Data-center power systems are entering a new phase as operators move toward 800V HVDC and denser DC-DC architectures, raising demand for high-voltage, low-loss, and high-frequency power devices. The shift is pushing suppliers toward integrated control, protection, and module solutions, while Taiwan manufacturers weigh how quickly they can adapt.
STMicroelectronics expects rapidly expanding AI infrastructure sales and stronger automotive demand to support growth into 2027, but its largest gross-margin gains will depend on customer approvals allowing older factories to close, CFO Lorenzo Grandi said at Citi's Global TMT Conference on September 9.
Analog Devices has agreed to acquire Alif Semiconductor in a US$1.35 billion all-cash deal, a move that could broaden access to faster, more energy-efficient edge AI systems for industrial, defense, health, and consumer applications worldwide. The deal underscores a shift toward local, real-time computing beyond data centers.
Arm unveiled its second-generation mobile Compute Subsystem (CSS), Arm CSS for Mobile 2, at its annual Arm Everywhere China event in Shanghai on September 8, further integrating AI computing capabilities across the CPU, GPU, and system interconnect while strengthening its presence in the Chinese market.
Taiwan's Ministry of Economic Affairs (MOEA) and National Science and Technology Council (NSTC) will hold a joint ministerial meeting in September to discuss subsidy applications submitted by domestic IC design firms for advanced chip development. If the applications are approved, the local IC design sector will receive a boost in its bid to raise the share of IC design industry output value generated by advanced-process chips from 45% to 47%.
For Samsung Electronics' two loss-making logic businesses, OpenAI's declaration that it is jointly producing and researching next-generation chips with the Korean company matters less as a new logo on the customer list than as a sign of where Samsung's advanced wafers actually go.
Chinese AI chip developer Cambricon said on September 8 that it has formally joined the PyTorch Foundation as a platinum member, the organisation's highest membership tier, securing a seat on its governing board.
SEMIFIVE has begun mass production of HyperAccel's data center inference accelerator using Samsung Foundry's 4nm process, marking the Korean custom-silicon company's first large-scale production program on the node as Samsung's Pyeongtaek SF4 line remains at full utilization.
HCLTech has opened a new semiconductor laboratory in Bengaluru, a move that could matter for chipmakers and device makers worldwide as they look for faster testing, qualification, and failure analysis. The facility is designed to shorten development cycles, support production quality, and strengthen India's role in the global semiconductor ecosystem.
Cambricon has joined the PyTorch Foundation as a Platinum member, taking one seat on its Governing Board and one on its Technical Advisory Council. The Linux Foundation announced the membership on September 8 at the combined KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 in Shanghai, where Alibaba Cloud also joined at the Platinum tier and Ant Group at Gold.
Arm is stepping up its role in AI infrastructure as cloud providers, original equipment makers, and software partners build systems for agentic AI, a shift that could affect data centers worldwide. The company said its latest Neoverse Compute Subsystems N4 and Arm AGI CPU aim to give customers more choice across performance, efficiency, and silicon design.
Capacity for advanced semiconductor processes and packaging remains tight, while mature-node chips required for AI-related peripheral applications are also facing mounting supply pressure. IC design houses and semiconductor industry sources generally expect the current capacity constraints to last longer than the chip supply crunch that followed the COVID-19 pandemic.
One of Japan's few private technology companies valued above US$1 billion is preparing to abandon a decade of engineering independence, because designing an AI chip is no longer the expensive part — manufacturing one at volume is.
Qualcomm's multi-generation agreement to build customized AI inference silicon and optical interconnects for Amazon gives the company something its data center roadmap has lacked since it re-entered the market: a named, US-scale hyperscaler willing to put Qualcomm parts into a production fleet. The commercial value was not disclosed; the strategic value is the point. Qualcomm has spent the past year buying its way into AI infrastructure and telling investors it belongs there. Amazon Web Services (AWS), the largest buyer of AI compute in the market, has now agreed.
Chinese GPU developer Biren Technology has teamed up with Shanghai University to establish an Integrated Circuit Intelligent Chip Joint Laboratory, expanding their cooperation into joint research on AI-chip optical interconnects and advanced thermal management.
Arm unveiled new platforms and ecosystem programs for cloud data centers, Physical AI and mobile devices on September 8, broadening its role beyond CPU intellectual property into a wider AI computing platform. The company said the push is aimed at supporting agentic AI, which requires systems to plan tasks, retrieve data, coordinate models and take actions on behalf of users.
ITRI held its 15th Laureate Ceremony on September 7, honoring three new laureates in 2026: TSMC executive vice president and co-COO Y.P. Chyn, Realtek chairman Sun-Chien Chiu, and superintendent of Taipei Veterans General Hospital Wei-Ming Chen.
MA-tek, a semiconductor testing and analysis company, reported August revenue of NT$565 million, up 17.78% from a year earlier and marking a record high for the sixth straight month. Revenue for the first eight months reached NT$4.196 billion, up 17.66% from the same period in 2025.
This week, DIGITIMES heads to Shenzhen to cover an expo showcasing the world of optoelectronics, a class of technologies that serves as the link between electricity and light. In today's era of pervasive digital connectivity, from smart glasses to cars to phones, this event is expected to give us a glimpse into the hardware supply chain behind these devices, and increasingly behind the global AI infrastructure build-out.
The 27th China International Optoelectronic Exposition (CIOE 2026) will run September 9-11 in Shenzhen, bringing together more than 5,000 exhibitors from over 30 countries and regions as AI infrastructure, high-speed optical communications and AR/VR applications reshape the optoelectronics industry.
Malaysia is evaluating Huawei Technologies' Ascend 910C accelerators as the compute backbone of its MYR2 billion (US$494 million) sovereign AI initiative, which is designed to keep national data under domestic control, Bloomberg reported on Sept. 7, citing people familiar with the matter. The number of chips under consideration remains unclear, and no supplier decision has been announced.
If Taiwan's government is determined enough, finding funding and land for the Industrial Technology Research Institute (ITRI) or the National Institutes of Applied Research (NIAR) is not necessarily the hardest part. The more difficult question comes later: who will shoulder the enormous and recurring cost of keeping advanced semiconductor research infrastructure running?
Huawei's Kirin 9050 Pro is the first commercial test of its Tau Scaling Law, with the company claiming a 55% increase in transistor density at the same process node while cutting NPU, GPU and CPU performance-core power by 66%, 58% and 41%, respectively.
Indian chip design company IndieSemiC is targeting shipments of 3 to 5 million locally made Wi-Fi modules in FY27-28, betting that security requirements for CCTV equipment will encourage camera makers to replace imported connectivity modules with products sourced in India.
