📁 Hardware

This Hardware archive tracks the practical side of local AI infrastructure: GPUs, NPUs, mini PCs, edge accelerators, memory bandwidth, and power efficiency tradeoffs that directly impact LLM inference quality. We prioritize benchmark-backed updates and deployment notes useful for real build decisions, from compact home labs to enterprise pilot clusters. Use this stream to compare total cost of ownership, thermal constraints, and model-fit scenarios across current devices, then deepen with our hardware pillar guide and connected LLM coverage.

GlobalFoundries is expanding its CPU intellectual property (IP) portfolio by acquiring assets from Synopsys. This strategic move repositions the company, transforming it from a mere chipmaker into a comprehensive IP and manufacturing partner in the semiconductor industry. The acquisition includes ARC and RISC-V technologies, crucial for developing new hardware solutions.

2026-01-15 Fonte

The Trump administration has formalized a 25% tariff on Nvidia's H200 chip sales to China. The tariff applies to specific semiconductors, marking an escalation in trade restrictions between the United States and China in the technology sector. This move could significantly impact Nvidia's market strategies and Chinese access to advanced artificial intelligence technologies.

2026-01-15 Fonte

OpenAI launches a new Request for Proposal (RFP) to strengthen the U.S. artificial intelligence supply chain. The initiative aims to accelerate domestic manufacturing, create jobs, and scale AI infrastructure.

2026-01-15 Fonte

Taiwan is accelerating its efforts to build a silicio photonics ecosystem. The initiative aims to position the island as a leader in this emerging field, crucial for future high-performance data communication and computing technologies. A significant impact on the local semiconductor industry is expected.

2026-01-15 Fonte

Chinese startup DeepSeek has unveiled a new technique called "Engram" to train large AI models, overcoming limitations imposed by US export restrictions on advanced chips. This approach aims to reduce reliance on GPU memory, enabling the handling of more complex models with less powerful hardware. The technique separates basic facts from complex calculations, optimizing the use of available computing power.

2026-01-15 Fonte

Nvidia introduced its next-generation AI chip platform, which will support a new robotaxi alliance. Key partners include Lucid Group, Nuro, and Uber, marking a significant step in the evolution of autonomous driving and robotics.

2026-01-15 Fonte

Chinese outfit Zhipu AI claims it trained a new model entirely using Huawei hardware, and that it’s the first company to build an advanced model entirely on Chinese hardware. If confirmed, this would be a significant achievement for the Chinese tech industry, which aims to reduce its reliance on foreign technologies. Details on the amount of hardware used were not provided.

2026-01-15 Fonte

A faulty 16-pin connector sidelined an expensive NVIDIA H200 Hopper GPU, crucial for data centers. A skilled technician successfully repaired the power connector, averting costly replacements and data loss. The issue involved damaged sense pins, resolved through precise soldering.

2026-01-15 Fonte

SK Hynix is investing nearly $13 billion to build a new facility dedicated exclusively to the production and testing of next-generation HBM memory. The facility, located in South Korea, aims to consolidate SK Hynix's leadership in the rapidly expanding market for high-bandwidth memory, essential for artificial intelligence applications.

2026-01-14 Fonte

The U.S. has formalized export rules for H200 and MI325X AI GPUs to China. The new regulations include several limitations on the total number of GPUs that can be shipped, with U.S. supply taking priority. Shipments to China will be evaluated on a case-by-case basis, but large volumes from AMD and Nvidia are not expected.

2026-01-14 Fonte

AI chip design is pushing advanced chip packaging to its limits. The increasing size and complexity of AI accelerators and HPC devices are making packaging a more critical factor. Workarounds exist for the limits of 2.5D packaging, but their viability is still years away.

2026-01-14 Fonte

Nvidia's DLSS 4.5 Super Resolution feature is now available to all users through the Nvidia app, having left beta. The update includes support for over 400 titles and introduces new presets, promising an enhanced and smoother gaming experience.

2026-01-14 Fonte

Phison's aiDAPTIV+ stack enables large Mixture of Experts AI models and agentic AI workloads to run on client systems with limited memory capacity, purportedly enabling 10x faster AI inference. Nvidia, AMD, MSI, and Acer systems were demoed showcasing aiDAPTIV+ capabilities.

2026-01-14 Fonte

Seagate has launched three new 32 TB hard drives using CMR (Conventional Magnetic Recording) technology. The new models belong to the SkyHawk AI and Exos families, starting at $699. All hard drives offer data transfer speeds of up to 285 MB/s and are covered by a five-year warranty.

2026-01-14 Fonte

Chinese customs officers have allegedly been instructed to block the import of Nvidia H200 chips, effectively banning these AI processors from entering the country. Sources indicate that Beijing might allow the import of these AI GPUs only under 'special circumstances'.

2026-01-14 Fonte

SK Hynix demonstrates its 48 GB HBM4 memory, designed to accelerate the performance of artificial intelligence applications. The memory features a 2,048-bit interface and speeds up to 10 GT/s, promising a significant increase in processing the most demanding workloads. This development underscores SK Hynix's commitment to innovation in the high-performance memory sector.

2026-01-14 Fonte

Sources suggest Beijing is limiting Nvidia H200 GPU purchases to entities meeting undefined "special circumstances." University R&D labs may be prioritized. This highlights geopolitical tensions in the semiconductor industry and the strategic importance of GPUs for technological innovation.

2026-01-14 Fonte