📁 Hardware

A recent analysis suggests that using Blackwell GPUs locally to run large language models (LLMs) could be more cost-effective than cloud API services. The study also includes GPU amortization costs, highlighting a quick cost break-even compared to cloud alternatives.

AMD will try to keep graphics card prices as low as possible for gamers during the DRAM shortage. Radeon GPU prices have already surged over 10%.

Leading developers of AI models from China are exploring ways to rent Nvidia's upcoming Rubin GPUs in the cloud. This option could overcome the challenges related to high costs, deployment complexity, and regulatory hurdles that limit the direct adoption of these powerful processing units.

Taiwan's IC design sector is expected to undergo a significant transformation by the end of 2025, driven by the increasing integration of artificial intelligence (AI). This evolution could lead to a split in the sector, with some companies specializing in AI-specific solutions, while others continue to focus on more traditional markets. The impact will be felt throughout the island's technological ecosystem.

According to a recent study, system-level design, energy efficiency, and total cost of ownership (TCO) are the key factors driving competition in the artificial intelligence hardware sector. Manufacturers are increasingly focusing on optimized solutions to reduce energy consumption and operating costs, which are crucial for the large-scale adoption of AI technologies.

The National Institute of Applied Research (NIAR) has launched a chip-level advanced packaging platform. The initiative aims to strengthen Taiwan's position in the semiconductor industry by enhancing manufacturing capabilities and technological innovation in advanced packaging. This strategic move could have a significant impact on the global semiconductor supply chain.

US export controls have been amended, paving the way for shipments of Nvidia GH200 and AMD MI325X chips to China. This decision could significantly impact the Chinese artificial intelligence market and access to advanced computing technologies.

Linux kernel 7.0 may mark a turning point for performance on Intel CPUs. An upcoming patch enables Intel Transactional Synchronization Extensions (TSX) by default on processors that are not vulnerable to side-channel attacks. This move promises to improve operating system performance without requiring manual configurations.

AMD EPYC 8004 "Siena" processors, focused on energy efficiency and cost optimization for servers, demonstrate significant performance increases on Linux. Recent benchmarks highlight how an up-to-date Linux software stack can further improve the performance of these chips over time.

SiFive has announced that it is adopting NVLink Fusion for its future RISC-V data center CPU designs. This will allow chips based on SiFive IP to directly connect to NVIDIA's GPU-based AI accelerators, opening new possibilities for high-performance computing and AI acceleration.

TSMC's chairman confirms strong long-term demand in the artificial intelligence sector. The company remains committed to expanding its manufacturing operations in the United States, in response to the growing demand for advanced chips for AI applications.

AMD discussed its graphics technologies, including FSR Redstone, at CES 2026. The company explained why the AI 400 lacks an RDNA 4 GPU, providing details on future strategies in graphics cards and artificial intelligence.

According to Gigabyte CEO Eddie Lin, the revenue per gigabyte of GDDR7 memory on Nvidia graphics cards will be the key factor determining product success and potential supply shortages in 2026. Nvidia's strategy will therefore depend on the profitability of GDDR7 memory.

Asus denies claims of discontinuing the RTX 5070 Ti and RTX 5060 Ti graphics cards. The Taiwanese company stated it has no plans to stop selling these models, while acknowledging that memory supply has impacted production and restocking.

Samsung plans to increase its DRAM production by 5% by 2026. This decision comes at a time of persistent supply tightness, with growing demand for artificial intelligence and data center applications. The production expansion aims to meet market needs and consolidate Samsung's position in the semiconductor industry.

Taiwan has opened an international tender for the purchase of a quantum computer. However, the US export ban remains in effect, a key element that could influence deliveries and technological choices. The initiative aims to strengthen the island's advanced computing capabilities.

US data centers, crucial for artificial intelligence, are facing significant delays in grid connection. This situation is driving the development of new power supply models to ensure the operation and growth of the sector. The delays jeopardize deliveries and the expansion of the infrastructure needed to support innovation in AI.

Taiwanese power semiconductor companies reported mixed results for 2025, amid significant shifts in global supply chains. Performance varied depending on market segments and corporate strategies adopted to address the new challenges in the sector.

Aeva secures a spot in Nvidia's autonomous driving ecosystem through its collaboration with LG Innotek. This strategic partnership strengthens Aeva's position in the autonomous vehicle sector, leveraging synergies with Nvidia and LG Innotek's expertise in advanced components. A step forward for autonomous driving.

The Trump administration has imposed a 25% tariff on high-end artificial intelligence chips, including Nvidia's H200 and AMD's MI325X, just one day after approving their export to China. This move, following a national security investigation, aims to incentivize domestic semiconductor manufacturing. The timing creates immediate complications for Nvidia, as chips destined for China must transit through the United States for testing, triggering the tariff. Exemptions are provided for US data centers and other domestic uses.