At COMPUTEX 2025, ASUS showcased its comprehensive AI infrastructure solutions, targeting the computing needs of data centers, edge computing, and a wide range of intelligent applications. With a robust product portfolio, ASUS brings to life the NVIDIA AI Factory vision, illustrating how AI will be deeply embedded into enterprise IT infrastructure. This evolution continues to drive forward applications and services that shape the future of AI and high-performance computing

Linxens, a global leader in component-based solutions for authentication and traceability, and Dracula Technologies, a pioneer in ambient light energy harvesting, have announced a strategic collaboration to develop next-generation autonomous IoT solutions for traceability and smart labelling applications. This collaboration paves the way for smarter and more sustainable connected devices

Technology news, product launches and show highlights from the event attendees available for download (PDF).

The rise of generative AI is accelerating digital transformation across traditional industries, and the construction sector is no exception. In an industry long dependent on hands-on expertise and experience sharing, construction companies face mounting challenges including labor shortages, rising costs, project delays, and increasingly stringent regulatory requirements. To address these issues, Shipeng Technology has developed an end-to-end smart construction solution centered on a "Digital Twin + AI Platform," powered by the robust computing capabilities of Amazon Web Services (AWS).Integrating AI, cloud, and digital twin technologies to drive smart constructionAccording to Shibo Lin, founder of Shipeng Technology, the company's core solution revolves around three key pillars: visibility, precision, and control. Covering the entire project lifecycle - from design and construction to operation and maintenance - it establishes a comprehensive digital twin management process. By replacing traditional 2D blueprints with 3D BIM (Building Information Modeling), design conflicts can be identified and resolved before construction begins. The AI platform integrates material and scheduling data, allowing on-site teams to query project progress and cost details using natural language. In addition, AR and sensor technologies enable real-time comparison between physical structures and digital models, ensuring quality and progress remain aligned. The result is a significant reduction in human error and delay risks, along with enhanced data-driven decision-making capabilities.Shipeng Technology leverages AWS cloud computing, AI/ML platforms, and IoT services to handle the massive volume of construction data and 3D model computations. By integrating sensor, image, and progress data in the cloud, the company builds high-precision digital twins that remain perfectly synchronized with the physical site. This architecture not only ensures flexibility, scalability, and security but also enables rapid model updates and data processing within minutes - greatly reducing operational costs and accelerating project delivery. Lin emphasized that AWS's global cloud infrastructure allows the team to focus on application innovation and AI model optimization, truly realizing the vision of bringing "smart construction from cloud to job site."Real-world projects demonstrate efficiency gainsIn a recent major construction project, regulatory changes required a complete redesign. Traditionally, the revision process would take four months. With Shipeng Technology's BIM solution and AWS support, the team completed the entire update and validation in just one week, reducing the timeline by over tenfold and saving substantial downtime costs.In another project, the Startup Terrace A6 Tower in Linkou adopted Shipeng Technology's digital asset management system. Tasks that once took three months - such as inventory verification - are now completed within five to ten minutes. Supported by AWS's real-time data integration and high-performance computing, the platform enables project managers to monitor on-site progress and asset conditions seamlessly. Lin noted that Shipeng Technology aims to make every data point from construction sites actionable, turning real-time insights into the foundation for smarter decisions.From Taiwan to the world: building a cloud ecosystem for smart constructionThrough its participation in the Startup Terrace Kaohsiung AWS Joint Innovation Center (JIC) project, Shipeng Technology established a presence in Kaohsiung's Startup Terrace and gained access to extensive local industry resources. The project's matchmaking and mentoring programs have helped the company connect with construction partners in southern Taiwan and refine its business presentation and client engagement capabilities. Lin remarked that this industry collaboration model - combining local partnerships with cloud-based coordination - not only accelerates digital transformation in southern Taiwan's construction industry but also strengthens Shipeng Technology's local market foothold.While deepening its presence in Taiwan, Shipeng Technology is actively expanding internationally. Following its strategy of "Taiwan R&D, Japan validation, and global deployment" , the team recently participated in a global startup competition and secured collaboration opportunities with Japanese partners to deploy smart construction solutions overseas. Given Japan's strong demand for construction automation and ESG management, the market serves as the company's first step toward globalization. Moving forward, Shipeng Technology will continue to establish international standards for smart building practices and bringing its digital construction technologies to the world.

Axelera AI is a startup targeting Edge AI applications. Leveraging its Metis AI Processing Unit (AIPU) chip, embedded hardware AI inference acceleration cards, and Voyager SDK software solutions, the company specializes in imaging application solutions for applications in edge nodes. Unlike AI systems developed in high-density cloud data centers, Edge AI computing offers energy efficiency, information security, low latency, and fast speed of response, rapidly powering real-time vision and smart detection applications for embedded AI systems.Axelera AI's technology offerings span 4K/8K image resolution real-time image recognition, security surveillance, campus safety, and drone imaging, pioneering a new model for the development of high-performance AI inference in diverse fields. The company's Metis chip boasts a computing performance of up to 214 TOPS (TeraOPS) of inference performance at INT8 precision. For example, using ResNet-50 Convolutional Neural Network (CNN) model doing object segmentation and analysis, a single Metis chip can achieve a high processing speed of 3200 frames per second (FPS), providing stable, real-time analysis capabilities while consuming only approximately 10W. This balance of performance and environmental benefits has attracted significant market attention.Meanwhile, the AI inference acceleration cards integrated with Metis chips are also a major product line. There are two types of accelerator cards that support M.2 and PCIe interfaces, which can quickly meet the needs of system integrators, OEM customers and industrial PC (IPC) makers. With these flexible connection interface, Axelera could engage the business with various embedded AI system makers.In addition to the hardware lineups, Axelera's software support is also a key factor in its success. The Voyager SDK software suite facilitates hardware and software integration for system integrators. There are open source software framework currently available on GitHub for computer vision and enabling customers to solve their AI business requirements. This Voyager SDK comes with a Model Zoo, a catalog of turnkey AI models, allowing customers either to import their own AI models or choose from Model Zoo models. These offerings significantly reduce developing time and excels in functional completeness and efficiency.Axelera AI's business model is based on collaboration with system integrators and OEM clients, encompassing joint development of AI projects and customized data architectures, providing clients with one-stop, high-performance Edge AI solutions. Axelera AI is working closely with major Taiwanese industrial PC vendors and OEM/ODM clients to develop key applications for financial and research institutions, medical solution providers and other public facilities that prioritize data security, privacy, and safety. This initiative aims to create emerging applications for Edge AI and unlock its significant value.

Empowered by the Startup Terrace Kaohsiung AWS Joint Innovation Center (JIC), eNeural is redefining the landscape of edge AI with two flagship innovations - AI-Craft, an automated AI development platform, and an edge AI self-learning solution. These breakthroughs aim to make AI truly self-learning and self-evolving, helping industries overcome long-standing barriers in deploying AI at the edge.As AI applications rapidly expand across industries, enterprises have realized that beyond computing power, the biggest barriers to AI implementation lie in the high cost and lengthy cycle of model training. For system integrators, success increasingly depends on the ability to integrate optimized AI models into lightweight edge devices to deliver reliable inference services in real time.A spin-off startup from National Yang Ming Chiao Tung University (NYCU), eNeural combines strong academic research foundations with practical AI R&D experience. The company has already forged partnerships with major electronics manufacturers including Quanta, Compal, and Lite-On, and is collaborating with a North American e-commerce logistics fleet on developing an AI-based collision warning system.According to eNeural CEO Eric Huang, edge devices are typically designed to be small, lightweight, and energy-efficient, which makes AI model optimization particularly challenging. "Our AI-Craft platform provides a one-stop, modular, and standardized AI development workflow," said Huang. "It allows developers to automatically generate optimized models as efficiently as products rolling off an intelligent production line."Edge AI self-learning accelerates model optimization and reduces costsFounded in 2022, eNeural focuses on proprietary AI technology R&D, with AI-Craft serving as the backbone of its innovation. The platform integrates four key technology modules - automated annotation, model pruning, model quantization, and generative AI data augmentation - to make AI development faster, more scalable, and more energy-efficient.Using pre-trained AI models and advanced image analysis algorithms, AI-Craft boosts data labeling efficiency by 10 to 100 times, significantly shortening AI project launch time. Model pruning removes redundant parameters to enhance computational efficiency by up to 90% without compromising accuracy, while reducing the memory and processor load.Model quantization converts floating-point operations into 8-bit, 4-bit, or even 2-bit precision, dramatically lowering power consumption and enabling smooth deployment of AI models on compact edge devices. Meanwhile, generative AI data augmentation automatically creates training data for rare yet critical conditions - such as night, rain, snow, or backlight - improving AI model robustness and generalization.Traditionally, AI models deployed on edge devices must send large amounts of data back to the cloud for retraining - a process that consumes time and resources. eNeural's Edge AI Self-Learning technology allows models to learn and recalibrate autonomously in real time, reducing the need for cloud data transfer."For example," Huang explained, "when fleet vehicles operate under different weather, lighting, or road conditions, the AI model can automatically adjust its decision logic to maintain accuracy. With federated learning, training results are securely shared across thousands of vehicles." This innovation reduces retraining cycles from months to weeks and significantly lowers AI operation and maintenance costs.AWS JIC Accelerates cloud transformation and strategic collaborationSince joining the AWS JIC in 2025, eNeural has leveraged the expertise of AWS specialists to upgrade its AI-Craft platform using Amazon SageMaker, a fully managed machine learning (ML) service, and Amazon Elastic Compute Cloud (Amazon EC2) GPU computing power. This transformation has turned AI-Craft into a cloud-based development solution, enabling developers to perform automated annotation, model pruning, and quantization in the cloud. With dynamic AWS resource allocation, eNeural can efficiently run multiple AI projects in parallel with greater scalability and cost efficiency.Participation in the AWS JIC program has also opened doors for new business collaborations. Huang shared that while earlier partnership talks with Advantech did not progress, joining the AWS JIC program led to the successful signing of a Memorandum of Understanding (MOU) between the two companies in 2025. Under this collaboration, eNeural will feature its two flagship AI solutions on Advantech's Marketplace, paving the way for deeper engagement with system integrators worldwide.Beyond Taiwan, eNeural is actively expanding into the U.S., Japan, and India. Leveraging AWS's global cloud infrastructure and innovation ecosystem, the company is pursuing a "software-driven core + hardware collaboration" strategy, collaborating with global players to unlock new opportunities in the rapidly growing edge AI market.