National Yang Ming Chiao Tung University in Taiwan has announced the creation of a new non-toxic blue-light material. This innovation could represent a significant step towards developing 3D displays that do not require special glasses, opening new frontiers for visual interaction and the visualization of complex data across various sectors.
Artificial intelligence is redefining the role of Central Processing Units (CPUs) in IT infrastructure. Recent statements from AMD, via CEO Lisa Su, highlight how AI is bringing CPUs back into focus, influencing deployment strategies and TCO considerations for AI workloads.
Foxconn reported revenues approaching $95 billion in the first four months of the year. This growth is significantly driven by the demand for AI server racks, a segment that fuels the company's financial outlook through the second quarter of 2026. This trend highlights the increasing importance of dedicated AI hardware for major manufacturers and its implications for on-premise deployment strategies.
A recent experiment showcased a significant performance boost in Large Language Model (LLM) inference on AMD Strix Halo hardware, leveraging `llama.cpp` with Multi-Token Prediction (MTP) support. The setup, featuring a system with 128GB of DDR5 at 8000MHz, achieved speeds between 60 and 80 tokens/s, nearly doubling performance compared to execution without MTP. These results highlight the potential of software optimization for self-hosted LLM deployments.
After nearly a decade, the SPEC consortium has introduced the SPEC CPU 2026 benchmark suite. This new version is set to redefine CPU performance evaluation standards, offering an updated perspective on the efficiency and power of modern AMD, Intel, and NVIDIA processors. The update is crucial for those designing on-premise infrastructures.
Astera Labs has introduced a high-speed connectivity solution for rack-scale AI systems, positioning itself as an alternative to Nvidia's NVSwitch. The technology promises compatibility with a wide range of accelerators, offering greater flexibility and potential benefits for on-premise deployments aiming to avoid vendor lock-in and optimize TCO.
QuantWare has closed a €152 million Series B funding round, the largest ever for a deeptech company or quantum processor firm in the Netherlands. The funds will be used to build the world's largest open-architecture quantum processor fab in Delft, with support from investors including Intel Capital and In-Q-Tel.
QuantWare has closed a €152 million Series B funding round. The capital will support the development and industrial-scale production of its quantum processors, including the VIO-40K™ architecture for high qubit capacity and the construction of KiloFab, a dedicated manufacturing facility in the Netherlands. The goal is to accelerate the transition of quantum computing from research to large-scale commercial deployment, strengthening Europe's position in the sector.
Intel has announced the appointment of Alex Katouzian, a former Qualcomm executive with 25 years of experience in mobile, compute, and extended-reality sectors. Katouzian will lead the new Client Computing and Physical AI group, reflecting Intel's strategy to attract talent from rival companies to restructure and enhance its core competencies. This move underscores Intel's commitment to distributed AI and local processing, crucial for on-premise and edge deployments.
A consumer PC bundle featuring an RTX 5080, 64GB of RAM, and a 9850X3D CPU raises questions about its suitability for on-premise LLM workloads. While such configurations can offer a starting point for local inference of smaller models, it's crucial to evaluate their limitations in terms of scalability, VRAM, and infrastructure requirements for enterprise deployment, highlighting the trade-offs compared to dedicated enterprise solutions.
A practical analysis reveals that a system equipped with two NVIDIA GeForce RTX 3090 GPUs, dedicated to Large Language Model inference, draws approximately 760W at the wall under load. This data, measured in a self-hosted context, offers crucial insights for CTOs and infrastructure architects evaluating operational costs and power requirements for on-premise deployments.
An enthusiast has faithfully recreated the Apple Lisa, the first commercial computer with a graphical user interface, using an FPGA board. The LisaFPGA project is not merely a tribute to computing history but also highlights the potential of Field-Programmable Gate Arrays as flexible hardware solutions for emulation and custom architecture development, relevant for those evaluating on-premise deployments and control over their technology stack.
Tesla is implementing a dual sourcing strategy for its AI5 chip, with Samsung among the partners, though production weight may not be equal. This move highlights the increasing importance of supply chain diversification for AI silicio, a critical factor for companies building on-premise infrastructure and seeking control over costs and performance.
Ardentec, a semiconductor testing company, has announced the commencement of testing activities for its AI-dedicated ASICs at its Longtan plant, scheduled to begin in the third quarter of 2026. This move highlights the growing importance of specialized AI chips and their implications for on-premise deployments, offering new perspectives in terms of efficiency and infrastructure control for companies managing AI workloads.
A recent DIGITIMES report suggests South Korea is exploring a "memory-led" strategy for artificial intelligence. This move indicates a potential alternative or competitive approach to Nvidia's current leadership in the AI hardware market, focusing on the development and integration of advanced memory technologies for AI workloads.
German deeptech eleQtron has secured €57 million in Series A funding, led by Schwarz Digits, to accelerate the industrial scaling of its trapped-ion quantum processors. The investment aims to transition its proprietary MAGIC technology, which uses microwave qubit control, from laboratory development to industrial deployment, expanding production capacity and system access, including cloud-based offerings.
The mobile Application Processor (AP) market is increasingly shaped by memory dynamics, creating a divide between Apple's stable performance and the volatility of the Android segment. The emergence of agentic AI is identified as the next catalyst, set to redefine hardware requirements and deployment strategies, especially for on-device AI inference.
Intel has announced the appointment of a former Qualcomm executive to lead its PC and physical AI unit. This strategic move underscores the company's commitment to developing artificial intelligence capabilities directly on client devices and at the edge, a critical area for data sovereignty and optimizing operational costs in distributed deployment scenarios.
MediaTek is consolidating its position in the ASIC sector, with an expansion that could lead the company to achieve 60% market share in the segment. This development reflects a strategic evolution towards higher-value-added solutions, with significant implications for on-premise AI infrastructures and Large Language Models workloads.
The tech industry is accelerating the development of DDR6 server memory, a strategic move to meet the growing demands of next-generation AI workloads. This evolution is crucial for on-premise deployments, where memory capacity and bandwidth directly influence performance, TCO, and data sovereignty, offering new opportunities for robust and scalable self-hosted architectures.