Server/Datacenter

A popular game archive service with around 390 TB of data, Myrient, will be closing its metaphorical doors later next month. Citing rapidly rising infrastructure costs, abusive usage, and lack of increase in donations, the service's founder has deemed it unsustainable.

Many ISP-provided routers are closed black boxes with limited functionality. In part 3 of my “Unbloated Network” project, I show how to install OpenWRT on an x86 PC to replace the Vodafone router with your own hardware. Step by step, we’ll set up the latest stable release, including image selection, Linux-based installation, and partition expansion.

Small mistake, big impact. A misplaced backslash caused a PowerShell script running on GPT to completely erase the hard drive contents for a Reddit user. A simple "typo" that could have happened to anyone reveals that it's not just vibecoding that is prone to errors.

After deciding against using the ISP router in part one, it’s time to focus on hardware: what actually works best for an OpenWRT router? A mini PC, Raspberry Pi, or a traditional router? And what about the switch and access point? In this part, I’ll walk through my hardware choices, including considerations around power consumption, architecture, and cost.

Provider routers are convenient—but that’s exactly their biggest problem. Poor Wi-Fi coverage, limited control, questionable privacy, and technical limitations are everyday issues for many users. In this new series, I’ll gradually replace my provider’s all-in-one device (the Vodafone Station) with a modular setup using an OpenWRT router, access point, and switch. In part 1, I’ll explain why I’m taking this approach and how the project is structured.

Lenovo’s North American President is unusually candid about the AI boom's impact. In a recent interview, he advised that partners commit to orders before the memory shortage depletes stock levels. Large and small businesses often find that servers, laptops, and desktops sell out within 24 hours.

A test by The Guardian indicates that ChatGPT is increasingly using Elon Musk’s Grokipedia as a source. Using the AI-generated website as a source for other AIs could lead to a "garbage in, garbage out" scenario in which unfiltered information is spread.

Two years of work lost with a single click—that is how Professor Marcel Bucher describes his experience in an article in Nature. He concludes that ChatGPT is a threat to consistent productivity. However, the chatbot offers a backup function that the researcher apparently failed to utilize.

Nvidia CEO Jensen Huang has revealed that Team Green overtook Apple in 2025 to become TSMC's biggest customer. Nvidia has had an excellent 2025, driven by strong demand for AI hardware, with a 62% Year-over-Year revenue growth for Q3 FY2026.

A high-ranking official at Micron recently shed light on the company's decision of pulling out from the consumer storage and memory market by shutting down its Crucial brand. While the AI gold rush is undeniably attractive for large companies, it appears that the average consumer is getting hammered in the process.

Reports are spreading that Microsoft Defender is blocking the popular community tool MAS, flagging it as a fake. In contrast, our own tests demonstrate that the script works flawlessly. Have affected users perhaps fallen victim to DNS tampering?

Razer has unveiled the Forge AI Dev Workstation, a custom-built desktop designed to handle AI development projects. The offering eliminates the need to figure out how to build systems with up to four powerful Nvidia GPUs, and Intel Xeon W or AMD Threadripper Pro CPUs.

Following the nearly half-million dollar crowdfunding success of the Mini-KVM, Openterface is back with the KVM-GO, a tiny KVM that reduces cable clutter by including all the hardware in the display connector.

I find the idea of self-hosting fascinating. I read about it regularly, and as someone who reviews PC hardware, I’m practically surrounded by ideal conditions—and gear—to finally get started. I’ve even tested devices that would be perfect for it. Yet despite thinking about it constantly, I still haven’t fully committed. Why?

Aaeon has introduced four new offerings carrying the UP brand name. Each of the mini PCs can be equipped with up to an 8-core Intel N355 CPU, and packs a decent array of standard as well as enterprise-focused I/O.

In a recent discussion on the Broken Silicon Podcast, veteran memory industry executive Dave Eggleston breaks down the key forces driving the current RAM supply crunch. He explains how the industry is reacting to massive memory procurement deals, how demand shifts are tightening DRAM availability, and what these pressures could mean for upcoming consumer hardware.

Another major memory and SSD storage maker has bad news for consumers. The Crucial brand is being sacrificed on the altar of AI profits after three decades in existence.

2026 could turn everything upside down. AI is no longer content with software and is now reaching for hardware, according to the conclusion reached by TrendForce. At the same time, rumors are rife. Apple is finally planning to launch a foldable iPhone, while new AI chips are running so hot that fans are becoming useless.

IceWhale, the company behind ZimaBoard and more, is joining the Black Friday rush with discounts of up to 40 percent across its lineup. Among other deals, buyers can save 20 percent on a ZimaBlade or ZimaCube. Orders ship within 72 hours.

On the surface, the internet seems to be decentralized. In reality, only a few global players ensure the functionality of websites and apps. Cloudflare, one of the largest, was rendered unavailable for hours on Tuesday, November 18th. What initially seemed like an attack turned out to be a simple mistake.

The 3D-stacked DRAM technology jointly developed by d-Matrix and AIchip has just been successfully validated and is ready for commercial launch. It will be featured in the upcoming Raptor inference accelerators from d-Matrix and the technology is supposed to boost inference speeds by up to 10 times when compared to similar HBM4-based accelerators.

Meta is investing over $1 billion in a new AI-optimised data centre in Beaver Dam, Wisconsin, which the company says will use upgraded infrastructure, improved cooling and clean-energy matching.

Artificial intelligence (AI) is the key factor behind the current boom of German data centers. According to a recent market analysis by digital association Bitkom, installed capacity will increase by a full 70% by 2030. The capacity exclusively used for AI applications is expected to quadruple from 530 megawatts to over 2,000 megawatts.

APT is set to gain Rust code in 2026 as Debian modernises its package manager. The coding language's presence in open-source projects keep growing.

Elon Musk has confirmed that SpaceX plans to build space-based data centers by scaling up its Starlink V3 satellites. With 1 Tbps-class capacity and launches planned as early as 2026, Starlink could soon evolve from broadband into orbital cloud infrastructure provider.

Qualcomm is expanding beyond smartphones with two new AI data center chips, the AI200 and AI250, built on its mobile neural processing tech. Promising significant efficiency gains and 768 GB memory support, these chips aim to challenge Nvidia’s dominance and power Saudi Arabia’s next-gen AI infrastructure.

In a significant step forward towards green computing, China has finished construction on a new underwater data center that is powered by offshore wind, uses zero freshwater, and dramatically reduces land use.

Intel will no longer be blindly contributing to the open-source software ecosystem, with recent comments from the company indicating that it will be using open-source as an "advantage to Intel." Intel was previously known for contributing substantial development efforts to open-source projects and releasing many of its AI acceleration tools and programs under open-source licences.

As previously announced, Ubuntu 25.10 became available earlier today as the final step before the next LTS release, which is scheduled for 2026. In addition to the 6.17 kernel and GNOME 49, this release also comes with the Ptyxis terminal, the Loupe image viewer, improved Bluetooth audio handling, and more.

Gneuton, a US-based company, has launched a patented thermal distillation system that repurposes waste heat from gas turbine-powered AI data centers into purified water, aiming to cut water costs and boost sustainability worldwide. The company plans pilot deployments in global AI hubs.

The memory semiconductor industry doldrums period is seemingly over as companies like SanDisk, Micron, or Samsung are now all intent to raise their prices. The AI-powered boom of chips and data center storage demand may be the culprit.

CheckMag From building a pico projector that mere mortals might actually be able to achieve, to running a web server on a Vape battery, the hacker community never ceases to amaze. Here are two of the week's best hacks.

Ubuntu 25.10 "Questing Quokka" is one step closer to the masses thanks to a new beta release. Targeting multiple platforms and available in quite a few versions, as usual, this release includes the 6.17 Linux kernel and most features that will come with the final release, expected to arrive on October 9.

At Huawei Connect 2025, the company introduced the Atlas 950 SuperCluster, delivering 524 FP8 ExaFLOPS for AI training and 1 FP4 ZettaFLOPS for inference. Built with over 500,000 Ascend 950DT NPUs, the system outpaces Oracle’s and xAI’s clusters while challenging Nvidia’s upcoming Rubin platform.

Google has pushed its “net-zero by 2030” pledge off the homepage and into its environmental report, while its data center page now frames the goal as a “moonshot.”

Thanks to Axiom Space and Red Hat, edge computing is now making its way to space. Known as AxDCU-1, the first orbital data center prototype is now being tested on the International Space Station. Instead of transmitting data to Earth for analysis, it crunches numbers locally and only sends down the results.

China’s cities are rolling out “computing power vouchers” that cover up to 80 percent of AI rental costs for SMEs, turning underused data centers into subsidized AI hubs.

China’s top tech firms are accelerating efforts to replace Nvidia processors, but most progress lies in running AI models, not training them.

Export controls and new data security rules keep Nvidia’s latest accelerators out of reach for many Chinese firms. Instead, refurbished A100s and second-hand H100s are powering inference clusters, extending the life of older hardware and reshaping the country’s AI hardware economy.

SK hynix has entered mass production of its 321-layer, 2-terabit QLC NAND, the first QLC to break 300 layers. A six-plane architecture and 3,200 MT/s I/O double transfer speeds apparently lift writes by up to 56 percent and reads by 18 percent, while cutting write power by more than 23 percent.

Nvidia has taped out six Rubin architecture chips at TSMC, featuring 3nm N3P chiplets, HBM4 support, and 4x reticle design. This comprehensive platform overhaul covers GPUs, CPUs, networking, and silicon photonics.

RIKEN, Fujitsu and Nvidia are joining forces on FugakuNEXT, Japan’s next flagship supercomputer. Targeting 600 EFLOPS FP8 and a hundred-fold performance boost over Fugaku, the system combines new Fujitsu CPUs and Nvidia GPU accelerators to power AI-driven science by 2030.

Beijing has quietly tightened pressure on tech giants to scale back Nvidia H20 chip purchases after U.S. comments angered senior officials.

The company will acquire up to 500 megawatts of nuclear energy under the agreement. This deal will allow an increase in the power output of the Hermes 2 Plant in Oak Ridge.

Nvidia shared an op-ed arguing U.S. export controls on its H20 GPUs curtailed American influence without halting China’s AI progress.

PQShield expands post-quantum cryptography hardware with UltraPQ-Suite and PQPlatform-TrustSys, building on its 2024 NIST-approved silicon debut.

Sandisk has unveiled the UltraQLC 256TB NVMe U.2 SSD, engineered for massive data storage at datacenters and AI training clusters. Built using the 8th gen BiCS QLC NAND architecture, it delivers faster performance with lower power consumption.

Cook confirms seven small acquisitions in 2025, amid broader AI investments that may strengthen on-device performance. The news comes as Apple's AI features currently available, such as those found on iPhones, face regular criticism.

To help ease pressure on the U.S. power grid, Google will begin cutting back electricity use at its AI data centers during times of high demand. The company has signed its first agreements with utility providers to temporarily slow down machine learning tasks, marking a new step in how tech giants manage growing energy needs.