Chen Guo believes that the key short-term tracking indicators may be the traffic of DeepSeek; in the medium term, if there is an explosive growth in edge AI and new application product traffic, then the market performance can reference Smart Phone and electric vehicles, mapping the changes in industrial prosperity from overseas to domestic.
Key Points
DeepSeek's low cost, open-source characteristics, and strong inference capabilities are expected to greatly promote the development of domestic AI applications and edge AI. In the Industry Chain, the sectors of computers/media (AI applications), robotics/electronics/self-driving (edge AI), and domestic computing power chains are likely to benefit significantly.
How to grasp the rhythm of the development? Before the fundamentals are verified, short-term key tracking indicators might be DeepSeek's traffic; in the medium term, if edge AI and new application products lead to an explosive growth in traffic, or if corporate profit margins improve (with preliminary verification in the first quarterly report), and if domestic AI investment significantly increases, then market performance may reference smart phones and electric vehicles, as it reflects the transition of industry prosperity from overseas to domestic.
Summary
The breakthroughs of DeepSeek can be likened to the Chat GPT moment in the USA; its low cost, open-source characteristics, and strong inference capabilities are expected to greatly promote the development of domestic AI applications and edge AI. DeepSeek has confirmed that the overseas computing power blockade against China is not insurmountable, but can be partially compensated for through efficient training methods.
Additionally, its high cost-performance characteristics can rebalance the cost-effectiveness between corporate AI investments and outputs. So far, multiple cloud vendors and various industry entities have integrated or fused with DeepSeek; at the same time, its strong inference capabilities are expected to accelerate the development of vertical model applications, resulting in an outburst of AI applications and edge AI. In the Industry Chain, the sectors of computers/media (AI applications), robotics/electronics/self-driving (edge AI), and domestic computing power chains are likely to benefit significantly.
Since the beginning of the year, under the promotion of DeepSeek, the TMT market has performed brightly. From a strategic perspective, how to grasp the subsequent rhythm of development? Based on historical experiences from smart phones, electric vehicles, as well as the recent domestic and international AI markets, we provide the following references:
1) Before the fundamentals are verified, refer to the overseas mapping trends of AI in 23H1, as well as the trending applications driven by Kimi, Doubao, etc. Short-term key tracking indicators may be the traffic of DeepSeek. The access volume data of Chat GPT in 2023 shows a significant positive correlation with the domestic technology trends. As the access volume of Chat GPT grew and hit a bottleneck in May, combined with the arrival of the earnings report season, the domestic technology sector lacks performance verification, showing signs of overheating, with the computer and media sectors experiencing peak adjustments.
Similarly, the rapid growth in traffic of the trending application Kimi in February 2024 was also the main driver of the media sector's explosion at that time. Trade indicators such as the TMT Index's transaction volume ratio touching the overheating range may exacerbate market fluctuations, but cannot determine the height of the market.
2) From a mid-term perspective, if the traffic of DeepSeek successfully transmits to related endpoints, such as creating a traffic explosion for new products, or improving corporate profit margins (basic fundamental verification will focus on the quarterly report at the earliest), along with a significant increase in AI investment domestically, then the market performance can reference Smart Phones and Electric Vehicles, transitioning from overseas mapping to domestic industry prosperity.
Reviewing the markets for Smart Phones and Electric Vehicles, the first phase was primarily overseas mapping and thematic speculation after the introduction of overseas blockbuster products. However, both have formed a second wave of investment in prosperity domestically and maintained a long-lasting duration, driven by the accelerated localization of overseas leading supply chains. More importantly, after the overseas business model took root, domestic blockbusters emerged, with penetration rates accelerating and driving substantial increases in industry chain capital expenditures.
At this point, the performance realization of the sectors is strengthening, making it easier for them to develop an independent market in a bottom-up manner compared to the overall market. The previous domestic AI investment cycle generally lagged behind overseas, with weak performance realization in the industry chain and the market driven mainly by themes; however, as DeepSeek and Doubao successively break through, domestic AI investments are increasing, and the prosperity of the domestic AI industry chain is underway.
DeepSeek promotes the acceleration of domestic AI industry development.
Fully welcoming the AI+ new era, various overseas business models have been established.
Chat GPT leads the explosion of the AI industry wave, as overseas technology giants have continuously increased their investments in AI-related areas since 2023, with overseas AI foundational large model technologies being iteratively updated. The AI industry wave represented by ChatGPT erupted at the end of 2022, and in the past two years, overseas large AI models have continued to iterate with breakthrough advancements in multi-modal, reasoning, and text-to-video AI large models.
Overseas technology leaders continue to increase their investments in the AI field, with capital expenditures entering an upward cycle since Q2 of 2023, and in Q3 of 2024, the four major technology companies ($Microsoft (MSFT.US)$/$amazon(AMZN.US)$/ $Alphabet-A(GOOGL.US)$/$Meta Platforms (META.US)$Total capital expenditure increased to 58.86 billion USD, year-on-year +59.1%; as of Q4 2024, Microsoft's and META's latest Earnings Reports show that capital expenditures still exceed expectations, with Microsoft expected to invest about 80 billion USD in constructing AI datacenters in fiscal year 2025, and META estimating its 2025 capital expenditures at 60-65 billion USD, a year-on-year increase of 54.6%-67.5%.
Overseas AI applications have exploded in performance, with various ToB business models validated and connected. With the improvement and expansion of foundational AI infrastructure, overseas application terminals are accelerating their outbreak, especially for various ToB business models which have connected and achieved performance validation. As of Q3 2023,$Applovin(APP.US)$、$Palantir(PLTR.US)$、$Snowflake(SNOW.US)$、$ Duolingo (DUOL.US) $、$Shopify(SHOP.US)$As AI application companies outperform expectations, the sectors involved include advertising, data governance, education, e-commerce, and more, with generally strong forward guidance.
Taking AppLovin as an example, its AI advertising engine model AXON 2.0 effectively improves advertising matching efficiency, achieving revenue of 1.2 billion USD in the latest quarter, a year-on-year increase of 39%; net income of 0.43 billion USD, a year-on-year increase of 300%. On the consumer side, ChatGPT's traffic data continues to show an upward trend.

Deepseek drives a confidence reassessment in the domestic AI industry.
DeepSeek's breakthrough progress can be compared to the Chat GPT moment in the USA. DeepSeek released the open-source model DeepSeek-R1, achieving significant breakthroughs in technology, with its mathematical/programming/reasoning capabilities comparable to OpenAI's o1 model, but the pre-training cost is less than one-tenth of that of OpenAI's GPT-4o model, and the API calling cost is only 2% of it. Data from Appfigures shows that on January 26, DeepSeek-R1 topped the App Store download charts and has maintained a leading position worldwide since, reaching the top in 160 countries or regions by February 4.
DeepSeek's low-cost, open-source, and high-performance characteristics are expected to significantly accelerate the development of edge AI and AI applications; in the industry chain, sectors such as computer/media (AI applications), robotics/electronics/self-driving (edge AI), and domestic computing chains are likely to benefit greatly.
Firstly, DeepSeek has confirmed that the overseas restrictions on China’s computing power are not insurmountable, but can be partially compensated for through efficient training methods.
Secondly, its cost-effective characteristics can rebalance the cost performance between AI investment and output. As of now, Huawei Cloud, $ tencent (00700.HK)$Cloud, $alibaba-W(09988.HK)$Cloud,$baidu (bidu.us)$Smart Cloud and other cloud vendors, Autel Intelligent Technology Corp.,Ltd., Jiangsu Hengrui Pharmaceuticals,$Geely(00175.HK)$Various industry entities, including China Securities Co.,Ltd., have integrated or merged with DeepSeek; its powerful reasoning ability is expected to accelerate the development of vertical model applications, accelerating the explosion of AI applications and edge AI.

In addition to DeepSeek, models like Doubao have also emerged. Domestic large models have made substantial breakthroughs over the past two years, with the overall capabilities of the first-tier large models now nearing GPT-4, and some models' Chinese capabilities are nearly on par with GPT-4. According to SuperCLUE tracking, the gap ratio between the top domestic and international models has narrowed from 30.12% in May 2023 to 1.29% by August 2024.
On the application front, Doubao, Kimi and others have started to shine. In February 2024, the visit count for the smart assistant Kimi doubled month-on-month, and on March 18, Moon's Dark Side announced that Kimi's smart assistant's ability to handle long context windows increased to 2 million words, which is the longest context input supported by large models that can be productized in the global market.
In November 2024, the AI smart application Doubao under ByteDance saw a surge in downloads, with its monthly active users reaching 59.98 million, second only to ChatGPT and ranking second globally; according to Quantum Bit data, as of the end of November, Doubao APP maintained an average daily new user download of 0.8 million, also ranking second globally.

The rhythm of the current technology market performance.
Since the beginning of the year, DeepSeek has driven impressive performance in the TMT sector. From a strategic perspective, how will the subsequent performance rhythm be grasped? Based on the historical context of Smart Phones, Electric Vehicles, and insights from the recent AI market, we provide the following reference:
From a vertical perspective, before the fundamentals are verified, reference can be made to the overseas mapping market of AI in the first half of 2023, as well as the popular application market driven by Kimi, Doubao and others; the key short-term tracking indicator may be the visit volume of DeepSeek.
Looking back at the explosive popularity of ChatGPT around the Spring Festival of 2023 and its impact on the domestic AI market, before validating the 24H1 earnings report data, the access data of ChatGPT showed a clear positive correlation with the domestic Technology market. As the growth of ChatGPT access volume hit a bottleneck in May, alongside the onset of the earnings season, the domestic Technology sector lacked performance validation, leading to signs of overheating in the market. The Computer and Media sectors gradually peaked and corrected. Similarly, in February 2024, the rapid growth of the popular application Kimi’s traffic was also a major driver for the Media sector's explosion and its independent market performance.

Trade indicators, such as the TMT Index transaction volume ratio reaching overheated ranges, may exacerbate market volatility but cannot determine the extent of the market rise. During the Internet+ market from 2013 to 2015, the TMT Index transaction volume ratio reached nearly 30% in the overheated range from July to September 2013, yet both the absolute return of the index and the relative returns (relative to Wind all-A) remained within the mid-stage of the market.
In September 2019, the TMT Index transaction volume ratio reached about 40% at a high point, but the market did not conclude until early 2020. Similarly, in the AI market at the beginning of 2023, the TMT Index transaction volume ratio hit a new high of over 40% in March, but this did not lead to the end of the market.

In the medium term, if DeepSeek's traffic explosion successfully transfers to relevant end-side applications, such as resulting in a surge in traffic for related products or an increase in gross margin (the earliest focus on fundamental validation would be the quarterly report), and successfully drives significant increases in domestic AI investments, then the market performance may reference the Smart Phone and Electric Vehicles sectors, reflecting a shift in domestic industry prosperity from overseas.
Similar to Smart Phones and Electric Vehicles, AI has also been a significant technological revolution achieved first by overseas technology giants, which realized technological breakthroughs and created popular products, followed by domestic following and successful commercial implementation. The development history of both may provide insights into the trends and market performance of the domestic AI industry.
Looking back at the historical performance of domestic Smart Phones and Electric Vehicles, it can be divided into two major phases:
1) The first phase is the overseas mapping and thematic speculation after the launch of overseas best-selling products. The core symbols that rose during this period were companies related to the overseas technology leader's industry chain. However, due to limited market share and sales volume, the performance realization feedback on the overall sector was actually weak. The market, apart from event catalysis, is also susceptible to macro fundamentals and market sentiment, exhibiting obvious thematic characteristics.
2) However, both sectors have formed a second wave of prosperous investment domestically and maintained a longer sustainability. On one hand, this was driven by the accelerated localization of the overseas leader's supply chain. More importantly, after the overseas business model was implemented, about 1-3 years later, domestic best-sellers began to emerge, penetration rates accelerated, and capital expenditure in the industry chain significantly increased. At this point, the related industry chain symbols had greatly expanded, performance realization in the sector strengthened, entering a stage of prosperous investment, and during this period, it was comparatively easier to develop independent market performance from the bottom up.

Smart Phone: 1) Mapping period: 2009-2010. In October 2009, the iPhone 3G and iPhone 3GS entered China, and in June 2010, Apple's revolutionary product, the iPhone 4, was officially released, leading the global smart phone industry craze and driving the performance of the electronic sector, particularly with significant price increases for suppliers like Shenzhen Laibao Hi-Tech, Goertek Inc., and Shenzhen Desay Battery Technology.
However, at that time, the share of domestic consumer electronics industry participating in Apple's supply chain was not high. Among the major core suppliers announced by Apple in 2012, only 8 mainland companies made the list, and the overall performance of the sector was limited, more reflecting overseas mapping logic. Subsequently, as monetary policy tightened and continued to strengthen from November 2010, suppressing the valuations of growth stocks, the structural market also came to an end.
2) Domestic industry trend initiation: 2013-2015. Domestic smart phone brands such as Xiaomi and OV gradually emerged, while Apple's core suppliers continued to expand, leading to a notable improvement in the performance of the consumer electronics sector. As the penetration rate of smart phones further increased, various popular applications emerged under the flow dividend, including online videos and mobile games, accelerating the explosion of the mobile Internet software sector.

Electric Vehicles: 1) Mapping period: 19Q4-20H1.$Tesla (TSLA.US)$The Shanghai factory was officially established, and Model 3 sales boomed. Although at that time, the actual supply ratio of most Tesla Concept stocks in the A-share market was still relatively low, and the domestic new energy vehicle market experienced continued negative growth year-on-year, the Wind Tesla Index still recorded impressive market performance under the mapping logic.
2) Domestic industry trend initiation: 20H2-2021. On one hand, the localization rate of components at Tesla's Shanghai factory continued to accelerate, rising from 30% at the beginning of 2020 to 95% in 2022, with the related industry chain continuing to expand;
On the other hand, Tesla played a catfish effect, promoting the rapid development of China's new energy vehicle industry, catalyzed by policies reinforcing support, upgrades in technology enhancing cost-effectiveness, and the launch of new products on the supply side.$BYD COMPANY(01211.HK)$、$NIO (NIO.US)$、$Xpeng Motors (XPEV.US)$、$Li Auto (LI.US)$As domestic Electric Vehicles brands have successively achieved a sales explosion since the second half of 2020, the penetration rate of domestic Electric Vehicles has rapidly increased, which simultaneously drives the quantity and price increase of the domestic lithium battery Industry Chain.

Returning to the current AI industry cycle, due to the slightly lagging progress of the domestic AI industry compared to overseas, the initial AI market performance (23H1) was mainly influenced by overseas industry trends. Benefiting from the global AI industry trend and the improvement of the domestic macro environment, the domestic AI market began to activate at the beginning of 2023, heavily reflecting the overseas logic, with theme attributes stronger than prosperity attributes.
Around the Spring Festival in 2023, ChatGPT became popular again, combined with an improving domestic macroeconomic environment and a basic enhancement in risk preferences, the domestic AI sector started to rise from February. Relevant targets in the overseas tech giants' supply chain led the way, followed by funds focusing on computing power, phenomenal application expectations, and large model directions.
From March to April, domestic generative large models, including Wenxin Yiyan, Riri Xin, and Tongyi Qianwen, achieved breakthroughs one after another. However, the market gradually became aware of the barriers and insufficient scarcity of large models, leading to a preliminary correction in the related concept, with the market increasingly focusing on the certainty of computing power + data direction. The first phase of the AI market operated primarily on the logic of overseas reflection, during which events such as the release of GPT-4 (March 14),$nvidia (NVDA.US)$the Q1 2023 earnings exceeded expectations significantly (May 25) became important catalysts for the market.

However, the pure logic of overseas reflection has clearly weakened since the second half of 2023. Since the second half of 2023, the AI sector has begun to differentiate internally, with the AI computing power export chain showing sustainability supported by performance, and key targets in the industry chain are expected to reach new highs in 2024; while the computer and media sectors have returned to their starting point in 2024.
From a macro perspective, the weak economic recovery in China has been confirmed in the second half of 2023, compounded by the continuous rise of U.S. debt, overall market risk preference has been disrupted; on a micro level, the Wind AI Index rose by as much as 45.4% from the high in February 2023 to June 2023, surpassing Microsoft (43.4%), and the domestic AI market speculation entered an obvious bubble stage. On June 20, 2023, Kunlun Tech shareholders significantly reduced their holdings, creating a severe impact on sector sentiment,
and after that, several AI-related companies announced shareholdings reductions, coupled with low fulfillment of performance forecasts in the Q2 2023 sector, domestic AI market speculation came to an end, leading to differentiation.
Since the second half of 2023, benefiting from the upward spending cycle of overseas tech giants on computing assets and the truly performance-delivering optical modules, core symbols are set to reach new highs in 2024; however, the overall valuation in the computer and media sectors has dropped, and although catalysts like the release of Sora and the domestic Kimi outbreak may lead to a rebound in 2024, the index still set a new low since 2023 in August and September, with relative market excess returns lagging in the first three quarters.

In the earlier domestic AI investment cycle, there has been an overall lag compared to overseas, with weak performance fulfillment in the Industry Chain and lack of sustained market trends. However, as DeepSeek and Doubao continue to make breakthroughs, they are expected to drive domestic investments in AI and boost the local AI Industry Chain. The construction of computing power in China accelerated in the second half of 2023, with the capital expenditure cycle of BAT warming up (36.23 billion in Q3 2024, up 117% year-on-year), yet the scale still has a significant gap compared to the four overseas giants, causing the domestic AI investment cycle to lag behind overseas.
By the end of 2024, with ByteDance's AI application exceeding expectations, its intent to invest further has become clearer. According to a report by the Financial Times in the UK, ByteDance plans to invest over $12.3 billion in AI infrastructure in 2025 (89.5 billion yuan). On December 26, Jiemian News learned that XIAOMI-W is working on building its own GPU cluster for AI large model investments. Since the outbreak of DeepSeek, many industries such as technology, machinery, autos, finance, and pharmaceuticals have expressed their intent to integrate or connect with large models, officially entering an accelerated phase of the AI+ trend.

In addition to the software industry, which is expected to see a traffic explosion and improvement in profit margins, the strong traffic and market scale advantage on the domestic C-end makes the commercialization of edge AI evidently advantageous. Currently, giants like ByteDance and XIAOMI-W have begun to layout edge AI aggressively, seizing the entrance of AI Agents. Besides mobile phones, PCs, glasses, and headphones, the potential base for edge AI is enormous, with home appliances, robots, smart cars, educational office equipment, and toys all set to benefit from the trend.
From January 7 to 10, CES 2025 will be held in Las Vegas, USA, focusing on the integration of AI with Consumer Electronics, Autos, Smart Home, and other fields. New products for AI glasses, AIPC, smart cars, robots, and AI toys will be displayed in abundance, while chip manufacturers will intensively release a series of new products; PC manufacturers are actively pursuing product innovation; traditional car companies in the auto sector will release multiple intelligent new products, showcasing the latest technological advancements and products from advanced driver assistance systems (ADAS) to fully self-driving technology.
From the financial report data of Q3 2024, unlike overseas, the net profit margins of domestic AI application-related companies have generally not achieved significant improvement, with gaming and marketing agencies seeing changes in gross margins of -0.3 percentage points and -0.5 percentage points respectively from Q2 to Q3. However, in aspects of edge AI, there have already been landed products in fields like AI smartphones, smart driving, and AI toys. From the performance of Q3 2024, the Consumer Electronics and smart driving Industry Chain has achieved relatively high prosperity, with the profits attributable to shareholders rising by +16.5% and +29.5% year-on-year respectively.

AI + Consumer Electronics: Continuous innovation in software and hardware, Apple has launched the AI-enabled Apple Intelligence, deeply integrated into its core products, while Android manufacturers are actively exploring edge applications of large models to optimize user experience. The enhanced user experience brought by AI smartphones is expected to trigger a wave of device upgrades. According to predictions from Canalys, the global penetration rate of AI smartphones will reach 16% in 2024 and 54% in 2028.
The launch of AI PCs is also expected to drive a surge in demand for PCs. In 2024, Lenovo plans to release multiple AI PC products, with Canalys predicting that the global shipment of AI PCs will account for 18% of total PC shipments in 2024 and will reach 40% in 2025.
AI + Self-Driving: Since 2024, with the reduction of costs in software and hardware, high-level self-driving has gradually become a standard for products in the 200,000 yuan range (such as P7+) and is sinking into the price range below 200,000 yuan. The year 2025 is expected to be a year for the scaled popularization of high-level self-driving, with high-speed NOA and high-level self-driving expected to become standards for products in the 100,000-200,000 yuan and above 200,000 yuan ranges, respectively. For instance, BYD plans to lower high-level self-driving solutions to models priced at 100,000 yuan in 2025. Referring to the experience of electric vehicles in 2020, the improvement in product strength and cost performance will significantly promote the acceleration of industry penetration.
AI Robots: 1) The construction of general foundational models for robots through AI large models, and advancements in reasoning chips are accelerating the response efficiency and accuracy of robots. Currently, foreign manufacturers are gradually releasing products, and mass production is imminent. At the Q4 2024 Tesla Earnings Conference, Musk has again raised the mass production guidance, stating that the internal goal is to produce 100,000 robots by 2025; plans to launch the second-generation version for mass production in 2026, guiding a monthly production volume of 10,000 units; as a next step, aiming to reach 100,000 units per month by 2027, exceeding market expectations.
2) Domestic policies strongly support industry development. In November 2023, the Ministry of Industry and Information Technology issued the "Guiding Opinions on the Innovative Development of Humanoid Robots," guiding the mass production of humanoid robot complete products in China by 2025, and significantly improving the completeness of the domestic industry chain by 2027. Various localities are gradually rolling out related policy arrangements.
In November 2024, Huawei (Shenzhen) announced the official operation of the Global Embodied Intelligence Industry Innovation Center.$Xpeng Motors-W(09868.HK)$Released the AI robot Iron.$XIAOMI-W(01810.HK)$、$UBTECH (09880.HK)$Companies like Yushu and Zhuimi are continuously iterating their robot products, and the domestic humanoid robot Industry Chain is on a fast track to development.

Author of this article: Chen Guo S1440521120006, Zheng Jiawen S1440523010001, Source: CSC Strategy Chen Guo Team, Original title: "Reviewing Previous Growth Bulls, How Will This Round of AI Market Play Out?"
Editor/ping