Last week, DeepSeek R1 was released, attracting significant global attention.
On January 27, DeepSeek topped the free app download charts in the Apple App Store for both China and the United States, causing its official website to crash due to overwhelming traffic that morning. The emergence of DeepSeek marks a “milestone” moment for AI in China. By creating an open-source AI model comparable to OpenAI’s o1 at an extremely low cost, investors are concerned about the profitability of AI and the strong demand for cutting-edge chips.
On the evening of the 27th, all three major U.S. stock indices fell collectively, with stocks like NVIDIA and Broadcom plummeting over 12% at the open due to the impact of DeepSeek, while TSMC dropped over 8%. By the close, NVIDIA had fallen about 17%, with a single-day market value evaporation of $589 billion, marking the largest single-day loss in history for a stock.
Notably, on January 28, the service status page of DeepSeek’s official website indicated that its online services had been subjected to large-scale malicious attacks. To continue providing services, registration methods other than +86 phone numbers were temporarily restricted, while registered users could log in normally. Additionally, in the early hours of January 28, the DeepSeek team officially released two multimodal frameworks—Janus-Pro and JanusFlow.
Global DeepSeek Craze and Heavy Losses in U.S. Tech Stocks
Since last weekend, a global craze for DeepSeek has erupted.
On January 27, the DeepSeek app topped the free app download charts in the Apple App Store for both China and the United States, surpassing ChatGPT in the U.S. download rankings, becoming a phenomenon.
It is worth noting that just a few days after the release of the fully open-source inference model DeepSeek R1, people have already come up with various crazy ideas using it. For instance, some have achieved AGI by running DeepSeek on multiple Macs at home, while others have implemented open-source o1 on consumer-grade hardware, and some have enhanced other models’ intelligence by leveraging DeepSeek’s inference capabilities.
Due to DeepSeek’s creation of an open-source AI model comparable to OpenAI’s o1 at an extremely low cost, investors are worried about the profitability of AI and the strong demand for cutting-edge chips.
On the evening of the 27th, in response to DeepSeek’s impact, tech stocks like NVIDIA and Broadcom plummeted. By the close, NVIDIA had dropped nearly 17%, with a single-day market value evaporation of $589 billion, marking the largest single-day loss in history for a stock. Broadcom fell 17.4%, Micron Technology dropped 11.67%, TSMC fell 13.23%, AMD dropped 6.37%, and Microsoft fell 2.14%.
Notably, last Friday, Goldman Sachs released a report stating that with the strong rise of China’s AI large model DeepSeek, hedge funds have begun to observe whether the AI models cultivated in the U.S. can sustain their momentum, and have started to withdraw their bets on tech stocks.
Bruno Schneller, Managing Director of Erlen Capital Management, stated, “The emergence of global competitors like China’s AI startup DeepSeek raises questions about the sustainability of the U.S.’s dominance in this field.”
However, Wedbush analyst Daniel Ives disagrees, stating that this sell-off is “another golden buying opportunity,” rather than a real threat to the AI revolution. “The real AI arms race is in artificial general intelligence, robotics, and autonomous technologies—areas where DeepSeek cannot compete at all.”
Dan Bin, chairman of Dongfang Hongyuan, who is heavily invested in NVIDIA, also expressed on social media, “Market sentiment is too extreme. The most direct impression is that you can create good AI without chips. But that’s not the case. Innovation in algorithms and continuous investment in computing power are not mutually exclusive. The direct impact of DeepSeek is a significant reduction in market inference costs, which is beneficial for application promotion; this is the core contradiction of future computing power demand, and cost reduction is the fundamental condition. Additionally, Dan Bin stated that he will explore more domestic AI opportunities in 2025.
Is the Era of AI ASICs Here?
In recent years, the competition among major tech giants in large models has been a battle of computing power, with capital expenditures on chips reaching hundreds of billions of dollars. The most advanced chips are not only in short supply but also face export restrictions, giving financially strong Silicon Valley tech giants a significant advantage.
However, DeepSeek’s ability to train breakthrough models with extremely low-cost chips is beginning to shake this status quo.
According to reports from New Intelligence, a team led by UC Berkeley PhD student Pan Jiayi and two other researchers replicated DeepSeek R1-Zero in the CountDown game. Through reinforcement learning (RL), a 3B foundational language model can self-validate and search, with costs of less than $30 (approximately 217 yuan), yielding impressive results.
A team led by Assistant Professor He Junxian at Hong Kong University of Science and Technology (co-authored by Huang Yuzhen and Weihao Zeng) replicated the training of DeepSeek-R1-Zero and DeepSeek-R1 on a 7B model using only 8K samples, achieving strong results in complex mathematical reasoning.
Additionally, the team at HuggingFace, the world’s largest open-source platform, officially announced the replication of all pipelines of DeepSeek R1. Once completed, all training data, scripts, etc., will be fully open-sourced.
Some have stated on social forums that while other major companies are scrambling to purchase H100 chips, DeepSeek quietly accomplished something: with less than $6 million in extremely low costs and under the conditions of 2048 chips with performance far below H100 and Blackwell’s H800 chips, the DeepSeek team created an open-source AI model with performance comparable to OpenAI’s o1, while Anthropic and OpenAI’s training costs reached as high as $1 billion.
Looking ahead, the emergence of DeepSeek R1, through a new low-cost model of “extreme compression + efficient reinforcement training + significant simplification of AI inference computing power,” heralds the arrival of the AI ASIC era.
ASIC (Application-Specific Integrated Circuit) refers to chips designed for specific tasks. In the AI field, ASIC specifically refers to chips customized for deep learning, neural network training/inference, and other AI tasks.
The “AI ASIC era” signifies a shift in AI technology from “software-driven” to a stage of deep synergy between software and hardware. With its advantages in computing power, energy efficiency, and customization, ASIC is becoming the “new infrastructure” for AI, driving breakthroughs in large models, autonomous driving, smart terminals, and other fields. In the future, as technologies like Chiplet and advanced packaging mature, ASIC will further reduce costs and become the core engine for the democratization of AI.
Using DeepSeek to Look Ahead to the Year of the Snake in A-Shares
Analyzing the Chinese stock market trends for 2025 requires a comprehensive consideration of multiple factors, including macroeconomic conditions, policy direction, industry trends, and international environment. Below, we provide a rational analytical framework from various dimensions:
1. Macroeconomic Fundamentals
If GDP growth stabilizes around 5% in 2025 (the target for the 14th Five-Year Plan), and corporate profit expectations improve, the stock market will receive fundamental support.
Key variables: the strength of consumption recovery (residential savings conversion rate), export resilience (global industrial chain restructuring), and the soft landing process of the real estate industry.
2. Policy Environment
Fiscal and monetary policy: If a moderately loose tone continues (such as expansion of special bonds, structural interest rate cuts), ample liquidity may benefit stock market valuations.
Reform dividends: Policies such as the deepening of the registration system, acceleration of state-owned enterprise reforms, and cultivation of data factor markets may reshape market structure.
3. Industry Differentiation and Structural Opportunities
1. Policy-driven sectors
Technological independence: Fields such as semiconductors, AI, and quantum computing that are “choke-point” areas will receive long-term investment from the state, and hard tech companies may see valuation premiums.
Green transformation: The new energy (photovoltaics, energy storage), electric vehicle industry chain, and carbon neutrality derivative industries (carbon trading, energy-saving technologies) will continue to benefit.
Security themes: Sectors such as food security, energy security, and information technology application innovation (ITAI) will receive policy support.
2. Consumer and cyclical sectors
Consumption recovery: If residents’ income expectations improve, optional consumption sectors such as food and beverages, healthcare, and cultural tourism may see a recovery.
Traditional industries: The real estate chain (building materials, home appliances) needs to observe marginal policy changes; resource sectors (non-ferrous metals, chemicals) are influenced by the global commodity cycle.
4. Market Structure and Capital Flow
1. Foreign Capital Flow
After the MSCI expansion window closes, the pace of foreign capital inflow may slow, but if the RMB exchange rate stabilizes and Sino-U.S. relations improve, it may still bring incremental capital.
Attention should be paid to the global liquidity environment (whether the Federal Reserve’s interest rate hike cycle has ended).
2. Strengthening of Domestic Capital Dominance
The proportion of long-term funds such as pensions and insurance entering the market may increase, potentially reducing market volatility and promoting value investment styles.
The proportion of retail investors is decreasing, and institutional pricing power is strengthening, leading to a more rational market.
5. Risks and Challenges
1. External Uncertainties
Geopolitical conflicts, recurring global inflation, and risks of economic recession in Europe and the U.S. may trigger a phase of foreign capital withdrawal.
2. Internal Concerns
Local debt pressure, risks in small financial institutions, and long-term issues such as population aging, if not handled properly, may suppress market sentiment.
The surge in the number of listed companies under the registration system may exacerbate individual stock differentiation, and the risks of “speculating on small and poor stocks” need to be heeded.
6. Scenario Analysis
Optimistic scenario: Moderate economic recovery + continuous policy support + easing external environment → The Shanghai Composite Index may challenge 4000 points.
Neutral scenario: Structural growth + localized risk release → The market may maintain a range-bound fluctuation (3000-3500 points).
Pessimistic scenario: Black swan events (such as global recession, geopolitical crises) → Short-term deep corrections may occur, but policy support is expected.
7. Investor Strategy Recommendations
1. Long-term allocation: Focus on policy dividends and industry trends (such as high-end manufacturing, digital economy), avoiding high-valuation bubble sectors.
2. Tactical operations: Pay attention to turning points in economic data (such as PMI, social financing) and signals released by important meetings (Central Economic Work Conference).
3. Risk hedging: Appropriately allocate safe-haven assets such as gold and government bonds to diversify single-market risks.
Summary
The Chinese stock market in 2025 is likely to present a “structural market,” with policy-driven and performance-stable industries expected to outperform the broader market, but caution is needed regarding the volatility brought about by the resonance of internal and external risks. Investors should downplay index fluctuations, delve into niche sectors, and maintain flexible positions to respond to uncertainties.
Last week, DeepSeek R1 was released, attracting significant global attention.
On January 27, DeepSeek topped the free app download charts in the Apple App Store for both China and the United States, causing its official website to crash due to overwhelming traffic that morning. The emergence of DeepSeek marks a “milestone” moment for AI in China. By creating an open-source AI model comparable to OpenAI’s o1 at an extremely low cost, investors are concerned about the profitability of AI and the strong demand for cutting-edge chips.
On the evening of the 27th, all three major U.S. stock indices fell collectively, with stocks like NVIDIA and Broadcom plummeting over 12% at the open due to the impact of DeepSeek, while TSMC dropped over 8%. By the close, NVIDIA had fallen about 17%, with a single-day market value evaporation of $589 billion, marking the largest single-day loss in history for a stock.
Notably, on January 28, the service status page of DeepSeek’s official website indicated that its online services had been subjected to large-scale malicious attacks. To continue providing services, registration methods other than +86 phone numbers were temporarily restricted, while registered users could log in normally. Additionally, in the early hours of January 28, the DeepSeek team officially released two multimodal frameworks—Janus-Pro and JanusFlow.
Global DeepSeek Craze and Heavy Losses in U.S. Tech Stocks
Since last weekend, a global craze for DeepSeek has erupted.
On January 27, the DeepSeek app topped the free app download charts in the Apple App Store for both China and the United States, surpassing ChatGPT in the U.S. download rankings, becoming a phenomenon.
It is worth noting that just a few days after the release of the fully open-source inference model DeepSeek R1, people have already come up with various crazy ideas using it. For instance, some have achieved AGI by running DeepSeek on multiple Macs at home, while others have implemented open-source o1 on consumer-grade hardware, and some have enhanced other models’ intelligence by leveraging DeepSeek’s inference capabilities.
Due to DeepSeek’s creation of an open-source AI model comparable to OpenAI’s o1 at an extremely low cost, investors are worried about the profitability of AI and the strong demand for cutting-edge chips.
On the evening of the 27th, in response to DeepSeek’s impact, tech stocks like NVIDIA and Broadcom plummeted. By the close, NVIDIA had dropped nearly 17%, with a single-day market value evaporation of $589 billion, marking the largest single-day loss in history for a stock. Broadcom fell 17.4%, Micron Technology dropped 11.67%, TSMC fell 13.23%, AMD dropped 6.37%, and Microsoft fell 2.14%.
Notably, last Friday, Goldman Sachs released a report stating that with the strong rise of China’s AI large model DeepSeek, hedge funds have begun to observe whether the AI models cultivated in the U.S. can sustain their momentum, and have started to withdraw their bets on tech stocks.
Bruno Schneller, Managing Director of Erlen Capital Management, stated, “The emergence of global competitors like China’s AI startup DeepSeek raises questions about the sustainability of the U.S.’s dominance in this field.”
However, Wedbush analyst Daniel Ives disagrees, stating that this sell-off is “another golden buying opportunity,” rather than a real threat to the AI revolution. “The real AI arms race is in artificial general intelligence, robotics, and autonomous technologies—areas where DeepSeek cannot compete at all.”
Dan Bin, chairman of Dongfang Hongyuan, who is heavily invested in NVIDIA, also expressed on social media, “Market sentiment is too extreme. The most direct impression is that you can create good AI without chips. But that’s not the case. Innovation in algorithms and continuous investment in computing power are not mutually exclusive. The direct impact of DeepSeek is a significant reduction in market inference costs, which is beneficial for application promotion; this is the core contradiction of future computing power demand, and cost reduction is the fundamental condition. Additionally, Dan Bin stated that he will explore more domestic AI opportunities in 2025.
Is the Era of AI ASICs Here?
In recent years, the competition among major tech giants in large models has been a battle of computing power, with capital expenditures on chips reaching hundreds of billions of dollars. The most advanced chips are not only in short supply but also face export restrictions, giving financially strong Silicon Valley tech giants a significant advantage.
However, DeepSeek’s ability to train breakthrough models with extremely low-cost chips is beginning to shake this status quo.
According to reports from New Intelligence, a team led by UC Berkeley PhD student Pan Jiayi and two other researchers replicated DeepSeek R1-Zero in the CountDown game. Through reinforcement learning (RL), a 3B foundational language model can self-validate and search, with costs of less than $30 (approximately 217 yuan), yielding impressive results.
A team led by Assistant Professor He Junxian at Hong Kong University of Science and Technology (co-authored by Huang Yuzhen and Weihao Zeng) replicated the training of DeepSeek-R1-Zero and DeepSeek-R1 on a 7B model using only 8K samples, achieving strong results in complex mathematical reasoning.
Additionally, the team at HuggingFace, the world’s largest open-source platform, officially announced the replication of all pipelines of DeepSeek R1. Once completed, all training data, scripts, etc., will be fully open-sourced.
Some have stated on social forums that while other major companies are scrambling to purchase H100 chips, DeepSeek quietly accomplished something: with less than $6 million in extremely low costs and under the conditions of 2048 chips with performance far below H100 and Blackwell’s H800 chips, the DeepSeek team created an open-source AI model with performance comparable to OpenAI’s o1, while Anthropic and OpenAI’s training costs reached as high as $1 billion.
Looking ahead, the emergence of DeepSeek R1, through a new low-cost model of “extreme compression + efficient reinforcement training + significant simplification of AI inference computing power,” heralds the arrival of the AI ASIC era.
ASIC (Application-Specific Integrated Circuit) refers to chips designed for specific tasks. In the AI field, ASIC specifically refers to chips customized for deep learning, neural network training/inference, and other AI tasks.
The “AI ASIC era” signifies a shift in AI technology from “software-driven” to a stage of deep synergy between software and hardware. With its advantages in computing power, energy efficiency, and customization, ASIC is becoming the “new infrastructure” for AI, driving breakthroughs in large models, autonomous driving, smart terminals, and other fields. In the future, as technologies like Chiplet and advanced packaging mature, ASIC will further reduce costs and become the core engine for the democratization of AI.
Using DeepSeek to Look Ahead to the Year of the Snake in A-Shares
Analyzing the Chinese stock market trends for 2025 requires a comprehensive consideration of multiple factors, including macroeconomic conditions, policy direction, industry trends, and international environment. Below, we provide a rational analytical framework from various dimensions:
1. Macroeconomic Fundamentals
If GDP growth stabilizes around 5% in 2025 (the target for the 14th Five-Year Plan), and corporate profit expectations improve, the stock market will receive fundamental support.
Key variables: the strength of consumption recovery (residential savings conversion rate), export resilience (global industrial chain restructuring), and the soft landing process of the real estate industry.
2. Policy Environment
Fiscal and monetary policy: If a moderately loose tone continues (such as expansion of special bonds, structural interest rate cuts), ample liquidity may benefit stock market valuations.
Reform dividends: Policies such as the deepening of the registration system, acceleration of state-owned enterprise reforms, and cultivation of data factor markets may reshape market structure.
3. Industry Differentiation and Structural Opportunities
1. Policy-driven sectors
Technological independence: Fields such as semiconductors, AI, and quantum computing that are “choke-point” areas will receive long-term investment from the state, and hard tech companies may see valuation premiums.
Green transformation: The new energy (photovoltaics, energy storage), electric vehicle industry chain, and carbon neutrality derivative industries (carbon trading, energy-saving technologies) will continue to benefit.
Security themes: Sectors such as food security, energy security, and information technology application innovation (ITAI) will receive policy support.
2. Consumer and cyclical sectors
Consumption recovery: If residents’ income expectations improve, optional consumption sectors such as food and beverages, healthcare, and cultural tourism may see a recovery.
Traditional industries: The real estate chain (building materials, home appliances) needs to observe marginal policy changes; resource sectors (non-ferrous metals, chemicals) are influenced by the global commodity cycle.
4. Market Structure and Capital Flow
1. Foreign Capital Flow
After the MSCI expansion window closes, the pace of foreign capital inflow may slow, but if the RMB exchange rate stabilizes and Sino-U.S. relations improve, it may still bring incremental capital.
Attention should be paid to the global liquidity environment (whether the Federal Reserve’s interest rate hike cycle has ended).
2. Strengthening of Domestic Capital Dominance
The proportion of long-term funds such as pensions and insurance entering the market may increase, potentially reducing market volatility and promoting value investment styles.
The proportion of retail investors is decreasing, and institutional pricing power is strengthening, leading to a more rational market.
5. Risks and Challenges
1. External Uncertainties
Geopolitical conflicts, recurring global inflation, and risks of economic recession in Europe and the U.S. may trigger a phase of foreign capital withdrawal.
2. Internal Concerns
Local debt pressure, risks in small financial institutions, and long-term issues such as population aging, if not handled properly, may suppress market sentiment.
The surge in the number of listed companies under the registration system may exacerbate individual stock differentiation, and the risks of “speculating on small and poor stocks” need to be heeded.
6. Scenario Analysis
Optimistic scenario: Moderate economic recovery + continuous policy support + easing external environment → The Shanghai Composite Index may challenge 4000 points.
Neutral scenario: Structural growth + localized risk release → The market may maintain a range-bound fluctuation (3000-3500 points).
Pessimistic scenario: Black swan events (such as global recession, geopolitical crises) → Short-term deep corrections may occur, but policy support is expected.
7. Investor Strategy Recommendations
1. Long-term allocation: Focus on policy dividends and industry trends (such as high-end manufacturing, digital economy), avoiding high-valuation bubble sectors.
2. Tactical operations: Pay attention to turning points in economic data (such as PMI, social financing) and signals released by important meetings (Central Economic Work Conference).
3. Risk hedging: Appropriately allocate safe-haven assets such as gold and government bonds to diversify single-market risks.
Summary
The Chinese stock market in 2025 is likely to present a “structural market,” with policy-driven and performance-stable industries expected to outperform the broader market, but caution is needed regarding the volatility brought about by the resonance of internal and external risks. Investors should downplay index fluctuations, delve into niche sectors, and maintain flexible positions to respond to uncertainties.
Editor: Gui Yanmin
Editor: Gui Yanmin
Proofreader: Su Huanwen
Millions of Users Are Watching
Just In! Trump Takes Major Action!
Good News! The CSRC Just Announced! Eight Departments Make Major Deployments! Trump Sets the Tone, Involving Tariffs on China! The Top Ten News Impacting the Market This Week
Suddenly, Collective Movement!
U.S., Major Changes!
This Weekend, A Surge! Musk, Major Plans!
Report Illegal and Bad Information: 0755-83514034
Email: [email protected]
