AI Hardware & Infrastructure

2026-04-02 • DigiTimes

IBM and Arm Partner on Dual-Architecture Hardware to Expand Enterprise AI Deployment

IBM and Arm have announced a collaboration focused on developing dual-architecture hardware. This initiative aims to broaden the deployment capabilities of artificial intelligence within enterprises, offering more flexible and performant solutions. T...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-02 • DigiTimes

Asahi Kasei Enters AI Chip Fiberglass Market, Challenging Nittobo

Asahi Kasei has announced its entry into the AI chip fiberglass market, a critical sector for advanced hardware component manufacturing. This move aims to challenge Nittobo's dominant position, signaling an intensification of competition in the AI ma...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-02 • DigiTimes

Nvidia Invests $2 Billion in Marvell for NVLink Fusion Integration with ASICs

Nvidia has announced a $2 billion investment in Marvell, aiming to integrate NVLink Fusion technology directly into ASICs. This strategic move seeks to enhance interconnection capabilities for custom chips, accelerating the development of optimized h...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-02 • DigiTimes

Chinese Companies Capture Nearly 41% of Domestic AI Accelerator Server Market

Chinese enterprises have secured a significant market share, almost 41%, in the domestic AI accelerator server sector. This highlights a growing local capability in providing critical infrastructure for Large Language Models (LLM) workloads and other...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-02 • DigiTimes

The Arm Architecture Redefines AI Servers: Towards a Post-x86 Era

Hyperscalers are re-engineering AI server CPUs by adopting the Arm architecture, signaling a potential shift from the x86 era. This transition promises greater energy efficiency and flexibility, with significant implications for TCO and data sovereig...

#Hardware #LLM On-Premise #DevOps

2026-04-02 • DigiTimes

Huawei's AI Strategy: Infrastructure at the Core

Huawei's 2025 annual report, featuring insights from Meng Wanzhou, highlights the company's AI strategy, which begins with foundational infrastructure. This vision underscores the importance of robust, scalable architecture to support the development...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-02 • DigiTimes

Microsoft Invests $6.5 Billion in Southeast Asia AI Buildout

Microsoft has announced a $6.5 billion investment to boost its artificial intelligence infrastructure in Southeast Asia, with a specific focus on Singapore and Thailand. This strategic move underscores the region's growing importance as a tech hub an...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-02 • Phoronix

AMD GPU Driver Optimizations Arriving with Linux 7.1

AMD is introducing new optimizations for its GPU drivers, including the DC Idle Manager and Multi-SDMA Engine, slated for the Linux 7.1 kernel. These updates aim to enhance the efficiency and performance of AMD graphics cards, a crucial aspect for on...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • TechCrunch AI

Meta: Hyperion AI data center powered by ten gas plants

Meta is building its next AI data center, named Hyperion. This critical infrastructure for artificial intelligence workloads will be powered by ten new natural gas plants. Meta's energy choice for a project of this magnitude raises questions about pr...

#Hardware #LLM On-Premise #DevOps

2026-04-01 • TechCrunch AI

Cognichip Raises $60M to Advance AI-Designed Chips for AI

Cognichip has secured $60 million in funding to pursue an innovative approach: leveraging artificial intelligence to design the very chips that power AI applications. The company aims to revolutionize the semiconductor industry by promising to reduce...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • The Next Web

Valar Atomics: $450 Million for AI Powered by Nuclear Energy, Rethinking Energy Scale

Isaiah Taylor, 27, has raised $450 million for Valar Atomics. The initiative aims to connect artificial intelligence with nuclear power, addressing the size problem of current reactors. The goal is to develop energy solutions better suited to the gro...

#Hardware #LLM On-Premise #DevOps

2026-04-01 • Tom's Hardware

DRAM and NAND: Prices Soar in Q2 Due to AI Server Demand

According to Trendforce, DRAM and NAND Flash prices are set for significant increases in Q2, with projected jumps of 63% and up to 75% respectively. These increases follow substantial hikes in Q1 and are attributed to the surging demand for AI server...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • Tom's Hardware

Gigabyte X870E Aorus Xtreme AI Top: The Hardware Foundation for On-Premise AI

The Gigabyte X870E Aorus Xtreme AI Top positions itself as a flagship motherboard designed for high-performance systems. Its architecture is relevant for building AI workstations or servers in self-hosted environments, where stability, connectivity, ...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • DigiTimes

Fujitsu and Rapidus: Japan's 1.4nm AI Chip Production Takes Shape

Fujitsu has announced plans for the production of cutting-edge AI chips, based on 1.4-nanometer technology. Manufacturing will take place in Japan, in collaboration with Rapidus, at the company's first fab in Chitose, Hokkaido. Operations are schedul...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • DigiTimes

Silicio Photonics and Advanced Packaging: Pillars for Future AI

Recent discussions at Touch Taiwan highlighted the increasing importance of Silicio Photonics (SiPh) and advanced packaging. These technologies are considered crucial for overcoming current hardware limitations and enabling the next generation of AI ...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • DigiTimes

SDI targets AI and heat spreader growth

SDI, a technology sector player, is directing its growth strategies towards artificial intelligence and heat spreader development. This move reflects the increasing demand for advanced thermal solutions, crucial for managing the heat generated by int...

#Hardware #LLM On-Premise #DevOps

2026-04-01 • DigiTimes

Micron Reportedly Developing Stacked GDDR to Meet AI Memory Demand

Micron is reportedly developing a new generation of GDDR memory using stacked technology to address the increasing demands of AI workloads. This innovation is crucial for the evolution of infrastructures hosting Large Language Models, directly impact...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • DigiTimes

Nvidia and Marvell: The $2 Billion Bet Redefining AI Alliances

Nvidia has invested $2 billion in Marvell, transforming a potential rival into a strategic partner. This move highlights the importance of collaborations for AI infrastructure, with significant implications for enterprises evaluating on-premise deplo...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • DigiTimes

Nvidia Aims for Full AI Stack Ownership with Three-System Strategy

Nvidia is expanding its offerings beyond GPUs, aiming to provide comprehensive AI solutions. This strategic move, based on a three-system approach, seeks to consolidate control over the entire AI pipeline, from computation to software. The goal is to...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • DigiTimes

China GPU Maker Biren Triples Revenue on AI Data Center Demand

Chinese GPU manufacturer Biren has reported impressive revenue growth, tripling its earnings due to increasing demand from artificial intelligence data centers. This trend highlights the strong expansion of the AI hardware market, with a particular f...

#Hardware #LLM On-Premise #Fine-Tuning

2026-04-01 • DigiTimes

Arm and Tesla Reshape AI Chip Market: Impact on Supply Chains and Memory

The AI chip landscape is undergoing a profound transformation, driven by the rise of Arm architecture and custom silicio development strategies from companies like Tesla. These shifts are redefining global supply chains and fueling a surging demand f...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • Tom's Hardware

Tryx Stage 360 AIO: The All-in-One Approach for On-Premise AI Infrastructure

The Tryx Stage 360 AIO is presented as an All-in-One solution promising a distinctive user experience, focused on design and quiet operation. For companies evaluating on-premise Large Language Model (LLM) deployment, adopting integrated systems can o...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • Tom's Hardware

Fujitsu Targets 1.4nm AI Chip: Design and Production Entirely in Japan with Rapidus

Fujitsu has announced its intention to develop a dedicated artificial intelligence chip, featuring a 1.4-nanometer manufacturing process. The project stipulates that both the design and manufacturing of the semiconductor will take place entirely in J...

#Hardware #LLM On-Premise #DevOps

2026-03-31 • Phoronix

Linux 7.1: The Evolution of Rust Graphics Drivers and Implications for AI Infrastructure

Recent Rust graphics driver changes for Linux 7.1, ahead of the April merge window, introduce new abstractions and infrastructure improvements. These developments aim to make drivers more robust and efficient, a crucial aspect for companies managing ...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • The Next Web

Oracle Cuts Thousands of Jobs to Fund AI Data Centers

Oracle is undergoing a significant workforce reorganization, with estimates suggesting up to 30,000 layoffs. The goal is to free up an estimated $8-10 billion to finance massive investments in AI infrastructure and data centers. These decisions, affe...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • Phoronix

Intel Panther Lake & Linux AI/LLM Debates Dominated Q1

The first quarter saw intense activity within the Linux landscape, with upcoming Intel Panther Lake processors and discussions surrounding Large Language Models (LLM) and artificial intelligence taking center stage. These topics generated significant...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • The Register AI

Agentic AI: Arm calls for new CPUs, Intel pushes back

Arm and Nvidia have unveiled specific CPUs designed to run agentic AIs, such as OpenClaw, suggesting a need for dedicated architectures. This view, however, is challenged by Intel, whose Data Center chief does not believe a radical shift in CPU desig...

#Hardware #LLM On-Premise #DevOps

2026-03-31 • Tech.eu

Nebius Announces 310 MW AI Mega Data Center in Finland

Nebius, a European AI infrastructure company, has announced the construction of a 310 MW data center in Lappeenranta, Finland, expected to be operational by 2027. The facility will be one of Europe's largest dedicated AI data centers, used for traini...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • The Next Web

Microsoft Commits Over $1 Billion to Cloud and AI Infrastructure in Thailand

Microsoft has announced an investment exceeding $1 billion in Thailand between 2026 and 2028. The initiative aims to bolster the country's cloud and AI infrastructure, encompassing data center construction, cybersecurity enhancement, the development ...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • DigiTimes

MediaTek and Airoha Strengthen Open Source Platform for Edge AI

MediaTek and Airoha are intensifying their collaboration on an open-source platform for the telecommunications sector. The initiative aims to compete with established players like Broadcom and Qualcomm, focusing specifically on developing solutions f...

#Hardware #LLM On-Premise #DevOps

2026-03-31 • DigiTimes

Lens Technology Shifts Focus to AI Servers, Robotics, and Aerospace

Lens Technology, known for its iPhone component manufacturing, is expanding its operations. The company is now concentrating on strategic sectors such as artificial intelligence servers, robotics, and aerospace. This move marks a significant diversif...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • ServeTheHome

Gigabyte Showcases NVIDIA Vera Rubin Platforms and More at GTC 2026

At GTC 2026, Gigabyte unveiled its latest hardware innovations, with a particular focus on new platforms built around the NVIDIA Vera Rubin architecture. These next-generation systems and components are designed to tackle the most intensive Large Lan...

#Hardware #LLM On-Premise #DevOps

2026-03-31 • DigiTimes

Chinese GPU Maker Moore Threads Secures $91 Million AI Cluster Order

Chinese GPU manufacturer Moore Threads has secured a $91 million order for an AI cluster. This deal highlights the increasing demand for dedicated artificial intelligence infrastructure and the emerging role of new players in the global LLM hardware ...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • DigiTimes

AI Computing Shifts from Training to Inference: Heterogeneous Architectures Take Center Stage

The AI computing landscape is undergoing a significant transformation, with a growing emphasis on model inference over training. This shift is accompanied by the widespread adoption of heterogeneous architectures, combining various hardware types to ...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • DigiTimes

CPU Resurgence Reshapes AI Chip Demand: Terafab Funding Questions Emerge

The AI chip market is undergoing a transformation, with an unexpected resurgence of CPUs beginning to redefine hardware requirements for artificial intelligence. This trend raises questions about future investments in manufacturing infrastructures li...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-31 • DigiTimes

Nvidia: Vera Rubin Design Unfinalized, Focus on Supply Chain Diversification

Nvidia is still finalizing the design of its "Vera Rubin" compute tray. This development phase coincides with a corporate strategy aimed at diversifying its supply chain. Nvidia's move highlights the importance of mitigating risks associated with the...

#Hardware #LLM On-Premise #DevOps

2026-03-31 • DigiTimes

The Rise of Custom Chips: Taiwan Responds to ASIC Demand for AI

The increasing demand for custom chips, known as ASICs, is prompting Taiwanese firms to strengthen their presence in this market segment. This trend reflects the need for more efficient and specialized hardware solutions to handle intensive LLM and A...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • DigiTimes

Samsung SDI strengthens LFP supply chain for US AI data centers

Samsung SDI is expanding its LFP cathode supply chain, targeting the growing US market for Energy Storage Systems (ESS) in AI data centers. This strategic move, involving Posco Future M, highlights the critical role of energy infrastructure in suppor...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • The Next Web

Rebellions Raises $400M for AI Inference Chips, Valued at $2.34 Billion

South Korean fabless AI chip company Rebellions, focused on AI Inference, has closed a $400 million pre-IPO funding round, reaching a $2.34 billion valuation. Backed by giants like Samsung and SK Hynix, the company targets US customers such as Meta a...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • TechCrunch AI

AI Chip Startup Rebellions Raises $400 Million for Inference Solutions, Challenges Nvidia

AI chip startup Rebellions has secured $400 million in a pre-IPO funding round, valuing the company at $2.3 billion. The firm, which plans to go public later this year, specializes in designing chips for AI inference, positioning itself as a new cont...

#Hardware #LLM On-Premise #DevOps

2026-03-30 • The Register AI

South Korea's Rebellions Raises $400M for Global Rack-Scale AI Platform

South Korean AI chip startup Rebellions, backed by SK Telecom, has secured $400 million in a pre-IPO funding round. This investment aims to support the global expansion of its new rack-scale compute platform, designed for enterprises and sovereign cl...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • DigiTimes

Tai-Tech Boosts Malaysia Production: Focus on AI Servers and Electric Vehicles

Tai-Tech, a passive component manufacturer, is expanding its factory in Malaysia. This move is driven by the increasing demand for AI servers and electric vehicles, both rapidly growing sectors. The company has also secured a new client in the United...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • DigiTimes

Arm Expands Beyond Licensing with New AI CPU Platform

Arm is redefining its traditional licensing business model by introducing an innovative CPU platform specifically designed for artificial intelligence workloads. This strategic move aims to offer optimized hardware solutions for AI, potentially influ...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • DigiTimes

Samsung Invests in AI Chip Design Startup to Boost Silicio Speed and Reduce Power Consumption

Samsung has announced a strategic investment in a startup specializing in artificial intelligence chip design. The primary objective is twofold: to accelerate the development of new silicio solutions and to drastically optimize energy efficiency. Thi...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • DigiTimes

DRAM Scaling Limits: New Memory Crucial for On-Premise AI

DRAM scalability is reaching its limits, while next-generation memories face delays. Atomera's MST technology promises to improve power and bandwidth efficiency, offering benefits comparable to a manufacturing node transition, a key factor for on-pre...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-30 • DigiTimes

AI compute shifts to inference, reshaping data center bottlenecks

DIGITIMES Research's analysis highlights a transition in the AI computing landscape: the focus is increasingly shifting towards inference. This change, presented at AI EXPO 2026 by Jim Hsiao, senior analyst, is redefining the challenges and bottlenec...

#Hardware #LLM On-Premise #Fine-Tuning

2026-03-29 • Tom's Hardware

New Cambridge chip slashes AI energy use

A new chip developed at Cambridge promises to drastically reduce the energy consumption of artificial intelligence systems. The component uses a new type of memristor with a switching current approximately one million times lower than conventional de...

#LLM On-Premise #DevOps

2026-03-29 • The Next Web

European investments: focus on AI infrastructure

Last week saw a surge of investments in Europe, with a particular focus on infrastructural layers. Funding spanned diverse sectors such as semiconductor physics, orbital logistics, defense systems, and artificial intelligence, signaling a strong inte...

#LLM On-Premise #DevOps

2026-03-29 • DigiTimes

Adata invests US$3 million in KonstTech to boost AI computing infrastructure

Adata has invested US$3 million in KonstTech to boost its AI computing infrastructure. The investment aims to strengthen KonstTech's capabilities in the sector, with a focus on expanding computing resources for AI workloads.

#Hardware #LLM On-Premise #DevOps

2026-03-28 • The Next Web

Kandou AI raises $225 million to bet on copper interconnects

Swiss company Kandou AI, specializing in copper-based chip-to-chip interconnect technologies, has secured a $225 million Series A funding round. The investment, led by Maverick Silicio, includes strategic participation from SoftBank, Synopsys, Cadenc...

2026-03-28 • Tom's Hardware

Meta to fund natural gas power plants for Louisiana AI data center

Meta partners with Entergy to build seven new natural gas power plants. The goal is to deliver 7 gigawatts of power to its planned AI data center in Louisiana, ensuring sufficient energy for compute-intensive operations.

#Hardware #LLM On-Premise #DevOps

2026-03-28 • ServeTheHome

Aivres Showcases NVIDIA Vera Rubin at NVIDIA GTC 2026

Aivres showcased NVIDIA Vera CPUs and Rubin GPUs at NVIDIA GTC 2026. Blackwell Ultra and BlueField-4 DPUs were also on display. The event offered a glimpse into NVIDIA's upcoming hardware architectures for advanced workloads.

#Hardware #LLM On-Premise #DevOps

2026-03-28 • DigiTimes

Holy Stone Enterprise expands Japan and Taiwan capacity, signaling tighter MLCC supply for AI power supplies

Holy Stone Enterprise is expanding its production capacity in Japan and Taiwan. This move signals a potential tightening of the supply of multilayer ceramic capacitors (MLCCs), critical components for power supplies used in artificial intelligence sy...

#Hardware #LLM On-Premise #DevOps

2026-03-27 • TechCrunch AI

AI Infrastructure: Real-World Resistance Emerges

The expansion of AI infrastructure into the real world is meeting resistance. An AI company offered an 82-year-old woman $26 million to build a data center on her land, but she refused. Tensions are rising regarding the territorial and social impact ...

#LLM On-Premise #DevOps

2026-03-27 • Phoronix

Intel Xe Driver Improves Memory Pressure / Out-Of-Memory Behavior For vRAM With Linux 7.1

The Intel Xe graphics driver for Linux is set to receive significant updates with the arrival of kernel 7.1. The changes involve a new user-space API designed to improve video memory (VRAM) management and prevent out-of-memory situations, optimizing ...

#Hardware #LLM On-Premise #DevOps

2026-03-27 • Phoronix

AMD ROCm 7.12 Tech Preview Brings More Consumer APU & GPU Support

AMD has released ROCm 7.12 as the newest tech preview, working towards the presumed ROCm 8.0 release. This release extends support to a greater number of consumer APUs and GPUs, expanding options for developers using the ROCm ecosystem.

#Hardware

2026-03-27 • DigiTimes

Taiwan's ALi bets on custom chips for 2026 turnaround

Taiwan's ALi (Acer Laboratories Inc.) is investing in the development of custom chips with the goal of a turnaround by 2026. The strategy focuses on specialized hardware solutions for emerging markets, with a particular focus on optimizing performanc...

#Hardware #LLM On-Premise #DevOps

2026-03-27 • DigiTimes

SK Hynix keeps HBM shipments steady, targets HBM4E sample this year

SK Hynix keeps HBM (High Bandwidth Memory) shipments steady and plans to release the first HBM4E samples by the end of the year. The Nvidia Vera Rubin AI platform highlights the growing demand for advanced memory in AI systems.

#Hardware #LLM On-Premise #DevOps

2026-03-26 • LocalLLaMA

Qwen 3.5 27B: 1.1M tok/s on B200s, configurations on GitHub

Qwen 3.5 27B achieved 1.1 million tokens per second using 96 B200 GPUs across 12 nodes, thanks to optimizations like DP=8 over TP=8, a context window reduced to 4K, FP8 KV cache, and MTP-1 speculative decoding. Scaling efficiency reached 96.5% on 12 ...

#Hardware #LLM On-Premise #DevOps

2026-03-26 • Phoronix

Open-Source Nouveau Performance With Linux 7.0 + NVK Mesa 26.1-dev vs. NVIDIA Linux Driver

New benchmarks compare the performance of the open-source NVIDIA driver (Nouveau with kernel driver and Mesa NVK Vulkan driver) on Linux 7.0 and Mesa 26.1-dev, against the open-source stack shipped by Ubuntu 25.10 and the official NVIDIA driver. The ...

#Hardware