Peak Quantum, a Munich-based quantum computing startup, has secured over €5 million in total funding, including €2.2 million in pre-seed capital. The company, a spin-off from the Technical University of Munich, focuses on developing superconducting quantum processors with intrinsic hardware-level error resilience. This approach aims to simplify systems and make quantum computers more practical. Peak Quantum will also operate a European quantum chip pilot line under the EU Chips Act.
Samsung Electro-Mechanics is reportedly planning a significant expansion of its MLCC embedded substrate production line in Vietnam. This move aims to bolster AI chip packaging capabilities, a crucial component for the efficiency and performance of AI hardware, with implications for the entire supply chain and on-premise LLM deployments.
Reports suggest that Samsung's 2-nanometer process yields remain below the levels required for mass production. This scenario could have significant implications for the availability and cost of next-generation chips, which are crucial for AI and on-premise deployments, influencing hardware procurement strategies.
JCET is driving innovation in advanced AI packaging, focusing on technologies like Co-Packaged Optics (CPO) and glass substrates. These developments aim to meet the growing demand for high-performance packaging solutions, which are crucial for the efficiency and power of next-generation AI chips, with direct implications for on-premise infrastructures.
Silicio Motion has broken ground on its new Taipei headquarters, with an anticipated opening in 2030. The facility is set to become the anchor for a dual-site operational strategy, an approach aimed at strengthening the company's resilience and production capabilities. Silicio Motion is a key supplier of NAND flash controllers for global technology infrastructure, including on-premise Large Language Model (LLM) deployments.
A recent technological innovation allows for the observation of CPU transistors in operation, utilizing terahertz radiation. This capability raises significant data security concerns, as it could theoretically enable the theft of information directly from the chip while it is running. The discovery, which involves players like ASML in the context of advanced semiconductor technologies, highlights new challenges for data sovereignty and the protection of on-premise infrastructures.
The HighPoint CopprLink ecosystem promises to redefine eGPU performance, surpassing current standards. Tested with an RTX 5090 GPU, it achieved near-native performance levels. This innovative solution, however, entails an additional cost of $2,300 for complementary hardware, introducing new considerations for those seeking external graphics power with data control and sovereignty requirements.
A significant update to Mesa's RadeonSI driver introduces improved support for OpenCL 3.0 capabilities on AMD APUs and SoCs with integrated Radeon graphics. This enhancement, leveraging the Rusticl driver, precedes the Mesa 26.1 release and aims to optimize performance and functionality for developers and architects utilizing these platforms, particularly in on-premise and edge deployment contexts.
Japan is funding the creation of a national ecosystem for artificial intelligence chips, with Rapidus at the core of this strategic initiative. The company's first facility, IIM-1, is under construction in Chitose, Hokkaido, marking a significant step towards technological autonomy in the AI sector and strengthening data sovereignty.
A recent benchmark explored the performance of the MiniMax-M2.7 Large Language Model, in its NVFP4 quantized version, on a self-hosted configuration equipped with two NVIDIA RTX PRO 6000 Blackwell GPUs. The results highlight a peak aggregate throughput of over 2800 tokens per second and offer crucial insights for CTOs and architects evaluating on-premise LLM deployments, providing concrete data on decode and prefill in scenarios with varying concurrency and context lengths.
Hygon, a Chinese chipmaker, is expanding its strategy in the AI compute sector, focusing on CPU and GPU integration. This move highlights the growing importance of optimized hardware solutions for artificial intelligence workloads, with significant implications for on-premise deployments and technological sovereignty.
The increasing demand for memory in AI workloads, particularly for Large Language Models, is leading to a surprising rediscovery of DDR3 technology in the PC market. This phenomenon highlights the challenges related to the cost and availability of newer memory types, prompting companies to consider alternative solutions to optimize TCO in on-premise deployments.
Taiwanese thermal solution providers are entering a structural growth phase in the AI server market, driven by the rapid adoption of liquid cooling. This technology is crucial for managing the high heat generated by next-generation AI accelerators, ensuring optimal performance and reliability. This infrastructural evolution has significant implications for on-premise deployments, impacting TCO and compute density.
Delta Electronics reported record revenue, driven by the increasing demand for AI servers and liquid cooling solutions. This trend highlights the importance of robust and efficient infrastructure for AI workloads, a key factor for companies evaluating on-premise deployments and seeking to optimize TCO and data sovereignty.
Taiwan has launched a program involving over 200 high-end devices to cultivate talent in advanced integrated circuit design. This initiative aims to solidify the island's position as a global silicio hub, crucial for the evolution of AI hardware and on-premise deployment strategies, directly influencing the availability and efficiency of solutions for Large Language Models.
A tech enthusiast successfully ran the classic video game Doom on an Agfa Compugraphic 9000PS printer controller, hardware approximately 40 years old. This feat, enabled by the integrated Motorola 68020 processor, highlights software optimization capabilities and the challenges of deploying on resource-constrained architectures, a topic relevant even for modern workloads like on-premise LLMs.
Linux 7.0 integrates support for three new AI-specific keys on keyboards, marking an evolution beyond the single Copilot key in Windows 11. Google developed both the HID specification and the kernel patch, indicating a growing standardization of user interaction with AI systems, potentially relevant for on-premise environments and data sovereignty control.
A new PCIe adapter, dubbed "platypus," enables the installation of half-height GPUs into full-height slots while adding two M.2 slots for SSDs. Leveraging PCIe bifurcation, this solution was demonstrated with a low-profile RTX 4060 and two storage drives, offering flexibility for on-premise configurations and space optimization in compact servers. The Gigabyte WindForce GeForce RTX 5070 12GB is also mentioned.
Gigabyte unveiled its W775-V10-L1 server in Taipei, a solution based on NVIDIA GB300 that introduces two significant innovations. The system features dual PCIe Gen6 M.2 slots, offering potential improvements in I/O performance, and a dedicated liquid-cooling tray. These features aim to optimize processing capabilities and thermal management, crucial aspects for demanding AI workloads in on-premise environments.
Murata Manufacturing has announced the production of seven new MLCCs (Multilayer Ceramic Capacitors) specifically designed for the automotive sector. These components are crucial for ensuring stability and reliability in autonomous driving systems and power supply solutions, meeting the growing demands for performance and robustness required by edge AI applications and next-generation vehicles.