AlphaWire

x_repost_queue

Main line in July: AI moves from computing power to bottleneck to make up for shortcomings The strongest market in the first half of the year is computing power

2026-07-07·x-repost-20260707-232008
Main line in July: AI moves from computing power to bottleneck to make up for shortcomings The strongest market in the first half of the year is computing power and storage But starting from July, funds will not just focus on the GPU itself, but will continue to dig into the industry chain: I will divide the focus of July into three levels: The first level: the hardest industrial catalysis 1.

Memory chip/HBM/DRAM AI servers eat up a lot of HBM, DRAM, and SSD demands. Storage is no longer just a cycle reversal, but one of the core bottlenecks in the expansion of AI computing power. For U.S. stocks, look at MU, for South Korea, look at Samsung and SK Hynix, and for A-shares, look at the direction of domestic storage and storage modules.

This line combines the logic of price increases, tight supply and demand, increased volume of AI servers, and domestic substitution. 2. Advanced packaging AI chips continue to be upgraded and cannot rely solely on advanced processes. Chiplets, HBM, CoWoS, and 2.5D/3D packaging will continue to increase the importance of advanced packaging.

Globally, look at TSMC, ASE, and Amkor, and A-shares look at the direction of advanced packaging, packaging and testing equipment, materials, and carrier boards.

Second layer: Material bottlenecks in packaging and computing power expansion 3. Glass substrate/advanced carrier board Glass substrates are not an isolated subject, but as advanced packaging continues to evolve, the material side is being repriced.

Once the market starts to speculate on next-generation packaging materials, glass substrates, ABF carrier boards, and advanced packaging substrates will all be put on the stage. This line looks at material factories, carrier board factories, and packaging equipment factories globally.

A-shares look more at the mapping of glass substrates, electronic glass, and packaging materials. 4. PCB upstream material end AI servers, switches, and high-speed interconnection have increasingly higher requirements for PCB materials.

Compared with ordinary PCB, what is more worth looking at are upstream materials, copper clad laminates, high-frequency and high-speed materials, and Low-Dk/Low-Df materials. Globally, look at high-speed PCB, CCL, and server chains; A-shares look at copper clad laminates, high-frequency high-speed materials, and AI server PCB supply chains. 5.

AI computing power metal Indium, tungsten, gallium, zirconium, and rare earths are essentially not just speculation in resource stocks, but the key material constraints behind the expansion of computing power. AI chips, advanced packaging, optical modules, power devices, and robots all use different key metals.

This line is highly elastic, but it depends more on price, supply and demand, and policy changes than just concepts. Layer 3: Overflow from data centers and terminal applications 6. CPO / Optical Communications / Optical Fiber As computing power clusters become larger and larger, data transmission will become a bottleneck.

High-speed, low-power consumption, and low-latency connections are inevitable upgrade directions for AI data centers. U.S. stocks look at optical modules, switches, and network chips; A-shares look at optical modules, optical chips, optical fiber cables, and CPO-related directions. 7.

Liquid-cooled server/data center power After the power consumption increases, heat dissipation and power supply are just needed. The logic of liquid cooling is straightforward: the stronger the server, the more important thermal management is; the larger the data center, the more important power, UPS, power distribution, energy storage, and transformers are.

US stocks look at data center power, temperature control and infrastructure; A shares look at liquid cooling, temperature control, power supply, power grid equipment and data center construction. 8. Physics AI / Humanoid Robot This is the direction AI is taking from the cloud to the real world.

Look at rotation in the short term, products and orders in the medium term, and industry trends in the long term. US stocks look at the underlying platform, complete machines, warehouse robots, machine vision and edge computing; A shares look at reducers, actuators, sensors, servo systems and structural parts.

Only personal research opinions, not investment advice

Full text

Main line in July: AI moves from computing power to bottleneck to make up for shortcomings The strongest market in the first half of the year is computing power

Main line in July: AI moves from computing power to bottleneck to make up for shortcomings The strongest market in the first half of the year is computing power and storage But starting from July, funds will not just focus on the GPU itself, but will continue

Main line in July: AI moves from computing power to bottleneck to make up for shortcomings The strongest market in the first half of the year is computing power and storage But starting from July, funds will not just focus on the GPU itself, but will continue to dig into the industry chain: I will divide the focus of July into three levels: The first level: the hardest industrial catalysis 1. Memory chip/HBM/DRAM AI servers eat up a lot of HBM, DRAM, and SSD demands. Storage is no longer just a cycle reversal, but one of the core bottlenecks in the expansion of AI computing power. For U.S. stocks, look at MU, for South Korea, look at Samsung and SK Hynix, and for A-shares, look at the direction of domestic storage and storage modules. This line combines the logic of price increases, tight supply and demand, increased volume of AI servers, and domestic substitution. 2. Advanced packaging AI chips continue to be upgraded and cannot rely solely on advanced processes. Chiplets, HBM, CoWoS, and 2.5D/3D packaging will continue to increase the importance of advanced packaging. Globally, look at TSMC, ASE, and Amkor, and A-shares look at the direction of advanced packaging, packaging and testing equipment, materials, and carrier boards. Second layer: Material bottlenecks in packaging and computing power expansion 3. Glass substrate/advanced carrier board Glass substrates are not an isolated subject, but as advanced packaging continues to evolve, the material side is being repriced. Once the market starts to speculate on next-generation packaging materials, glass substrates, ABF carrier boards, and advanced packaging substrates will all be put on the stage. This line looks at material factories, carrier board factories, and packaging equipment factories globally. A-shares look more at the mapping of glass substrates, electronic glass, and packaging materials. 4. PCB upstream material end AI servers, switches, and high-speed interconnection have increasingly higher requirements for PCB materials. Compared with ordinary PCB, what is more worth looking at are upstream materials, copper clad laminates, high-frequency and high-speed materials, and Low-Dk/Low-Df materials. Globally, look at high-speed PCB, CCL, and server chains; A-shares look at copper clad laminates, high-frequency high-speed materials, and AI server PCB supply chains. 5. AI computing power metal Indium, tungsten, gallium, zirconium, and rare earths are essentially not just speculation in resource stocks, but the key material constraints behind the expansion of computing power. AI chips, advanced packaging, optical modules, power devices, and robots all use different key metals. This line is highly elastic, but it depends more on price, supply and demand, and policy changes than just concepts. Layer 3: Overflow from data centers and terminal applications 6. CPO / Optical Communications / Optical Fiber As computing power clusters become larger and larger, data transmission will become a bottleneck. High-speed, low-power consumption, and low-latency connections are inevitable upgrade directions for AI data centers. U.S. stocks look at optical modules, switches, and network chips; A-shares look at optical modules, optical chips, optical fiber cables, and CPO-related directions. 7. Liquid-cooled server/data center power After the power consumption increases, heat dissipation and power supply are just needed. The logic of liquid cooling is straightforward: the stronger the server, the more important thermal management is; the larger the data center, the more important power, UPS, power distribution, energy storage, and transformers are. US stocks look at data center power, temperature control and infrastructure; A shares look at liquid cooling, temperature control, power supply, power grid equipment and data center construction. 8. Physics AI / Humanoid Robot This is the direction AI is taking from the cloud to the real world. Look at rotation in the short term, products and orders in the medium term, and industry trends in the long term. US stocks look at the underlying platform, complete machines, warehouse robots, machine vision and edge computing; A shares look at reducers, actuators, sensors, servo systems and structural parts. Only personal research opinions, not investment advice

← Back to archive