Post
The AI Era, Huge Changes in the Memory Market, and Apple's InfluenceAs the paradigm shifts from AI learning to inference, significant changes are expected in the memory market. In particular, if Apple fully enters the on-device AI inference market, new memory specifications optimized for smartphones could shake up the existing market. This analysis delves into how to prepare for these changes amidst the competitive landscape between Samsung and SK Hynix.
1. 📈 The Advent of the AI Inference Era and the Shifting Center of Gravity in the Memory Market
· The memory market landscape is changing as the paradigm in AI data centers shifts from 'learning' to 'inference.'
· In the past, due to the high value of HBM, investing in and expanding HBM production was the standard. However, the value of HBM is no longer as overwhelmingly dominant, leading to a competitive dynamic with general-purpose memory.
· This is leading to a game of strategic positioning among the global top three memory players (Samsung Electronics, SK Hynix, and Micron).
· In the past, due to the high value of HBM, investing in and expanding HBM production was the standard. However, the value of HBM is no longer as overwhelmingly dominant, leading to a competitive dynamic with general-purpose memory.
· This is leading to a game of strategic positioning among the global top three memory players (Samsung Electronics, SK Hynix, and Micron).
00:00 - 00:29
1. 📈 The Advent of the AI Inference Era and the Shifting Center of Gravity in the Memory Market
· The memory market landscape is changing as the paradigm in AI data centers shifts from 'learning' to 'inference.'
· In the past, due to the high value of HBM, investing in and expanding HBM production was the standard. However, the value of HBM is no longer as overwhelmingly dominant, leading to a competitive dynamic with general-purpose memory.
· This is leading to a game of strategic positioning among the global top three memory players (Samsung Electronics, SK Hynix, and Micron).
· In the past, due to the high value of HBM, investing in and expanding HBM production was the standard. However, the value of HBM is no longer as overwhelmingly dominant, leading to a competitive dynamic with general-purpose memory.
· This is leading to a game of strategic positioning among the global top three memory players (Samsung Electronics, SK Hynix, and Micron).
00:34 - 01:51
2. 🧠 Increasing Importance of Inference Computation and Memory Bottlenecks
· The importance of the 'inference' stage in AI data centers is growing, playing a crucial role in processing user input and generating information.
· 'Token' processing is becoming critical in the inference process. Currently, even the most advanced generation of HBM is struggling to handle hundreds of thousands of tokens.
· As a result, memory, rather than GPUs, is becoming the bottleneck for inference. Since HBM alone has limitations, technical solutions are needed to bypass this or increase memory capacity.
· 'Token' processing is becoming critical in the inference process. Currently, even the most advanced generation of HBM is struggling to handle hundreds of thousands of tokens.
· As a result, memory, rather than GPUs, is becoming the bottleneck for inference. Since HBM alone has limitations, technical solutions are needed to bypass this or increase memory capacity.
02:02 - 03:12
3. 🔗 Memory Vertical Integration Structure and the New Necessity for HBM-DRAM Interconnection
· The memory vertical integration structure consists of several stages, including computation cores, data processing, and data storage. Closer to the core, speed is faster but capacity is smaller, while further away, speed is slower but capacity is larger.
· HBM and DRAM have played important roles in this structure. However, as inference computation intensity increases, HBM alone is insufficient, and discussions are emerging about the need for new technical solutions between HBM and DRAM.
· HBM and DRAM have played important roles in this structure. However, as inference computation intensity increases, HBM alone is insufficient, and discussions are emerging about the need for new technical solutions between HBM and DRAM.
03:16 - 05:17
4. 💰 Changes in the HBM Market and Strategic Shifts Towards Increased Demand for General-Purpose Semiconductors
· SK Hynix's record-high performance is attributed to the high profitability of its HBM business, while Samsung Electronics' HBM share is still small.
· In the past, due to the high value of HBM, even general-purpose memory manufacturers preferred HBM production. However, the situation is changing as the price gap per gigabyte of HBM narrows compared to general-purpose memory.
· Consequently, the consensus is shifting towards a strategy of securing cash by lowering costs and increasing production capacity for general-purpose semiconductors, rather than making massive investments in HBM. This is leading to a strategic standoff in the global memory industry.
· In the past, due to the high value of HBM, even general-purpose memory manufacturers preferred HBM production. However, the situation is changing as the price gap per gigabyte of HBM narrows compared to general-purpose memory.
· Consequently, the consensus is shifting towards a strategy of securing cash by lowering costs and increasing production capacity for general-purpose semiconductors, rather than making massive investments in HBM. This is leading to a strategic standoff in the global memory industry.
05:18 - 07:15
5. 🔄 Investment Decisions in the Memory Industry: HBM vs. General-Purpose Semiconductors
· The memory industry faces a decision: whether to invest secured cash into next-generation HBM production facilities or to increase the production capacity of general-purpose semiconductors.
· HBM production is feasible only for a few companies, and competition among them is fierce. SK Hynix is contemplating whether to maintain its HBM-centric strategy, and its lower-than-expected performance could be interpreted as a signal of the market's shifting focus.
· The semiconductor industry, despite its substantial profits, requires continuous investment in equipment upgrades and R&D. This implies the need for investments amounting to tens of trillions of won periodically.
· HBM production is feasible only for a few companies, and competition among them is fierce. SK Hynix is contemplating whether to maintain its HBM-centric strategy, and its lower-than-expected performance could be interpreted as a signal of the market's shifting focus.
· The semiconductor industry, despite its substantial profits, requires continuous investment in equipment upgrades and R&D. This implies the need for investments amounting to tens of trillions of won periodically.
07:19 - 09:35
6. 🤝 Samsung vs. SK Hynix: Integrated Strategies for Foundry and Memory Businesses
· A competitive landscape is forming between the TSMC-SK Hynix alliance and the Samsung Foundry-Samsung Memory combination.
· Samsung Foundry's technological competitiveness is similar to TSMC's, but it has weaknesses in optimizing fabless custom processes and securing access channels. However, leveraging its in-house process technologies (2nm, 3nm) could be advantageous for producing next-generation memories like HBM4E.
· SK Hynix has advanced its packaging technology through collaboration with TSMC and may expand its foundry partnerships in the future. Additionally, securing in-house base die production capabilities is necessary to enhance its bargaining power and increase investments.
· Samsung Foundry's technological competitiveness is similar to TSMC's, but it has weaknesses in optimizing fabless custom processes and securing access channels. However, leveraging its in-house process technologies (2nm, 3nm) could be advantageous for producing next-generation memories like HBM4E.
· SK Hynix has advanced its packaging technology through collaboration with TSMC and may expand its foundry partnerships in the future. Additionally, securing in-house base die production capabilities is necessary to enhance its bargaining power and increase investments.
09:39 - 11:24
7. 📱 The Rise of the On-Device AI Inference Market and Apple's Potential
· The next phase of AI will move beyond online inference to on-device AI, which operates embedded in the real world.
· Apple is expected to play a significant role in the on-device AI inference market, leveraging its energy-efficient chip design and robust hardware ecosystem.
· New memory specifications optimized for smartphone form factors are likely to emerge, potentially causing a major upheaval different from the memory specifications in the existing PC/server market. This could create a situation similar to the past when Japanese semiconductor companies failed to adapt to changes in the mobile market.
· Apple is expected to play a significant role in the on-device AI inference market, leveraging its energy-efficient chip design and robust hardware ecosystem.
· New memory specifications optimized for smartphone form factors are likely to emerge, potentially causing a major upheaval different from the memory specifications in the existing PC/server market. This could create a situation similar to the past when Japanese semiconductor companies failed to adapt to changes in the mobile market.
11:25 - 15:37
8. 🍎 Apple's Hardware Optimization Strategy and Samsung's Response Plan
· Based on the high energy efficiency of its M-series chips, Apple could create a new market by demanding the development of low-power memory specifically for AI.
· Samsung, with its high market share in Galaxy smartphones, can respond to Apple by leveraging synergies between its foundry and memory divisions. Integrated development of Exynos chips and mobile DRAM is crucial.
· Apple's hardware optimization strategy is challenging, but if successful, it has the potential to significantly grow the smartphone-based AI market. Samsung must actively respond to these changes to seize the opportunity.
· Samsung, with its high market share in Galaxy smartphones, can respond to Apple by leveraging synergies between its foundry and memory divisions. Integrated development of Exynos chips and mobile DRAM is crucial.
· Apple's hardware optimization strategy is challenging, but if successful, it has the potential to significantly grow the smartphone-based AI market. Samsung must actively respond to these changes to seize the opportunity.
15:41 - 18:43
9. 💰 Performance Bonus Disputes in the Semiconductor Industry and Future Investment Strategies
· While there are recent disputes over performance bonuses due to high profits in the semiconductor industry, this also represents a positive aspect, showcasing the success of advanced strategic industries.
· The semiconductor industry must reinvest a significant portion of its profits into capital expenditures for the future, such as equipment upgrades and R&D investments. R&D costs are also increasing significantly due to rising equipment prices and growing technical complexity.
· Therefore, companies need a balanced strategy that secures cash, provides appropriate cash and stock-based compensation (like RSUs) to motivate employees, and simultaneously secures funds for future investments. This is a critical challenge for both Samsung and SK Hynix.
· The semiconductor industry must reinvest a significant portion of its profits into capital expenditures for the future, such as equipment upgrades and R&D investments. R&D costs are also increasing significantly due to rising equipment prices and growing technical complexity.
· Therefore, companies need a balanced strategy that secures cash, provides appropriate cash and stock-based compensation (like RSUs) to motivate employees, and simultaneously secures funds for future investments. This is a critical challenge for both Samsung and SK Hynix.