AI Compute

Twitter

Facebook

Copy Link

Key Facts

Key PlayersNVIDIA, Microsoft, Google, AWS, CoreWeave

Top News

Torus Foundry Secures $500M to Deploy Modular AI Data Centers Using Existing Grid Capacity

Salt Lake City-based Torus raised $500 million in Series C funding led by Magnetar to launch Foundry, prefabricated AI inference computing units that can be deployed at existing sites in 12 to 16 weeks. The modular data centers range from 1 to 20 megawatts and use available grid capacity, bypassing the need for new power infrastructure as US data center developers face mounting power constraints.

Technology3 sources

Torus Foundry Secures $500M to Deploy Modular AI Data Centers Using Existing Grid Capacity

Technology3 sources

Microsoft Auto SR Exits Preview: 120 FPS AI Upscaling Arrives for ROG Ally X Handheld Gaming

Microsoft officially launched Automatic Super Resolution for ROG Xbox Ally X and X20 handhelds, introducing 120 FPS Performance Mode and undocked AI-powered upscaling. The update removes preview restrictions, adds 1080p support, and brings Auto HDR compatibility to enhance gaming visuals without requiring external displays.

Technology4 sources

Microsoft Auto SR Exits Preview: 120 FPS AI Upscaling Arrives for ROG Ally X Handheld Gaming

Technology4 sources

AI Data Centers Trigger Fiber Optic Shortage, Squeezing Ukrainian Drone Operations

AI data centers are consuming massive amounts of fiber optic cables, creating a global shortage that's hitting Ukraine's drone warfare capabilities hard. Ukrainian defense firm BlueBird Tech reports 6-10X price increases as Russia secures 37 million miles of cable while Ukraine struggles with supply chain constraints and Chinese suppliers refusing sales.

Business2 sources

AI Data Centers Trigger Fiber Optic Shortage, Squeezing Ukrainian Drone Operations

Business2 sources

AMD Launches Zen 6 EPYC Processors With 256 Cores, Challenges Nvidia's Agentic AI CPU Claims

AMD unveiled its 6th Gen AMD EPYC processors featuring Zen 6 architecture with up to 256 cores and 512 threads, targeting agentic AI workloads. Chief Zen architect Mike Clark and VP Madhu Rangarajan dismissed competitor claims that Nvidia Vera and Arm CPUs were purpose-built for agentic AI, arguing CPU development cycles make such claims unrealistic.

Technology5 sources

AMD Launches Zen 6 EPYC Processors With 256 Cores, Challenges Nvidia's Agentic AI CPU Claims

Technology5 sources

US Power Demand Set for Record Highs in 2026 and 2027 as AI Data Centers Drive Unprecedented Surge

The U.S. Energy Information Administration forecasts US power demand will climb to historic levels, reaching 4,370 billion kWh by 2027. AI data centers and broader electrification trends are fueling this surge, marking three consecutive years of record high electricity consumption while reshaping America's energy landscape.

Business2 sources

US Power Demand Set for Record Highs in 2026 and 2027 as AI Data Centers Drive Unprecedented Surge

Business2 sources

Stanford AI100 Report Warns AI Scaling Era Faces Limits as Investment Outpaces Economic Returns

Stanford's One Hundred Year Study on Artificial Intelligence released its third report warning that the scaling era driving AI development is hitting constraints. The AI100 report finds investment in data centers and chips outpacing earlier booms while AI is not yet paying its way economically, despite reshaping billions of lives with limited public involvement.

Policy4 sources

Stanford AI100 Report Warns AI Scaling Era Faces Limits as Investment Outpaces Economic Returns

Policy4 sources

Mercor Doubles Valuation to $20bn as AI Agent Training Shifts to Enterprise

San Francisco-based Mercor raised $750 million at a $20 billion valuation, doubling from $10 billion in October 2025. The AI training startup connects over 5 million subject-matter experts with companies to train and evaluate AI models and agents, now expanding from frontier labs to enterprises.

Startups6 sources

Mercor Doubles Valuation to $20bn as AI Agent Training Shifts to Enterprise

Startups6 sources

Hedra Launches Model Factory to Accelerate Physical AI Model Training with Recursive Self-Improvement

Hedra, backed by Andreessen Horowitz, has launched Model Factory—an autonomous training pipeline that uses recursive self-improvement to help robotics, manufacturing, biology, and defense teams build and continuously optimize custom AI models for real-world applications without massive retraining costs.

Technology2 sources

Hedra Launches Model Factory to Accelerate Physical AI Model Training with Recursive Self-Improvement

Technology2 sources

CoreWeave pivots to prepayments and asset-backed debt to fuel AI infrastructure expansion

CoreWeave reported strong third-quarter results with 168% revenue growth but widening losses as interest expenses hit $846 million. The neocloud provider is abandoning unsecured debt markets, instead relying on $9 billion in customer prepayments and chip-backed loans to finance its ambitious AI infrastructure buildout.

Business27 sources

CoreWeave pivots to prepayments and asset-backed debt to fuel AI infrastructure expansion

Business27 sources

SpaceX targets 10GW AI computing capacity in 18 months, orders 100GW solar manufacturing for satellites

SpaceX AI expansion accelerates with plans to add 10GW data center capacity by 2027—matching Google's cumulative development. Elon Musk orders 100GW solar manufacturing equipment from Germany's SINGULUS TECHNOLOGIES for Starmind AI satellites, targeting a $28.5 trillion market with space-based computing infrastructure.

Technology4 sources

SpaceX targets 10GW AI computing capacity in 18 months, orders 100GW solar manufacturing for satellites

Technology4 sources

Australian Startup Tests Quantum Computer to Tackle AI Data Centers' Energy Crisis

Australian startup Diraq is conducting the world's first field test of a silicon chip quantum computer in a commercial data center in Sydney. Founder Andrew Dzurak says quantum computing will be ready to address gigawatt-scale AI energy consumption within five years, with their system drawing less than 20 kilowatts of power.

Technology2 sources

Australian Startup Tests Quantum Computer to Tackle AI Data Centers' Energy Crisis

Technology2 sources

Microsoft Patent Reveals Plan to Pay Users for AI Training on Idle GPUs

Microsoft filed a patent for a distributed AI training system that would use idle gaming PCs to process micro-training tasks. The opt-in system would monitor GPU utilization, temperatures, and power draw, compensating users with credits, software access, or cash for their computational contributions.

Technology6 sources

Microsoft Patent Reveals Plan to Pay Users for AI Training on Idle GPUs

Technology6 sources

AMD Ryzen AI Halo Gets Project Zenith Ready-to-Code Setup for Local AI Development

AMD now offers Microsoft's Project Zenith on Ryzen AI Halo workstations powered by Ryzen AI Max+ PRO 495, delivering a preconfigured Windows 11 developer setup with up to 192GB unified memory. The ready-to-code environment includes Visual Studio Code, Python, and AI Playbooks, enabling developers to run models with up to 300 billion parameters locally without cloud infrastructure.

Technology2 sources

AMD Ryzen AI Halo Gets Project Zenith Ready-to-Code Setup for Local AI Development

Technology2 sources

Developer Creates 'Ram-Altman' App to Email Sam Altman About Unused RAM on Your PC

A UK developer launched Ram-Altman, a tongue-in-cheek Linux script that automatically emails OpenAI CEO Sam Altman when more than 10% of a computer's RAM sits unused. The satirical response follows Altman's controversial podcast remarks about DRAM in personal computers being a waste of resources.

Entertainment4 sources

Developer Creates 'Ram-Altman' App to Email Sam Altman About Unused RAM on Your PC

Entertainment4 sources

Microsoft VP releases open-source GPU task manager to track VRAM and AI workloads

Microsoft VP Scott Hanselman launched gputaskman, an open-source GPU task manager that reveals per-process GPU usage and VRAM consumption across platforms. The tool provides detailed breakdowns of AI and LLM workloads, including model weights and KV cache data, addressing gaps in Windows Task Manager's GPU monitoring capabilities.

Technology3 sources

Microsoft VP releases open-source GPU task manager to track VRAM and AI workloads

Technology3 sources

Prev 123 4…105 Next