshaped the memory industry's profit structure. HBM margins are 3-5 times higher than consumer-grade memory products, with NVIDIA, Microsoft, and Google having already secured long-term orders through 2027. Driven by profits, leading manufacturers continue reallocating capacity from smartphone and PC DRAM/NAND to HBM and enterprise storage, creating a reinforcing cycle where 'the hotter AI gets, the pricier consumer electronics become'—the very industrial root of the current consumer electronics price surge.

Price Hikes Ripple Through the Entire Chain: From Servers to Huaqiangbei

The memory chip price surge is cascading through the industry chain to consumer endpoints. Huaqiangbei, China's electronics hub, felt the impact first. According to CCTV Finance, major graphics card brands implemented warehouse lockdowns following new price hike notices from global manufacturers, rapidly tightening spot market supply. RTX 5070 retail prices jumped from 4,500-5,000 yuan last year to 6,000-6,500 yuan currently, while RTX 5080 soared from 8,000 yuan to nearly 12,000 yuan—'changing daily.' NVIDIA notified graphics card makers of GPU package price hikes affecting both GDDR7 and previous-gen GDDR6 memory.

Smartphone and PC memory costs have surged similarly. TrendForce data shows 12GB smartphone memory costs rose from 200 yuan to 600 yuan, while 1TB SSD prices tripled. The 'disappearance of sub-5,000 yuan laptops' became a trending topic, with popular gaming laptops rising over 5,000 yuan in six months. Even with subsidies, high-end notebook prices universally exceed 10,000 yuan. Consumer upgrade thresholds have risen sharply, with distributors expecting this price wave to last at least a year.

On the flip side, the entire industry chain is seeing profitability recovery. Memory manufacturers' profit elasticity has been fully unleashed in this cycle, benefiting upstream design, module packaging, and materials/equipment sectors. As Yuan Yu, Deputy Director of CCID Consulting's Electronics Information Industry Research Center, observed, HBM's 3-5x profit margins over consumer memory have drawn manufacturers to continuously reallocate consumer-grade capacity, propelling the entire memory industry into a prosperous phase of rising volume and prices.

Supercycle Far From Peaking, Industrial Value Undergoes Accelerated Revaluation

The sustainability of this memory supercycle is increasingly supported by data. TrendForce sharply raised its 2026 global memory market forecast in May: from $551.6 billion to $889.3 billion, with 2027 expected to surpass $1.28 trillion (44% YoY growth). DRAM revenue is projected to reach $618.7 billion in 2026 (+303% YoY), while NAND flash memory markets are forecast at $270.6 billion (2026) and $379.4 billion (2027).

Structural demand shifts provide long-term growth support. As AI applications transition from model training to Agentic AI centered on inference, KV Cache capacity requirements expand proportionally with context windows, directly boosting HBM and DRAM demand. The world's top nine cloud providers are projected to increase 2026 capex by 79%, shifting investment logic from 'on-demand expansion' to large-scale AI infrastructure deployment.

TrendForce data shows server-related applications now account for 60-65% of DRAM market demand, a share expected to exceed 65% by 2027 and approach 70% by 2028, tightening the bond between memory chips and AI computing power.

In this 'century battle,' China's memory industry chain is also seizing accelerated growth opportunities. Domestic memory chip leaders are expanding production capacity with steadily improving manufacturing yields, while design, manufacturing, equipment, and materials ecosystems take shape, enhancing China's voice in the global memory supply landscape.

From an industry cycle perspective, the upward trajectory of memory chips remains clear due to the trifecta of AI-driven demand, supply rigidity, and long-term order lock-ins. However, the sustainability of high prices and the pace of downstream demand realization remain variables requiring ongoing industry observation—this memory 'century battle' has only just begun its prologue.

- End -