London-based startup Fractile has secured $220 million in funding, led by Accel, with Pat Gelsinger joining as an angel investor. The capital will be used to bring its innovative inference chips, which integrate compute and memory on the same die, into production. This development follows early discussions with potential customer Anthropic, underscoring market interest in dedicated AI hardware solutions.
UK startup Fractile has closed a $220 million Series B round to develop next-generation inference hardware. The company aims to resolve the growing bottleneck related to the time and cost of producing useful outputs at scale for Large Language Models, addressing memory bandwidth limitations in current architectures.
AMD has announced the expansion of its Ryzen 9000 PRO processor line with six new SKUs. For the first time, these workstation-oriented chips integrate 3D V-Cache technology, promising significant performance improvements. With TDPs reaching 170W, the new models will be available through OEM partners later this year, offering powerful options for professional environments requiring high computing capabilities.
Intel has released a new version of its open-source Compute Runtime, version 26.18.38308.1. The update extends OpenCL and Level Zero support across the company's integrated and discrete graphics hardware, introducing Xe3P enablement and support for future Nova Lake P architectures. This release is crucial for developers and infrastructure architects aiming to optimize Intel GPU performance in on-premise environments, especially for AI and LLM workloads.
L&T Semiconductor Technologies and Synopsys have signed a strategic multiyear agreement for the design of AI-enabled power modules. This collaboration aims to develop fundamental hardware solutions for AI infrastructure, with significant implications for the efficiency and reliability of high-performance computing systems, essential for Large Language Models workloads and other artificial intelligence applications.
Samsung Foundry is experiencing a significant resurgence, driven by the increasing demand for artificial intelligence chips. The adoption of HBM4 technology and advancements in 4-nanometer manufacturing processes are key factors redefining its position in the semiconductor market, with direct implications for on-premise LLM deployment strategies.
Google has unveiled Googlebook, a new line of premium laptops that marks a shift beyond Chromebooks. These devices, arriving this autumn, integrate Android with Gemini at the operating system level, transforming the cursor into an AI agent. This move reflects Google's view that a browser-only system is no longer sufficient for current needs, focusing on pervasive artificial intelligence.
Google is set to launch Googlebooks, a new line of Android-powered laptops deeply integrated with Gemini Intelligence. These devices, expected later this year, introduce innovative features like the "Magic Pointer," marking an evolution in the company's approach to personal computing, while Chromebooks remain on the market.
Managing noise in high-performance computing systems, such as those used for AI workloads, presents a complex challenge. Components like cases, fans, and All-in-One (AIO) liquid cooling systems are crucial for heat dissipation but are also primary sources of noise. This aspect becomes particularly relevant in on-premise environments, where the integration of AI hardware requires careful evaluation of trade-offs between performance, thermal efficiency, and acoustic impact.
China's next-generation Loongson 3B6600 (CPU) and 9A1000 (GPU) chips aim to match the performance of Intel 12th Gen and AMD RX 550 by 2027. This development highlights China's ambition to strengthen its technological autonomy in the semiconductor sector, with implications for data sovereignty and on-premise deployment strategies.
Applied Materials and TSMC have announced a collaboration at the EPIC Center to accelerate the development of chips dedicated to artificial intelligence. This initiative aims to optimize manufacturing processes and foundational technologies, with significant implications for the efficiency and availability of AI hardware, crucial for companies evaluating on-premise deployments and data sovereignty management.
VSO Electronics is targeting significant growth, driven by the increasing demand for specialized cables for AI infrastructure. The company also plans to activate a new in-house leak-detection line by late 2026, consolidating its production capabilities and quality control in a rapidly evolving market.
AcBel Polytech, OmniOn, and Kinpo Group have formed a strategic partnership to develop power supply solutions specifically for the growing artificial intelligence market. This initiative aims to address the demand for robust and efficient infrastructure, essential for intensive LLM workloads and on-premise deployments, where power efficiency and thermal management are critical factors for TCO.
A user demonstrated an open-loop tap water cooling method for a DGX system, keeping GPUs below 68°C at 95% utilization. The setup handles a Qwen3.5-122b-a10B LLM with Q6_K precision, utilizing 110 GB of memory and an 80k context window, achieving 18.77 tokens/second for continuous vision analyses. This highlights the challenges and creative solutions for on-premise AI deployments.
The open-source "R300g" driver for ATI (AMD) Radeon R300 and R500 series GPUs, dating back over two decades, is set to receive a significant code restructuring in 2026. This effort, led by a single community developer, highlights the longevity and dedication of open-source projects, ensuring support and improvements even for hardware considered obsolete.
Rumors suggest AMD is preparing an entry-level RDNA 4 GPU, the RX 9050, featuring 8GB of VRAM and 2048 cores. This potential addition to the Radeon lineup could offer new options for lighter AI workloads and on-premise deployments, balancing cost and capability for specific inference needs.
AMD has released significant updates for its AMDGPU kernel driver on Linux, introducing support for HDMI 2.1 Fixed Rate Link (FRL) and Display Stream Compression (DSC). These enhancements enable higher resolutions and refresh rates, solidifying the open-source driver's position as a robust solution for AMD hardware in environments demanding advanced graphics performance and infrastructural control.
System76 has unveiled the Thelio Major workstation, a high-end Linux system built entirely on AMD hardware. Featuring AMD Ryzen Threadripper 9000 series processors and Radeon AI PRO R9700 graphics, this machine offers a powerful, open-source solution ideal for developers and professionals requiring high performance for intensive workloads, including those related to artificial intelligence. It provides complete control over the operating environment and data sovereignty.
The technology industry, particularly in the field of artificial intelligence, is evolving at an unprecedented pace. For CTOs and infrastructure architects, keeping up means understanding the implications of new hardware developments and deployment strategies. This requires an in-depth analysis of on-premise options, costs, and data sovereignty, all crucial aspects for informed decisions.
Intel and SK Hynix shares surged following reports of a potential strategic chip packaging partnership. The collaboration would involve SK Hynix testing Intel's 2.5D EMIB technology for High Bandwidth Memory (HBM) integration. This move highlights the increasing importance of advanced packaging technologies for AI and LLM applications, with significant implications for performance and efficiency in next-generation hardware.