As the AI computing race accelerates and semiconductor manufacturing capacity is increasingly redirected toward advanced memory, the memory industry is shifting from price-driven competition to a battle over resource allocation and supply security. The rapid expansion of high-bandwidth memory (HBM) production continues to consume advanced fabrication capacity, constraining the supply of conventional memory products and creating persistent shortages across the market. This concentration of high-end production has widened the supply-demand gap, driving the average selling prices of certain premium memory products to levels eight to ten times higher than previous norms.Against this backdrop, TEAMGROUP General Manager Gerry Chen warned that the industry could face a period of unprecedented scarcity by 2027. He described the situation as one where "even customers willing to accept higher prices may still face supply risks," potentially echoing the severe chip shortages seen during 2021–2022. Chen noted that even buyers willing to pay premium prices could face significant procurement challenges if they lack long-term strategic relationships with original chip manufacturers, leaving them highly vulnerable to supply disruptions.In response, TEAMGROUP initiated a strategic transformation in 2025, redirecting 80% of its resources from the consumer segment toward long-lifecycle, high-reliability applications. The company is now focused on mission-critical sectors, including industrial automation, enterprise infrastructure, defense, and healthcare, while also expanding into military-grade storage and high-density computing applications. Leveraging three decades of partnerships with leading global suppliers and its extensive experience in IPC design-in collaboration, TEAMGROUP has strengthened supply-chain resilience and enhanced supply assurance amid ongoing market volatility, positioning itself as a trusted enabler of secure and reliable AI infrastructure.Building supply chain resilience amid persistent memory shortageChen emphasized that the exponential growth of AI-driven demand is outpacing semiconductor capacity expansion, a process that typically requires three to three-and-a-half years to complete. As a result, supply constraints in DRAM and SSD markets are becoming a structural challenge rather than a cyclical phenomenon. In this environment, competitive advantage will increasingly depend on supply assurance, allocation capabilities, and long-term supplier relationships rather than price alone.Leveraging decades-long partnerships with leading memory manufacturers and extensive IPC expertise, TEAMGROUP has established a stable supply framework for industrial PCs and other long-lifecycle applications requiring seven to ten years of product availability. The company works closely with customers to identify alternative component sources, optimize system configurations, and mitigate demand distortion and bullwhip effects across the supply chain. These capabilities have strengthened resilience for customers in mission-critical industries such as automotive, healthcare, and defense.T-CREATE EXPERT P35S & TEAMGROUP INDUSTRIAL P250Q secure data in defense & finance. Credit:TEAMGROUPTEAMGROUP's portfolio spans commercial-grade storage solutions for general confidential data protection, as well as industrial-grade products engineered for deployment in aircraft, naval vessels, unmanned systems, and other high-reliability platforms. The company has showcased these solutions at major industry events, including Embedded World and Japan IT Week.Physical destruction technology enables one-click data destructionThe T-CREATE EXPERT P35S, recipient of the 2026 COMPUTEX Best Choice Award, exemplifies TEAMGROUP's philosophy that "physical destruction is the ultimate safeguard." According to Chen, true data security lies in the complete physical destruction of storage media rather than relying solely on software-based deletion methods.When activated, the SSD's built-in boost chip instantly releases a high-voltage current that physically breaks down the insulation layer of the NAND flash ICs, permanently destroying the storage medium within 2.4 seconds and making data recovery extremely difficult. For everyday protection, AES-256 encryption provides a first line of defense, creating a comprehensive security framework that integrates both hardware- and software-based safeguards.The design is engineered to eliminate any possibility of residual data. Its proprietary circuit architecture uses a zoned sequential-conduction mechanism, enabling high-voltage current to destroy memory chips in a controlled chain reaction. TEAMGROUP's patented "power-resume destruction" technology further ensures destruction integrity by automatically completing the process when power is restored if an interruption occurs during the destruction sequence.To prevent accidental activation, the device incorporates a two-stage fail-safe button requiring deliberate user action. The technology also supports remote activation through 4G/5G networks, radio communications, or SMS commands, allowing data destruction to be triggered from several kilometers away. This capability helps prevent sensitive information from being compromised even if a device is lost, stolen, or captured.The patented technology is currently available in two products: the T-CREATE EXPERT P35S, a portable secure storage device, and the industrial-grade TEAMGROUP INDUSTRIAL P250Q-M80 M.2 high-speed SSD, designed for classified computing environments, financial institutions, and defense applications. Together, they provide a robust cybersecurity safeguard for the defense, finance, and R&D sectors.TEAMGROUP showcases a variety of enterprise-grade SSDs at COMPUTEX. Credit: TEAMGROUPAdvanced specs pave the way for agentic AILooking ahead to the convergence of AI and physical robotics, Chen noted that technology ultimately serves human needs. As AI systems continue to collect and process vast amounts of behavioral, operational, and contextual data, demand for on-premises storage is expected to grow significantly.TEAMGROUP keeps data stable in harsh settings & inspires young teams to innovate. Credit: TEAMGROUPOne example is the emerging long-term care robotics market. These systems must not only deliver healthcare-related capabilities but also retain medical records and behavioral data over extended periods. Such requirements place stringent demands on SSD endurance, reliability, and long-term stability, underscoring the importance of local data storage. Similarly, enterprises are increasingly reluctant to move highly sensitive information—including financial data, proprietary business models, and operational intelligence—to the cloud. As a result, agentic AI systems capable of autonomous decision-making within on-premises environments are gaining strategic importance.To support next-generation AI applications, TEAMGROUP has become one of the industry's early innovators in LPCAMM2 and SOCAMM2 memory technologies. LPCAMM2 enables LPDDR memory, traditionally soldered directly onto notebook motherboards, to become modular and upgradeable, while SOCAMM2 introduces a compact, low-power architecture designed for AI servers and advanced computing platforms, offering an alternative to conventional RDIMM. Together, these technologies offer greater flexibility, serviceability, and power efficiency for future AI systems.Beyond supplying low-power memory components, TEAMGROUP has evolved into a comprehensive solution provider, helping customers maintain data integrity and system stability in demanding operating environments characterized by vibration, thermal challenges, and continuous workloads. At the same time, the company encourages its young talent to pursue bold innovation while investing in next-generation DDR6 R&D, laying the groundwork for future AI infrastructure and intelligent edge computing applications.Amid the ongoing memory supercycle, TEAMGROUP has demonstrated strategic resilience and innovation beyond the role of a traditional module manufacturer. By combining supply-chain expertise, advanced physical-level security technologies, and next-generation memory architectures, the company aims to help customers safeguard critical digital assets while preparing for the next phase of AI-driven transformation.For more information, please visit.

On the opening day of COMPUTEX 2026, AIC Inc. hosted a high-level strategic panel session at its booth, focusing on overcoming the "memory wall" challenge. Industry giants and key strategic partners, including NVIDIA and VAST Data, joined AIC for a presentation on their latest platforms designed to eliminate bottlenecks in Large Language Model (LLM) inference and intensive AI workloads, marking a critical evolution in active AI storage driven by Agentic AI in 2026.In his opening remarks, Michael Liang, CEO and President of AIC, outlined the new challenges facing AI infrastructure as AI applications enter the "Long Context" era. The transition to long-context and Agentic AI has completely shifted the primary AI infrastructure bottleneck from raw computational speed to massive data movement and memory bandwidth constraints—a hurdle commonly known as the "Memory Wall."Liang emphasized that the shift toward autonomous AI agents executing task decomposition and multi-step APIs is fundamentally transforming data center demands and reshaping underlying AI infrastructure. Consequently, AIC is actively collaborating with NVIDIA and VAST Data to develop advanced, AI-native storage solutions. By integrating the NVIDIA Vera BlueField-4 STX Storage Processor into its hardware platforms, AIC is building the essential infrastructure required to eliminate bottlenecks and accelerate workloads for Agentic AI applications.NVIDIA Ecosystem Scales Agentic AI Adoption WorldwideJason Hardy, NVIDIA's Vice President of Storage Technology, highlighted the significance of "Agentic Inferencing", a key theme from the NVIDIA GTC Taipei keynote during COMPUTEX 2026. Agentic AI requires more than faster compute; it demands fast, secure access to context memory so agents can reason across long sessions, large datasets, and complex workflows.NVIDIA Vera BlueField-4 STX addresses this paradigm shift. It enables a new class of AI-native storage infrastructure for context memory, built with Vera-based BlueField-4, NVIDIA Spectrum-X Ethernet, NVIDIA DOCA, NVIDIA Dynamo, and NVIDIA AI Enterprise. This foundation provides NVIDIA's storage partners, such as AIC, with the essential building blocks to keep agent context and inference data close to the compute path, significantly improving throughput, responsiveness, and infrastructure efficiency.NVIDIA is actively building a robust partner ecosystem around the NVIDIA Vera BlueField-4 STX architecture, spanning storage, systems, cloud infrastructure, and security sectors. Key partners like AIC are collaborating closely with NVIDIA to integrate, validate, and bring this next-generation infrastructure to market. Hardy emphasized that these close alliances will help customers optimize resource utilization, reduce costs, accelerate response times, and enhance security during large-scale deployments, thereby ushering in the era of Agentic Inferencing.VAST Data and AIC Hard-Soft Integration Optimizes AI InfrastructureEchoing the new design of NVIDIA Vera BlueField-4 STX, VAST Data CTO Andy Pernsteiner emphasized that Agentic AI requires sophisticated mechanisms for managing and optimizing massive-scale KV caching to persistent memory. This avoids redundant, expensive prefill computations across multi-turn, long-context sessions, while providing new storage platforms that support confidential computing and data protection for highly sensitive information. VAST Data integrates seamlessly with NVIDIA's BlueField-4 DPU architectures and Dynamo routing frameworks to offload, share, and reuse KV cache context across wide GPU clusters.The strategic hardware-software partnership between VAST Data and AIC pairs AIC's advanced server hardware with VAST's software intelligence to build next-generation AI infrastructure and context memory storage platforms. Integrating NVIDIA Context Memory Storage (CMX) platform, featuring the NVIDIA Vera BlueField-4 STX storage processor, effectively resolves GPU KV cache bottlenecks. By utilizing fast NVMe arrays as a shared, high-bandwidth context tier, the solution significantly increases tokens-per-second throughput and energy efficiency for long-context, multi-turn AI inferencing.AIC Embraces NVIDIA Vera BlueField-4 STX for Agentic AIAs Liang stated in a post-event interview with DIGITIMES, the company has successfully built its storage server business since 2014. By continuously reinvesting 15% of its annual revenue every year into R&D and early-stage development of new architectures, AIC has positioned itself as a key player in developing next-generation, Agentic AI-native storage infrastructure.Today, AIC is established as a Solution Advisor in the NVIDIA Partner Network (NPN). AIC also is building upon a strategic partnership with VAST Data that began seven and a half years ago. To meet the surging demand for AI data centers, AIC's strategic expansion in Yangmei, Taiwan, and Haiphong, Vietnam, directly targets the skyrocketing global demand for artificial intelligence data centers. These state-of-the-art manufacturing footprints allow the company to scale production of AI servers and high-density storage while seamlessly integrating computing, networking, and security into unified infrastructure platforms.The new facilities anchor AIC's global supply chain and position the company to meet the intense deployment needs of cloud service providers and enterprise customers. This empowers customers to maintain a competitive lead and achieve greater success amidst the AI wave.

The megatrend in electronic design today is end-to-end collaboration across ICs, packaging, PCBs, systems, data centers, and physical applications, with rapidly evolving artificial intelligence playing an increasingly critical role.In early June, Graser Technology held its annual technology forum, Graser TECHTALKS 2026, under the theme "AI in Sync: Intelligent Design, Accelerated Manufacturing." The event focused on how AI connects design, analysis, and manufacturing workflows. It brought together industry speakers, in-house engineering experts, and customer representatives to share professional insights and real-world experience, outlining a new paradigm for electronic design workflows and industrial applications in the AI era.In her opening remarks, Graser Chairwoman Lillian Pan said the company has, for more than 30 years, upheld the principles of fast response, professional service, and long-term partnership, helping customers turn ideas into products faster. She added that Graser will continue promoting the leverage of AI across engineering workflows, introducing advanced design tools, and supporting Taiwan's semiconductor and electronics industries in remaining globally competitive.AI as a Design Workflow CollaboratorIn the first keynote, "Paradigm Shift of System Design in the AI Era," Michael Shih, Corporate Vice President for APAC and Japan at Cadence, said electronic design is facing a new level of complexity as Moore's Law becomes harder to sustain and the cost of advanced process technologies and system integration continues to rise.He noted that the challenge is no longer limited to designing a single chip. Instead, engineering teams must increasingly solve complex issues across chips, advanced packaging, PCBs, system-level design, and multiple physical domains. Against this backdrop, Cadence has been expanding its focus from IC design into packaging, PCB design, multiphysics simulation, data centers, and system analysis, evolving from a traditional EDA tool provider into an Intelligent System Design platform company.Shih explained that Cadence's Intelligent System Design platform brings together AI, EDA and IP, system design and analysis, and computational software. This enables engineering teams to perform simulation, analysis, optimization, and design verification at the system level. Within this framework, Cadence is pursuing AI in two directions: Design for AI, which helps customers build AI infrastructure, and AI for Design, which embeds AI directly into design solutions. In other words, AI is not only an application enabled by advanced ICs and systems; it is also becoming a core collaborative capability within the electronic design process.A major part of this shift is the introduction of agentic AI into design workflows. Shih said Cadence is bringing AI agents into front-end design and verification, digital implementation, and custom and analog design processes.These AI agents can help engineers understand design goals, break down tasks, execute workflows, and accelerate iterative design cycles. Their value goes beyond labor savings: by automating repetitive and time-consuming work, AI agents allow design teams to explore feasible options faster, shorten development cycles, and reduce the time and cost pressures created by rising complexity of designs.Shih noted that, for example, many companies must complete large numbers of board designs every year, involving repetitive yet expertise-intensive tasks such as placement, routing, layout, and design checks. By introducing AI into these workflows, engineers can spend more time on system architecture, reliability, and innovation. This suggests that design automation in the AI era is moving beyond point-tool acceleration toward broader efficiency gains across ICs, packaging, PCBs, and system-level simulation.AI Deployment Through System IntegrationFocusing on system integration design trends in the AI era, Eric Kao, Business Development Director at Giga Computing, shared his perspective from the data center infrastructure side. He noted that as enterprises adopt AI agents and generative AI applications, inference workloads are growing rapidly, pushing data center architectures originally optimized for AI training to shift.This shift is also redefining the role of the CPU. Because AI agent workflows involve task decomposition, step-by-step planning, API calls, tool invocations, and other logic-heavy and I/O-intensive operations, the CPU is no longer just a supporting component next to GPUs or accelerators. Instead, it is becoming the control and orchestration hub inside the AI data center.Kao pointed out that future AI infrastructure will move toward more refined heterogeneous computing configurations. Effectively managing different platforms and resources—and matching the right hardware to the right models and workloads—will become a critical system design challenge.Giga Computing's own technology roadmap also reflects this transition. According to Kao, the company has expanded from server motherboards and system development into HPC, OCP, GPU servers, liquid cooling, heterogeneous computing platforms, and broader AI infrastructure services. This shows that competition in AI data centers is shifting from standalone server specifications to integrated capabilities across racks, cooling, networking, software, POD design, and system-level simulation.Po-Ting Lin, Professor in the Department of Mechanical Engineering and Director of the Center for Intelligent Robotics (CIR) at National Taiwan University of Science and Technology (NTUST), approached AI from the perspective of physical system applications. He shared his team's experience applying AI to obstacle-avoiding path planning for robotics.Lin explained that when a robot encounters nearby people or obstacles during operation, it must quickly determine a safe trajectory to avoid collisions. Traditional optimization methods can be used to search for safe paths, but they often require significant computation time. By incorporating AI models, the system has the potential to greatly shorten response time.Lin emphasized that robot obstacle avoidance is not about taking the longest possible detour. The goal is to find a path that avoids obstacles just enough while maintaining task efficiency. NTUST's robotics research covers human-robot collaborative robotic arms, UAV inspection, and dual-arm robotic systems, with a common focus on balancing safety and operational efficiency.Through the insights shared by these two speakers, it is evident that bringing AI into real-world applications depends not only on a single chip or algorithm but also on the integration of computing, software, sensing, simulation, and physical systems.Intelligent Tools and Simulation Integration Across the Design FlowThe afternoon sessions of Graser TECHTALKS 2026 focused on two major tracks: electronic system design automation and multiphysics simulation. Graser's engineering team highlighted the latest advances in Cadence Allegro/OrCAD X 25.1 and Allegro X AI, demonstrating how automation and AI-assisted design can improve PCB development workflows.The program also featured technical experts from AIC, Supermicro, and Cadence, who shared practical insights into power integrity, electrothermal co-simulation, AI server system design, and multiphysics optimization, spanning packaging to system-level design, using Cadence Sigrity, Clarity, Celsius 3D, Sigrity HPC, and Aurora.Graser also presented updates to its in-house software portfolio, including GraserWARE, GIMS, and CAMPro, addressing requirements such as circuit reliability checks, component and BOM management, and manufacturing data validation.Building on features introduced last year, the company added several practical tools to GraserWARE MSAPack, including simulation schedule management, stackup format conversion, S-parameter port-naming optimization, temperature-dependent material parameter fitting, and automatic Power Tree generation. These capabilities help streamline SI/PI simulation workflows while improving analysis efficiency and data consistency.The key takeaway from Graser TECHTALKS 2026 is that in the AI era, design competitiveness goes beyond upgrading individual tools—it depends on how effectively organizations can synchronize design, analysis, verification, and manufacturing data to enable faster, more agile system-level development.