Looking ahead to the electronics market in 2025, the application of AI technology in consumer electronics and smart vehicles will deepen further, driving the semiconductor market to maintain an annual growth rate of over 10%, with particularly strong demand for AI and automotive chips. Concurrently, consumer electronics are trending towards sustainability, with increased promotion of eco-friendly products. The integrated circuit industry is benefiting significantly from the push of AI, 5G, and electric vehicle markets, leading to notable technological advancements. Additionally, emerging technologies like Generative AI and service robots will become new highlights in the market.
This report provides trend analysis and market outlooks on hot topics and fields including humanoid robots, smart wearables, autonomous driving, green computing, smartphones, generative AI, domestic substitution, 3nm automotive chips, memory, and digital simulation.
Trend 1: Continuous Breakthroughs in Humanoid Robots
Humanoid robots, built upon multidisciplinary foundations and integrating advanced technologies like artificial intelligence, high-end manufacturing, and new materials, are poised to become disruptive products following computers, smartphones, and new energy vehicles. They serve as a crucial indicator of a nation's high-tech strength and development level.
The rapid development of the humanoid robot market stems from two key factors. Firstly, the iteration of large AI models and capabilities for autonomous learning and optimization enable robots to better understand human instructions, engage in complex dialogues, and even predict and fulfill user needs, significantly enhancing service quality and interaction experiences. Secondly, it results from the maturity and refinement of the upstream, midstream, and downstream industry chains, coupled with government policy support and guidance.
Capital markets also favor the humanoid robot sector. As of November 2024, China's humanoid robot field recorded 49 financing events, primarily concentrated in Beijing, Shenzhen, and the Yangtze River Delta region, with a total financing exceeding 5 billion RMB. Among the 33 humanoid robot companies that secured funding, 26 (nearly 80%) were founded between 2022 and 2024.
Financing activity is equally vigorous internationally. Figure AI secured $675 million (approx. 4.73 billion RMB) in February this year, with investors including Microsoft, NVIDIA, OpenAI, and Amazon founder Jeff Bezos. In late October 2024, Agility Robotics completed a $150 million (approx. 1.05 billion RMB) Series C funding round, achieving unicorn status.
Particularly noteworthy is the breakthrough in key technologies like AI, machine learning, and computer vision, presenting significant development opportunities for a large cohort of humanoid robot companies highlighting "Embodied Intelligence." It is widely believed that research guided by this theory enables humanoid robots to better adapt to complex and changing environments and tasks, possessing capabilities for autonomous learning, perception, and decision-making.
Although humanoid robots can be categorized by application scenarios, surveys indicate that most Chinese companies believe manufacturing, especially automotive manufacturing, will be the first to achieve true commercialization of humanoid robots, with quality inspection and component assembly identified as clear demand scenarios. In contrast, the highly anticipated home service application ranks lower.
Trend 2: Smart Wearables Regain Momentum, Market Structure Evolves
With rapid technological advancements, smart wearable devices are increasingly becoming indispensable in our daily lives. From smartwatches to fitness trackers, VR headsets to AR glasses, hearables to smart rings, the variety and functionality of wearables continue to expand, offering consumers unprecedented convenience and experiences.
After fluctuating development in recent years, the smart wearable market is expected by International Electronic Business to reach a new peak in 2025, with global shipments projected to hit 800 million units.
The primary growth driver comes from consumers' increasing focus on health and fitness. Smartwatches and fitness trackers help users better manage their health by monitoring key metrics like heart rate, sleep quality, steps, and calorie consumption. Furthermore, fueled by the proliferation of 5G and IoT development, wearables are becoming more interconnected and intelligent, providing highly personalized services and experiences across healthcare, sports monitoring, communication, and entertainment.
Additionally, the integration of AI and generative AI promises more precise data analysis and predictions for wearables, such as monitoring key health indicators like blood pressure. We anticipate seeing more innovative applications in 2025, like biometric authentication or personalized fitness plans generated via machine learning algorithms.
Two segments are particularly noteworthy for 2025. The first is children's smartwatches, which integrate critical functions like communication, location tracking, emergency calling, and health monitoring, meeting parents' growing concerns about child safety and health, indicating rising demand. The second is smart wearables for the elderly. Amidst global population aging trends, these devices can help seniors maintain independence and access timely medical assistance through features like health monitoring, fall detection, emergency response, and daily activity assistance, also showing strong growth potential.
As the market matures, the competitive landscape is shifting. Traditional consumer electronics giants like Apple, Samsung, and*continue to lead, while an increasing number of startups and emerging brands are gaining prominence, challenging incumbents with innovative products and solutions.
Trend 3: Autonomous Driving Market Presents a Dichotomy
In 2024, enthusiasm for autonomous driving cooled somewhat overseas, partly due to the massive R&D investments required before large-scale commercialization and persistent profitability challenges. Examples include Samsung Electronics hitting the brakes on autonomous car research and Apple halting its "Apple Car" project in early 2024. These moves reflect the technical difficulties and the challenge of achieving significant returns on investment.
However, the picture in China is different. Chinese manufacturers continue to invest heavily in autonomous driving, demonstrating confidence in the technology.
Policymakers in China actively promote autonomous driving development, introducing policies to support testing, demonstration applications, and commercial operations. Multiple cities have launched intelligent connected vehicle (ICV) road test demonstrations, opened test roads, completed road intelligence upgrades, and installed roadside units (RSUs), creating a favorable policy environment. Notably, the launch of robotaxi services like "Luobo Kuaipao" (萝卜快跑) in cities like Wuhan and Guangzhou has spurred investment enthusiasm.
Technologically, the rise of end-to-end autonomous driving systems has concentrated focus. Unlike traditional modular architectures, end-to-end systems use a unified deep learning model to map sensor inputs directly to vehicle control outputs, reducing intermediate steps and improving real-time performance and accuracy.
At the enterprise level, Chinese companies like Baidu Apollo, NIO, XPeng, Huawei, and Xiaomi are actively deploying autonomous driving technology, accelerating the digitalization and intelligence of the automotive industry through innovation and unique product design.
Looking ahead to 2025, the autonomous driving market is expected to achieve positive growth. Global autonomous vehicle shipments are forecast to reach tens of millions, while sales of L2+ and above intelligent vehicles in China are projected to exceed 10 million units, with a penetration rate of 50%. L4 vehicles will begin entering the market, and China is poised to become the world's largest autonomous driving market.
Notably, the potential for rapid growth extends beyond autonomous driving tech developers to upstream and downstream players in the supply chain. Driven by Chinese autonomous vehicles going global, Chinese companies with highly competitive products in areas like in-vehicle touchscreens and power transmission stand to gain significant new development benefits.
Trend 4: Green Computing Becomes More Urgent
Schneider Electric's 2023 Energy Management Research Center whitepaper states that AI power consumption reached 4.5 GW in 2023, with a projected CAGR of 25%-33% until 2028 – 2-3 times faster than the overall data center growth rate. Therefore, AI power demand is expected to reach 14-18.7 GW by 2028, accounting for 15%-20% of total data center power. This doesn't even account for AI power dispersed across more edge and endpoint devices.
This level of consumption is equivalent to the average annual power usage of some countries. Considering the energy consumption across the entire AI chip design and manufacturing chain, and reports of Sam Altman potentially investing trillions in AI chip infrastructure, the total energy footprint of AI technology could be staggering.
If AI power consumption exceeds a certain proportion of the power infrastructure's supply capacity, public livelihood will be affected. Hence, aligned with the broader energy-saving and carbon-reduction trend, "Green Computing" and "Green Data Centers" are now hot topics. Solutions are being proposed at several levels.
Even beyond power devices and chips, for digital chips, especially AI chips, joint optimization of chip design/manufacturing and software algorithms is crucial to improve efficiency and reduce the power consumption per unit of compute. Furthermore, for large-scale cluster computing across chips, boards, and nodes, interconnect and memory are key bottlenecks; advancements in technologies like in-memory computing and optical interconnect are essential developments under this trend.
Concurrently, some chip companies are promoting system-level solutions like cold plate and immersion liquid cooling. At events like WAIC, many AI chip companies showcased liquid-cooled POD cabinet solutions. For instance, Enflame (燧原科技) highlighted achieving PUE ≤ 1.15 in lab data centers through excellent performance linearity in thousand-card deployments combined with cluster-wide liquid cooling.
It's understandable that chip foundries, EDA/IP suppliers, and chip design firms are all discussing achieving higher efficiency from a "system" perspective – encompassing silicon, devices, units, chips, packaging, systems, software, and applications – to realize green computing.
Another layer involves investment in energy infrastructure. Foreign media previously revealed OpenAI's "US AI Infrastructure Blueprint" includes a grand vision for nuclear power. Jensen Huang has also mentioned in interviews that data centers need renewable energy, and nuclear is a great option. Oklo, where Sam Altman is Chairman, plans to deploy its first SMR (Small Modular Reactor) by 2027. Google has signed an agreement with nuclear startup Kairos Power to power data centers with SMRs...
Many AI tech giants are making capital investments in nuclear energy. This underscores the anticipated future demand for green energy and green computing.
Trend 5: Smartphones Fully Embrace Generative AI, Ecosystem Consolidates
Starting with iOS 18, Apple is rolling out generative AI features for iPhone in select regions, including voice-to-text for Notes, AI photo editing, email writing tools, key summaries for emails and calls, and natural language-based photo/video search...
Apple is not only investing heavily, using Google TPUs to train both server-side and on-device AI models, but also specifically building an AI data center based on its own chips for on-device AI applications – creating "Private Cloud Compute" clusters in the cloud while performing hybrid AI inference both locally and in the cloud.
Regardless of opinions on the value of "AI phones," this has become a serious business direction. Apple's entry into the "AI phone" concept essentially marks the shift of mobile AP SoC competition into the generative AI era. The battlefield for mobile AP SoCs, operating systems, and OEMs has expanded to include competition in AI models, AI development ecosystems, and AI applications.
For users, the shift is evident: new iPhones now start with 8GB RAM, and new MacBooks typically have at least 16GB RAM – largely because Large Language Models (LLMs) and other large models required for generative AI demand significant memory space. This indicates that the adoption of generative AI on smartphones will advance rapidly over the next 1-2 years, creating new opportunities for mobile chips and various peripheral components.
Further highlighting this trend is MediaTek's Dimensity 9400 chip, launched in October 2024, which aggressively incorporates more specialized AI acceleration units within its NPU to support multimodal input, on-device LoRa training, and even on-device short video generation. MediaTek is also actively collaborating with smartphone OEMs, large model providers, OS vendors, and ISVs to build its own AI ecosystem.
As the smartphone generative AI battle intensifies in 2025-2026, the AI ecosystem on phones is expected to gradually consolidate from a fragmented state. AI technology stacks, features, and functionalities will also tend towards standardization.
Trend 6: Generative AI Moves to the Edge
NVIDIA's Research team, focused on cutting-edge technologies, has long pursued "Efficient AI" research. Techniques like model pruning, sparsification, and AWQ weight quantization aim to run large models on edge or endpoint devices – not just AI PCs, but also platforms like Jetson Nano.
Research like Efficient AI forms the foundation for bringing generative AI to endpoint devices like PCs and phones. But this is not the whole story of "Generative AI at the Edge."
At the 2024 WAIC (World AI Conference), International Electronic Business observed numerous small AI chips claiming the ability to run LLMs in their specs. For example, Axera's (爱芯元智) AX630C, with INT8 compute of 3.2 TOPS and typical power under 1.5W, can run the Qwen2 0.5B model at speeds around 10 tokens/second. While Axera stated specific application directions are still uncertain, the possibilities are vast.
Throughout 2024, nearly all AI chip companies interviewed unanimously agreed that AI, including generative AI, will move to the edge and into everything. Given the relatively large size of models and the constraints of storage, I/O, and compute power on edge devices, deep collaboration between AI chip architecture design and software algorithm optimization will become critical.
VeriSilicon (芯原) mentioned at its 2024 AI Tech Symposium that its NPU and GPU IPs, initially targeting AI vision, speech, and graphics, now cover natural language and span AR/VR, autonomous driving, PCs, smartphones, wearables, and robots. VeriSilicon stated its next move is to "Go Transformer" – the structural foundation of contemporary large models.
Consequently, many edge and endpoint AI chips are incorporating "Transformer engines." VeriSilicon also mentioned optimizing its NPU IP for Transformer acceleration, including mixed-precision data format support and implementing 4-bit and 8-bit weight quantization to drastically reduce bandwidth consumption. These efforts will help bring generative AI to the edge – not just to cars, PCs, phones, and robots, but potentially also to more compute-constrained embedded applications.
Trend 7: Increasing Domestic Substitution in China
The proportion of domestically produced chips substituting for imports in China has been rising steadily. In 2023, China's chip production surged to 351.4 billion units, a 6.9% YoY increase, achieving a self-sufficiency rate of 23%. Production jumped further to 135.4 billion units in the first four months of 2024, showing strong momentum. Benefiting from the trend of edge AI deployment, demand is expected to grow significantly in sectors like domestic CIS, memory interface chips, niche memory, analog chips, and cloud/edge compute chips.
Trade friction causing global supply chain fragmentation, coupled with the need for supply stability driving domestic sourcing, is expected to fuel continued growth in demand for domestic foundries, equipment, and packaging/testing facilities over the next 2-3 years. Currently, across the eight major stages of chip manufacturing, China's domestic semiconductor equipment industry has made significant breakthroughs in areas like thermal processing, thin-film deposition, etching, and cleaning, with client progress reaching 14nm, and etching machines achieving 5nm capability.
In the first half of 2024, mainland China's semiconductor equipment procurement reached $25 billion, with domestic equipment accounting for about $8 billion. The substitution rate for domestic equipment in processes like resist stripping, cleaning, etching, and packaging has surpassed 30%, reflecting both massive demand and the accelerating pace of domestic substitution.
Advanced packaging technology is crucial globally and for China, addressing challenges from Moore's Law slowdown and supply chain issues caused by export controls. Chinese companies have solid capabilities here. Despite limitations in front-end wafer manufacturing, advanced packaging holds broad growth prospects and is a key driver for China's market development.
Global expansion ("going out") and intelligence are driving sustained growth in China's consumer terminal industry. Automotive electronics is a key focus, with overseas expansion being a top priority for companies in 2025. The rapid development of China's domestic automotive industry enhances the competitiveness of its supply chain companies internationally, presenting new opportunities for leading firms.
However, overall, from a global semiconductor supply chain regional share perspective, the US, Europe, and others hold the majority. China only holds a certain share in mid-stream wafer manufacturing and packaging/testing. Self-sufficiency remains relatively low, with reliance on others ("受制于人") persisting in upstream areas like EDA/IP, equipment, high-end process nodes, high-performance computing (HPC), and some key materials.
Trend 8: 3nm Automotive Chips Begin Deployment
Recall that*announced mass production of its first-generation 3nm (N3, aka N3B) in late December 2022, followed by its second-generation N3E in Q4 2023, and its third-generation N3P in the second half of 2024. According to TSMC's roadmap, N3X will launch in 2025, and N3A (automotive-grade) in 2026.
Typically, advanced nodes like 5nm and 3nm are associated with smartphone and AI chips. Indeed, companies like Apple, Qualcomm, NVIDIA, and AMD currently dominate TSMC's 3nm capacity, with other vendors queuing for orders. Notably, nearly two years after the first mass production of 3nm chips, 3nm technology is finally entering the automotive chip domain.
In October 2024, MediaTek revealed during its "Dimensity 9400 Launch" that its first "Dimensity AI Cockpit Chip CT-X1" would enter mass production and deployment in vehicles in 2025. The CT-X1, based on TSMC's 3nm technology, boasts a CPU performance of 260K DMIPS and hardware-level GPU performance of 3,000 GFLOPS. Its on-device generative AI NPU offers 46+ TOPS, supporting multi-modal generative AI models with up to 13 billion parameters. Leveraging this performance, CT-X1 supports up to 10 displays, 16 cameras, 8K 30fps video playback/recording, 9K resolution display, and communication technologies like 5G and Wi-Fi 7.
As a cockpit chip, CT-X1 does not adhere to automotive-grade standards. Under highly integrated E/E architectures, computing power is the scarcest resource. The long development cycles for automotive-grade chips struggle to meet rapidly growing compute demands. Against this backdrop, automakers like Tesla and BYD are also adopting non-automotive-grade chips in their cockpits.
Of course, automotive chips using TSMC's N3A process are also emerging. On November 13, 2024, Renesas Electronics announced its next-generation automotive multi-domain fusion SoC, the R-Car X5 series. The first SoC in this series, R-Car X5H, is based on a Chiplet architecture and uses TSMC's 3nm automotive-grade process (N3A) – developed specifically for automotive SoCs and compliant with AEC-Q100 Grade 1 reliability standards.
R-Car X5H delivers up to 400 TOPS AI performance and excellent TOPS/W efficiency, alongside GPU processing power up to 4 TFLOPS and CPU performance exceeding 1,000K DMIPS. With 38 Arm cores, powerful AI, and GPU capabilities, a single chip can simultaneously support multiple in-vehicle applications including ADAS, In-Vehicle Infotainment (IVI), and gateways. R-Car X5H samples will be available to select automotive customers in H1 2025, with mass production planned for H2 2027.
Furthermore, back in October 2023, Japanese chip designer Socionext announced it had begun developing 3nm ADAS and autonomous driving custom SoCs, also utilizing TSMC's N3A process, with mass production expected in 2026.
These developments indicate that 3nm process technology is gradually becoming a new standard in the automotive chip field, providing robust computing power for smart car development. As more 3nm-based automotive chips enter the market, the transformation towards smarter and more electric vehicles will accelerate, offering consumers safer and more intelligent driving experiences.
Trend 9: Memory Market Seeks New Equilibrium Amid Price Volatility
After memory prices began recovering from their trough in Q4 2023 and experienced a strong rebound in H1 2024, different memory products started diverging in H2, leading to a "dichotomous" market situation in Q4. What trends are expected for the memory market in 2025?
For NAND Flash, consumer SSDs account for over 80% of the market, while enterprise SSDs make up nearly 15%. Starting from Q3 2024, the performance of different flash products began to diverge. Enterprise SSDs, buoyed by demand from AI servers, outperformed consumer SSDs, maintaining slight growth in H2 2024. In contrast, the sluggish consumer electronics market, particularly poor performance in smartphones and laptops, led to declining contract prices for consumer SSDs starting in Q3. Pessimism about the consumer electronics market in Q4 2024 and Q1 2025 suggests NAND Flash prices will remain weak.
Furthermore, while edge AI applications are expected to scale commercially in 2025, NAND Flash has yet to receive a significant boost from this trend. International Electronic Business data indicates that at least for Q4 2024 and Q1 2025, the overall NAND Flash price trend will be downward. Concurrently, memory manufacturers, learning from the industry-wide losses in 2023, are proactively controlling capacity amid oversupply. Consumer SSD prices are expected to rebound in H2 2025.
To better meet AI's data transmission demands, NAND Flash technology is evolving towards higher capacity, enhanced security, lower latency, and lower power consumption. Balancing capacity and cost has become a key focus for the NAND supply chain, with more QLC NAND products expected to emerge.
For DRAM, which constitutes over 50% of the memory industry, there are three main types: Standard DDR, Mobile DDR (LPDDR), and Graphics DDR (GDDR, HBM). Compared to standard DDR4/DDR5, Graphics DDR represented by GDDR and HBM offers significantly higher bandwidth.
Graphics DDR is designed for high-performance computing and graphics processing. GDDR has progressed to GDDR6X, with GDDR7 emerging, primarily used in high-end gaming GPUs and professional workstations. HBM and its iterations (HBM2, HBM2E, HBM3, HBM3E) are widely used in HPC and AI accelerators due to their ultra-high bandwidth and energy efficiency. With semiconductor process advancements, Graphics DDR will further increase density and efficiency to meet data processing needs, seeing broader application alongside 5G, AI, and VR development.
Price trends for Mobile DDR (LPDDR) and Standard DDR (DDR4/DDR5) are influenced by multiple factors and generally decline with fluctuations. Mobile DDR demand remains stable due to mobile device refresh cycles; new generations start high-priced but decrease with maturity and scale, often priced higher than Standard DDR due to power/performance requirements. Standard DDR4 is mature with stable prices, gradually being replaced by DDR5, which starts expensive but is expected to decrease as production ramps and demand increases. Overall prices are impacted by market supply/demand, raw material costs, capacity adjustments, and the global economic climate.
Trend 10: Digital Simulation Transforms More Traditional Industries
Before a rocket launch, R&D traditionally goes through four stages: Feasibility Study - Detailed Design - Prototype Testing - Flight Model Testing. The most expensive stage is Prototype Testing – manufacturing all components for physical verification.
SpaceX overturned this conventional approach. They conduct extensive simulation testing during the Detailed Design phase to achieve faster and more cost-effective rocket launches. The key lies in SpaceX's significant investment in simulation technology, making it a leader in aerospace, reportedly 10 years ahead of competitors.
Beyond aerospace, digital simulation currently accounts for only about 20% of the aircraft design process, with the remaining 80% relying on physical testing. In fields like life sciences and drug discovery, simulation's share in the R&D process is often less than 1%.
Compared to the semiconductor industry, where most design occurs in software/virtual environments, many traditional industries face a digital transformation imperative. The proliferation of digital simulation will help these sectors achieve higher efficiency and lower costs.
For the semiconductor and electronics industry, especially players like EDA/IP vendors and software/solution providers, this represents significant opportunities in technologies like digital simulation and digital twins. It also aligns with the broader trend of semiconductor and system integration – or the convergence of chips and systems.
This convergence manifests in several ways: system companies like Apple, Tesla, and AWS designing their own chips; traditional chip companies like NVIDIA and Qualcomm moving towards system-level design; and EDA/IP vendors finding vast new opportunities in system design due to the low digitalization levels in various application fields. Recent developer conferences from major EDA firms have heavily emphasized exploring these new system-related markets.
This represents a significant, unignorable opportunity for the electronics and semiconductor industry. Applications like NVIDIA Omniverse in the industrial metaverse for shipbuilding, railways, healthcare, and manufacturing are manifestations of this trend – part of the broader digital transformation sweeping across all industries and society.
