Hong Kong-listed semiconductor stocks also rallied with gains expanding to 11%. Tianyue Advanced and Montage Technology jumped more than 6%

The semiconductor industry is known for the so-called "cyclical curse", featuring drastic price swings every 3 to 5 years. From the super cycle of memory chips in 2017-2018, the global chip shortage from 2020 to 2022, to the AI-fueled memory price surge in 2025-2026, every cycle brings violent shocks across the whole industrial chain.

The current cycle is marked by unprecedented structural differentiation:

• Consumer-grade memory faces supply shortages due to squeezed production capacity, with DDR4 particle prices jumping by as much as 369% within half a year
• 3. Accelerated localization of semiconductor equipment yet core technical barriers still exist
• 4. Circular economy concepts are gaining traction, and used mobile phone recycling has become a new way to ease chip shortages

Such structural differentiation not only reflects the evolution of technical routes, but also reveals that the semiconductor industrial chain is shifting from a simple manufacturing-sales model to a full ecosystem featuring independent equipment development, optimized manufacturing and recycling.

II. Upstream Breakthroughs: Industrial Significance of Domestic Etching Equipment Reaching 5nm Process

2.1 AMEC’s Flagship Dual Equipment Launch

On March 25, 2026, at the SEMICON China exhibition, two etching equipment models released by Advanced Micro-Fabrication Equipment Inc. (AMEC) sent shockwaves across the industry:

• • Primo Angnova™ ICP Etching System : Designed exclusively for logic and memory chips at 5nm and below, with an aspect ratio of 100:1, sub-angstrom (0.2Å) precision control and 40% higher single-chamber throughput
• • Primo Domingo™ High Selectivity Etcher : It solves world-class challenges in 3D device manufacturing and delivers a silicon nitride-to-silicon oxide etching selectivity ratio of 500:1

The groundbreaking value of these two equipment models lies not only in their technical parameters, but also in theircommercial verification:

Specific transmission mechanisms:

1. 1. Lower procurement costs : Domestic equipment is generally 30-40% cheaper than imported counterparts, directly cutting wafer fab CAPEX expenditure
2. 2. Optimized operating costs : Localized service networks provide 24-hour response, shortening spare part delivery cycles from 2 weeks to 48 hours
3. 3. Accelerated technical iteration : Joint R&D mechanisms between equipment suppliers and wafer fabs enable customized optimization for specific processes, targeting a production yield of 99.9%
4. 4. Supply chain security : Over 80% of core components are independently developed, eliminating reliance on a single supplier

2.3 Market Scale and Competitive Landscape

The global etching equipment market size is projected to reach USD 17.38 billion in 2026, among which etching equipment for 3D semiconductors accounts for over 60%. The current competition landscape follows a pattern of "three industry leaders plus one rising challenger":

AMEC adopts a differentiated strategy focusing on the 3D memory sector. It has captured more than 35% market share at ChangXin Memory and become the world’s third-largest etching equipment supplier.

III. Midstream Turbulence: Truth and Impacts of Memory Price Slumps

3.1 The "Memory Earthquake" Triggered by Google TurboQuant

On March 26, 2026, Google launched the TurboQuant compression algorithm, claiming it can cut the memory footprint of key-value cache (KV Cache) during large language model inference by at least six times. Right after the announcement, global memory giants lost over USD 90 billion in market capitalization within two trading days:

• • Micron Technology fell 3.4%, erasing USD 15.166 billion in market value
• • SK Hynix plunged 6.23%, losing USD 29.38 billion in market value
• • Samsung Electronics dropped 4.71%, wiping out USD 38.45 billion in market value

Nevertheless, the truth behind this "memory earthquake" is far more complex than it appears on the surface.

3.2 Structural Divergence: Collapsing Spot Prices vs Resilient Contract Prices

Spot Market (Consumer-Grade):

• • The price of a single 16GB DDR5 module plunged from a peak of RMB 1,200 to roughly RMB 700, representing a 27% monthly decline
• • The average retail price of DDR5 in Germany fell 7.2% month-on-month, with high-end models down nearly 19%
• • Corsair VENGEANCE 32GB memory kits dropped from USD 490 to USD 379.99

Contract Market (Enterprise-Grade):

• • Conventional DRAM contract prices jumped 90%-95% quarter-on-quarter in Q1 2026
• • NAND flash contract prices rose by 55%-60%
• • Samsung and SK Hynix maintain price hike expectations for Q2 amid robust demand from major enterprise clients

3.3 Transmission of Impacts to End-Consumer Electronics Markets

The impact of falling memory prices transmits asymmetrically across consumer electronics sectors:

1. 1. PC Market : DIY assembly demand is temporarily suppressed, while branded PC makers face eased cost pressure
• • Memory prices surged over 300% from late 2025 to early 2026, driving a more than 60% slump in DIY market sales
• • After the current price correction, a single 16GB DDR5 module is projected to fall to the RMB 300-400 range by late 2026
2. 2. Mobile Phone Market : Storage’s share of total costs has soared from 10-15% to 30-40%
• • All Redmi K90 models are priced RMB 300-600 higher than their previous generation
• • OPPO, VIVO and other brands rolled out synchronized price hikes, with mid-to-low thousand-yuan phones bearing the largest price adjustments
3. 3. Server Market : AI demand remains robust, keeping HBM prices firm
• • A single AI server requires 8 to 10 times more memory than a traditional server
• • HBM production capacity is prioritized, squeezing the supply of consumer-grade DDR5 by 25%

3.4 Jevons Paradox: Efficiency Gains Ultimately Boost Total Demand

Morgan Stanley’s analysis points out that the influence of TurboQuant technology is limited:

• • It only optimizes temporary cache for inference workloads and has no impact on HBM and weight storage required for model training
• • The Jevons Paradox will take effect: lower memory consumption per task may drive wider adoption of AI across more application scenarios

"Compression algorithms never truly cut total purchasing volumes. The Jevons Paradox tells us that efficiency improvements usually lead to higher overall consumption." — Shenzhen News Network, "Will AI Algorithms Pop the Flash Memory Bubble?"