Quantum Motion, a London-based company specializing in silicon-CMOS spin qubits, has secured $160 million in funding. The investment is led by the European Union's new Scaleup Europe Fund, marking its first major late-stage venture commitment. This move, announced post-Brexit, highlights the EU's continued support for European quantum technologies, recognizing Quantum Motion as a key player in the quantum hardware landscape.
Silex Microsystems, a MEMS foundry backed by Bure Equity and Creades, successfully debuted on Nasdaq Stockholm. The offering, priced at SEK 81 per share and oversubscribed multiple times, valued the company at SEK 8.9 billion, with shares opening sharply higher. Prominent investors acquired approximately three-quarters of the deal.
SpaceX is investing $119 billion in a new Terafab facility in Texas, aiming to achieve independence in the production of AI-specific chips. This strategic move highlights the increasing importance of controlling the hardware supply chain for large-scale AI operations and data sovereignty.
Nvidia and Corning have formed a strategic partnership to increase optical component manufacturing capacity in the United States tenfold. This initiative aims to strengthen the AI infrastructure supply chain, crucial for data centers and on-premise Large Language Model deployments, ensuring greater resilience and control over the supply of critical technologies.
Fositek, a Taiwanese component manufacturer, is experiencing a significant increase in demand for its AI server cooling solutions. This trend highlights the critical importance of thermal management for infrastructures hosting artificial intelligence workloads, a key factor for on-premise deployments and Total Cost of Ownership (TCO) control.
A recent Counterpoint Research study places Snapdragon at the top of chipset trust rankings in India. While primarily reflecting the consumer market, this data raises questions about the perception of silicon reliability, a critical factor for enterprises evaluating Large Language Model (LLM) deployments on-premise or at the edge, where hardware choice impacts performance, security, and TCO.
An in-depth analysis explores the performance of the Qwen3.6-35B-A3B model, optimized with Multi-Token Prediction (MTP), on local hardware configurations. Initial tests show modest speed increases (6% for Q4, 2.5% for Q8) compared to 27B models, where gains were significantly higher. However, an external report indicates more substantial improvements (up to 50% for Q8) on different setups, suggesting that the optimization's effectiveness heavily depends on the hardware architecture and specific implementation.
Cambridge-based startup Nyobolt has closed a $60 million Series C funding round, achieving a billion-dollar valuation. Its success is driven by ultrafast batteries capable of charging in seconds and lasting 20,000 cycles. Contrary to common expectations, these batteries do not power cars but rather Symbotic's SymBot autonomous mobile robots, an AI robotics company listed on Nasdaq, used in warehouse logistics.
Clément Delangue of Hugging Face has shared an analysis of the 100 most popular hardware setups used on the platform. This study offers crucial insights for CTOs and infrastructure architects evaluating Large Language Model deployment, highlighting the importance of hardware choices for performance, TCO, and data sovereignty in self-hosted and on-premise contexts.
The upcoming Linux 7.2 kernel release will integrate a new power management module for AMDGPU graphics and AMDKFD compute kernel drivers. This update, expected during the June merge window, aims to better align the power management behavior of Radeon GPUs on Linux with that already present in Microsoft Windows, optimizing efficiency and performance for intensive workloads.
SpaceX, Elon Musk's space company which also includes his AI entity, xAI, is considering an initial investment of $55 billion, with a potential to grow up to $119 billion, to build a semiconductor factory in Texas. This strategic move, revealed in documents filed with Grimes County, underscores Musk's ambition to vertically integrate the production of critical hardware for his operations, including artificial intelligence.
A comparative analysis of Linux drivers for workstations examines the performance of the open-source Nouveau driver against the proprietary NVIDIA R595 solution. The test, conducted on an HP Z6 G5 A workstation, highlights the dominant position of the official NVIDIA driver for RTX (PRO) hardware, while acknowledging the continuous evolution of Nouveau awaiting the Nova kernel driver. Driver choice is crucial for those managing on-premise deployments, impacting control, performance, and TCO.
SpaceX has filed paperwork for the construction of a new semiconductor fabrication facility, named Terafab, in rural Texas. The projected investment for this new structure is approximately $55 billion. Combined with an existing packaging operation in Bastrop, SpaceX's total commitment to chip manufacturing in Texas could reach $119 billion, marking a strategic expansion into supply chain control.
AMD is expanding and specializing its EPYC processor lineup, with the Zen 7 architecture already under development. The goal is increased customization to meet the evolving demands of AI and hyperscale workloads, both in cloud and on-premise environments. This strategy aims to optimize performance and efficiency for a wide range of applications.
AMD has released a new update for its open-source ROCDXG library, enhancing ROCm compatibility within the Windows Subsystem for Linux (WSL). This expansion aims to extend ROCm support on WSL2 to a broader range of Ryzen processors, offering developers a more robust platform for AI and HPC application development in local environments.
Apple has quietly removed the 128GB unified memory configuration from the Mac Studio, reducing the maximum capacity to 96GB. This decision, affecting the Early 2025 model, is attributed to supply constraints and a surging demand for local AI processing capabilities. The reduction in maximum available memory raises questions for developers and enterprises aiming for on-premise Large Language Models deployment, highlighting trade-offs in adopting self-hosted solutions.
Intel engineers are preparing a significant update for the Linux kernel graphics drivers, targeting kernel version 7.2. The main new feature is the introduction of "Panel Replay Tunneling," a functionality aimed at improving graphics management and energy efficiency. This development underscores the importance of robust and updated software infrastructure for on-premise deployments, ensuring stability and maximizing hardware performance.
The artificial intelligence landscape is shifting from model training to serving, creating new avenues for chip startups. The heterogeneity of inference workloads, which demand a variable mix of compute, memory, and bandwidth, allows new players to specialize. Some companies focus on disaggregated architectures for prefill and decode, while others propose integrated or innovative solutions like optical accelerators.
The CEO of AEM highlights how artificial intelligence is radically transforming the semiconductor testing sector. This evolution presents new challenges and opportunities for the industry, driving the adoption of more efficient and automated solutions, with significant implications for deployment infrastructure and data sovereignty.
Vanguard International Semiconductor (VIS) is joining the CoWoS supply chain, crucial for AI chips. An interposer foundry in Singapore, backed by TSMC, strengthens the production of essential components for high-bandwidth memory integration. This development is significant for the availability of advanced AI hardware, impacting on-premise deployment strategies and technological sovereignty.