TrendForce: Unveiling New Opportunities in Tech Innovation for 2025

TrendForce has outlined 10 key trends shaping the technology landscape in 2025. Highlights include:

Generative AI Leads the Way: Humanoid and Service Robots Receive Major Upgrades

Robotics is poised to remain a central focus in 2025 as advancements in AI and robotics accelerate, driven by major players like NVIDIA and Tesla. The development will emphasize training machine learning, digital twin simulation platforms, collaborative robotics, mobile robotic arms, and humanoid robots to adapt to various environments and enable seamless human-machine interaction.

Humanoid robots, backed by substantial investment from U.S. and Chinese manufacturers, are set to begin mass production starting in 2025. The global humanoid robot market is forecast to achieve a staggering CAGR of 154% between 2024 and 2027 and potentially exceed US$2 billion in market value. Industrial robots primarily focus on tasks like arm-based picking; service robots, powered by generative AI, will enable multi-modal interactions, information retrieval, text summarization, and scheduling. These advancements will boost their mobility, companionship, and versatility, positioning service robots as the next frontier in robotics innovation.

Market Standardization Driven by Technological Advancements: AI Notebook Penetration to Reach 21.7% in 2025

AI-enabled notebooks are expected to become standard in the coming years with rapid advancements in technology. By 2025, AI notebook penetration is projected to reach 21.7%—climbing to nearly 80% by 2029. This surge will significantly contribute to the adoption of Arm-based architectures, which offer greater energy efficiency and scalability compared to traditional x86 architectures.

Arm-based notebooks are set to steadily increase their market share as edge inference demands grow and energy efficiency remains a priority. The growing popularity of Windows on Arm systems will provide consumers access to high-performance, low-power AI notebooks.

Although AI applications currently rely heavily on cloud computing, TrendForce anticipates that groundbreaking advancements in Edge AI will further drive the adoption of AI notebooks. Edge AI will empower notebooks by enabling local processing. This will allow them to handle real-time tasks such as voice commands and image recognition more efficiently, overall enhancing the user experience. Localized computing also ensures data privacy, particularly for sensitive information and fostering greater consumer trust in AI notebooks. As AI technologies mature, Edge AI will unlock new productivity potential in smart offices and automated workflow management and cater to diverse user needs.

AI Server Shipments to Grow Over 28% in 2025, with HBM 12hi Yield Improvement in Focus

The growing demand from CSPs and enterprise clients for AI infrastructure is driving substantial growth in the AI server market. In 2024, global shipments of AI servers—including those equipped with GPUs, FPGAs, and ASICs—are projected to grow by 42%. By 2025, fueled by strong demand from CSPs and sovereign cloud operators, annual shipment growth is expected to exceed 28%, accounting for 15% of the overall server market.

Starting in 2025, HBM3e 12hi will emerge as the mainstream stack height, with NVIDIA’s B300 and GB300 platforms expected to adopt this confirmation. SK hynix is leveraging Advanced MR-MUF technology in its 12hi generation, introducing a mid-temperature pre-bonding process during each die stacking step. SK hynix aims to effectively control die warpage by optimizing MUF materials and extending the processing time.

Conversely, Samsung and Micron are sticking with the TC-NCF stacking architecture for their 12hi products. While this method offers easier warpage control, it faces challenges such as longer process times, greater cumulative stress, and reduced thermal dissipation performance. These limitations introduce significant uncertainties in achieving rapid yield improvement during mass production.

As the adoption of the 12hi stack height extends across HBM3, HBM3e, HBM4, and HBM4e generations (spanning 2027–2029), the timeline for mass production will stretch over several years. Consequently, improving and stabilizing the yield rates of 12hi manufacturing processes will be a critical priority for suppliers throughout 2024 and beyond.

Advanced Processes and AI Drive Semiconductor Innovation and Strong Growth in CoWoS Demand in 2025

As semiconductor manufacturing progressed beyond the 7 nm process node, the adoption of EUV lithography pushed FinFET architecture to its physical limits at 3 nm. This prompted a divergence in advanced process technologies. In 2023, both TSMC and Intel continued using FinFet for their 3 nm products. Meanwhile, Samsung sought to lead the transition to GAAFET with its MBCFET architecture, initiating production in 2022. However, widespread adoption of Samsung’s approach has yet to take hold.

TSMC will introduce its nanosheet transistor architecture at the 2 nm node by 2025, while Intel plans to adopt its RibbonFET technology for its 18A node. Samsung will continue refining its MBCFET-based 3 nm process as it aims to achieve large-scale production by 2025. This marks the official entry of all three players into the GAAFET race, targeting superior transistor control through four-sided gate contact and promising higher performance, lower power consumption, and increased transistor density per unit area.

The growing need for customized chips and larger packaging areas driven by AI applications is simultaneously boosting CoWoS demand in 2025. There are a number of key developments expected in the CoWoS market next year: Firstly, NVIDIA’s demand for TSMC CoWoS is projected to rise to nearly 60% of TSMC's total CoWoS output, driving a near doubling of TSMC’s CoWoS monthly production capacity to approximately 75–80K units by the end of 2025. 

Secondly, the launch of NVIDIA’s Blackwell platform in the first half of 2025 will significantly increase demand for CoWoS-L—surpassing CoWoS-S—with CoWoS-L expected to account for over 60% of the total. Lastly, CSPs are ramping up investments in the development of ASIC AI chips. Major players such as AWS are anticipated to significantly increase their demand for CoWoS in 2025.

AI as a Double-Edged Sword in 2025: Enhanced Cybersecurity Defense and Threat Detection to Counter Complex Attacks
    
The current focus of global cybersecurity lies in the hardware and software of the cloud-driven IoT era. The complexity of both attack and defense strategies has grown significantly compared to the past as various technologies continue to advance, prompting companies to gradually shift their focus from IoT to AI. Meanwhile, generative AI is seeing two major trends in enhancing cybersecurity defenses through empowering operators and accelerating threat detection.

The first trend enables operators to identify and respond to major risks through automated translation and data consolidation, allowing them to interact using natural language. The second facilitates faster identification of vulnerabilities by guiding users and providing operational recommendations that significantly reduce detection cycles.

However, generative AI is being equally exploited by hackers to strengthen offensive tactics such as enumeration analysis and phishing. Analyzing the risks associated with the creation of LLMs reveals several critical challenges that include faulty outputs from input manipulation, vulnerabilities introduced during training, lack of comprehensive access control, and excessive autonomy in functionality. As 2025 approaches, these risks represent pressing cybersecurity challenges that enterprises must address when developing AI-driven products and services. 

AMOLED Expands into Medium-Sized Applications, Boosting Notebook Market Penetration to 3%

In 2024, Apple officially launched its iPad Pro series featuring RGB AMOLED panels, signaling the expansion of RGB AMOLED technology into medium-sized product applications. Beyond tablets, the trend of integrating AMOLED panels into notebooks is also gaining momentum. While Apple plans to introduce AMOLED panels to its MacBook series between 2026 and 2027, the company has already begun encouraging panel makers to scale up investments. They are now transitioning RBG AMOLED production lines from 6th-generation to 8.6 or 8.7-generation configurations to meet anticipated demand.

This strategic shift has spurred other brands to accelerate their market positioning, leveraging existing production lines to capture early opportunities. By 2025, the market size for AMOLED-equipped notebooks is projected to surpass 6 million units, with an estimated penetration rate reaching 3%.

Vision Pro Transforms VR/MR from Entertainment to Productivity Tool; LEDoS Near-Eye Display Sets Milestones for AR Weight and Visual Experience

The most pivotal development in the VR/MR headset market in 2024 will be Apple’s launch of the Vision Pro, which redefines VR/MR devices by shifting their primary use case from entertainment and leisure to productivity tools. This repositioning is expected to inspire other manufacturers to release innovative new products. The Vision Pro’s display leverages OLEDoS technology, delivering resolutions exceeding 3,000 PPI and making it the preferred near-eye display solution for high-end VR/MR devices. TrendForce forecasts that VR/MR device shipments will reach 37 million units by 2030.

AR glasses, positioned as assistive devices, will also see renewed market interest in 2024 as it’s fueled by advancements in AI technology. Meta’s Orion, though not a mass-produced device, integrates LEDoS displays and SiC-based waveguides, achieving an impressive 70-degree FOV and setting a new benchmark for lightweight design with a weight under 100 grams. Alongside LEDoS, current near-eye display technologies for AR glasses include OLEDoS, LCoS, and laser beam scanning (LBS), offering diverse solutions that enhance the flexibility of AR hardware design. TrendForce projects AR device shipments will reach 25.5 million units by 2030.

Miniaturization and Low-Cost Production of CubeSats Drive Global Communication and IoT Revolution in 2025

With the guidance of 3GPP Release 17 for satellite application, the number of CubeSats within low Earth orbit constellations has been growing exponentially. Emerging satellite companies are leveraging low-cost production techniques for small CubeSats and deploying large-scale satellite constellations to provide global low-latency satellite communication coverage.

Looking ahead to 2025, satellite miniaturization is set to accelerate. Small and medium-sized emerging satellite operators are utilizing modular satellite platforms and commercial-off-the-shelf components to launch large-scale CubeSat production, which is set to significantly reduce production costs. In parallel, these companies are deploying CubeSat constellations for Space Situational Awareness (SSA), focusing on monitoring and clearing space debris. Additionally, the development of satellite IoT applications is progressing rapidly, which is supporting remote IoT device monitoring in areas such as agricultural sensors and driving innovation in connectivity for remote and underserved regions.

Modular End-to-End Model Production and Level 4 Robotaxi Commercialization Accelerate in 2025

Autonomous driving, a key application area for Edge-AI, is seeing rapid advancements, with Tesla spearheading the adoption of end-to-end (E2E) models. This trend is prompting other automakers to accelerate their investments in AI technology and computational power, with 2025 expected to mark the beginning of mass production for this architecture. However, most manufacturers are anticipated to adopt modular end-to-end models, which offer advantages in interpretability and debugging.

E2E models are data-driven and heavily reliant on diverse datasets. Generative AI, due to its openness and creativity, plays a crucial role in producing varied and rare scenarios to train these models, effectively addressing the long-tail problem in data distribution. 

Advancements in AI technology are also extending into the commercial sector. Level 4 autonomous robotaxis are poised for accelerated replication and commercialization as regulatory frameworks gradually improve. Nonetheless, challenges in both electrification and autonomous driving are heightened by geopolitical factors, which complicate technology and commercial expansion efforts.

EVs and AI Data Centers Drive Battery and Energy Storage Innovations in 2025

The growth of the EV market is slowing, with BEVs experiencing the most pronounced deceleration. By 2025, the BEV growth rate is projected to shrink to 13%. Range anxiety remains a significant barrier to BEV adoption, prompting the industry to focus on solving this challenge. In battery technology, CATL has introduced LFP batteries with a 4C charging rate, capable of delivering 600 km of range with just 10 minutes of charging. These batteries are expected to see broader market adoption by 2025. Additionally, semi-solid-state batteries entered mass production in 2024 and are expected to accelerate integration into vehicles by 2025, while fully solid-state batteries are projected for commercialization post-2027.

In terms of charging infrastructure, megawatt-level charging systems tailored for commercial trucks and passenger vehicles will launch in 2024, driving the development of high-power charging technologies. These advancements aim to alleviate range anxiety and meet the growing demand for faster charging and extended range.

Simultaneously, as charging technologies advance, automakers are also working to improve overall EV performance and user experience to adapt to market changes and remain competitive. By 2024, smart connectivity and autonomous driving features will see widespread adoption in EVs, significantly enhancing energy efficiency, intelligence, and safety.

Furthermore, the rapid expansion of AI-powered data centers is driving a surge in demand for advanced energy storage systems. With continuous technological advancements and declining costs, global energy storage installations are expected to reach 92 GW/240 GW by 2025, representing an annual growth of 25% in capacity and 33% in energy. The swift growth of AI technologies has significantly increased electricity demand, making energy storage systems essential for stabilizing renewable energy output and providing backup power during outages, and thereby improving data center reliability.

As the data center industry continues its robust growth, the construction of new facilities is expected to remain steady and create vast opportunities for next-gen energy storage systems. This trend underscores the crucial role of energy storage in supporting the expanding infrastructure needs of the AI and renewable energy ecosystems.

TrendForce Roadshow Japan – AI Innovations and Opportunities
Join us on December 12 at Villa Fontaine Grand Tokyo Ariake for an insightful seminar hosted by TrendForce. Our analysts will present on key topics, including foundry, HBM, and AI server. If you are interested in attending, please contact us for more details. We look forward to welcoming you!