Novosti Vijesti: 29 kolovoz 2025 - 8 listopad 2025

October 14, 2025, 12:25 AM EDT. OpenAI is partnering with Broadcom to design and deploy roughly 10 gigawatts of custom AI chips and systems, a scale that could power a large city and underscores how power-hungry the AI boom has become. The alliance follows earlier deals with Nvidia and AMD as OpenAI expands its compute capacity for a growing user base-ChatGPT now reaching about 800 million weekly users, with rival Sora 2 generating headlines for rapid growth. Deployment is slated for the second half of 2026. The project would consume roughly the energy of 8 million U.S. households, prompting concerns over data center energy use and the broader share of electricity going to tech infrastructure. Broadcom's stock rose, and OpenAI could embed hardware learnings to accelerate future models and services.

October 14, 2025, 12:24 AM EDT. Vodafone has reported a major outage affecting its UK network, with customers unable to access broadband, 4G and 5G services after around 3pm BST. The disruption spanned the country, and even Vodafone's official network status pages were reportedly offline amid the surge in traffic. A spokesperson apologized and said the company is working to resolve the issue as quickly as possible. Some customers saw problems on other providers' networks, as users checked DownDetector for reports. Vodafone has not yet offered a firm timeline for restoration, but engineers are investigating the fault and coordinating with suppliers. The outage highlights how dependent modern homes and businesses are on mobile and fixed connectivity, and how swiftly such disruptions spread online.

October 14, 2025, 12:23 AM EDT. An independent developer tests Google Gemini to replace a traditional Kanban board. The piece explains how Gemini can automatically triage tasks, generate cards, and update statuses, enabling a more fluid, AI-powered workflow. It covers setup steps, data migration, and integration with existing tools, plus tradeoffs like learning curve and data governance. Benefits highlighted include faster triage, fewer context switches, real-time visibility, and reduced manual overhead. The article also offers practical tips for teams eyeing an AI-assisted board and cautions against overreliance on automation. In short, Google Gemini can act as a new kind of workflow assistant that reshapes how developers plan, track, and deliver work, with Kanban as a starting point rather than a fixed container.

October 14, 2025, 12:22 AM EDT. EcoFlow's Delta Pro Ultra X is billed as the most powerful home energy backup. The system offers 12kW of continuous output and can scale to 36kW with three inverters in parallel, and battery capacity from 12kWh to 180kWh. That means a typical home could run for about six days, or weeks for critical loads, during outages. It works with EcoFlow's third-generation Smart Home Panel (200 A; up to 32 circuits) and supports Time-of-Use charging, as well as solar, gas generators, and EV chargers. It's modular and portable, with installation in under seven days via EcoFlow-certified electricians; bundles start at $7,999 (inverter + two batteries), rising to $10,899 with Smart Home Panel 3. 425 installer partners nationwide.

October 14, 2025, 12:21 AM EDT. Traditional data centers are evolving into AI factories where power infrastructure determines location, scale, and feasibility. The proposed blueprint combines an 800 VDC power distribution system with integrated, multi-timescale energy storage to meet rising, volatile AI workloads. As GPUs cluster with high-bandwidth interconnects (e.g., NVIDIA NVLink), the move from copper cabling becomes essential to avoid overwhelming current and copper volume. The jump from NVIDIA Hopper to Blackwell demonstrates how a 75% rise in GPU power, a 72-GPU NVLink domain, and a 50x performance gain push rack power density from tens of kW to well over 100 kW, with megawatt-per-rack on the horizon. Traditional 54 VDC is impractical due to losses and copper needs. AI workloads introduce a new volatility profile: synchronous cycles across thousands of GPUs create facility-wide load swings, demanding a fresh architectural approach.