Cloud
Google Uses Gemini 4 Argon to Optimize 200K-Line Python Agent Development Kit Overnight
Google’s Gemini 4 Argon accelerated a 200,000-line Python codebase, cutting build times and improving runtime efficiency for its Agent Development Kit.
AI infrastructure
20 stories
More in AI infrastructure
- Cloud
Google and Meta Release TorchTPU: Native PyTorch on TPUs
Google and Meta open-source TorchTPU, enabling native PyTorch execution on TPUs with a new compiler backend and runtime.
- Technology
AI Power Demands Surge During Climate Week, Prompting Grid Strain and Investment Shifts
AI workloads at Climate Week sparked a surge in data-center electricity demand, reshaping climate-tech investment and highlighting grid reliability challenges.
- Cloud
GitLab Duo self-hosted AI expands with Microsoft Foundry
GitLab Duo self-hosted AI now runs on Microsoft Foundry, giving enterprises on-premise control over large language model deployments and security policies.
- Technology
Nscale files $3 billion IPO targeting $35 billion valuation
British neocloud provider Nscale filed to list on the NYSE on September 22, 2026, seeking to raise $3 billion and achieve a $35 billion valuation.
- Cloud
Google AX Orchestrator Open-Sourced: A Kubernetes-Style Orchestrator for Autonomous AI Agents
Google AX orchestrator offers a Kubernetes-style control plane for autonomous AI agents, letting teams reuse familiar tooling while adding safety.
- Technology
Data Center Opposition Rises as AI Investment Hits Pennsylvania
A Data & Society report shows why Pennsylvania residents are pushing back on data center projects, highlighting trust gaps, economic concerns.
- Cloud
DoorDash Deploys Multi-Agent LLM System to Automate Cleanup of 60,000 Feature Flags
DoorDash’s new multi-agent LLM pipeline generated pull requests for 45 of 50 stale flags in under 14 minutes each, cutting technical debt at scale.
- Technology
Crusoe $3.9 billion data center funding powers megacenters and modular AI factories
Crusoe secures $3.9 billion in Series F financing to scale Texas megacenters and launch truck-able Spark AI factories, reshaping the compute landscape.
- Technology
Single-Phase Direct Liquid Cooling Proven for Ultra-Dense AI Compute
Single-phase direct liquid cooling delivers reliable thermal control for 1,000-W CPUs and 100-kW racks, securing data-center efficiency for the next decade.
- Development
GitHub Copilot HydraFusion Boosts Coding with Multi-Model Routing
GitHub Copilot's HydraFusion research preview introduces runtime model orchestration that boosts coding assistance performance beyond current LLM limits.
- Technology
Nvidia revenue growth 70%: Jensen Huang outlines 2027 outlook
Jensen Huang details Nvidia revenue growth 70% target for fiscal 2027, driven by AI-first hardware, Grace-Blackwell supernodes.
- Tech & AI
AI Data Center Accountability: Lessons from the $3.2 B Lake Mariner Fire
A fire at New York's $3.2 B Lake Mariner AI hub exposes how layered ownership erodes safety, jobs, and cost responsibility.
- Tech & AI
AI Data Center Accountability: Lessons from the $3.2 B Lake Mariner Fire
The Lake Mariner AI data center fire highlights accountability gaps, safety lapses, and the tangled corporate web behind a $3.2 billion project.
- Technology
AI model downtime: Four major models experience rare overlapping outages
A rare instance of AI model downtime saw ChatGPT, Claude, Grok, and Gemini all suffer overlapping outages, highlighting shared infrastructure risks.
- Cloud
Foundry Model Router expansion drives latency gains across 28 regions
The Foundry Model Router expansion adds 26 new regions, refreshes its model pool, and lowers inference latency while addressing data-sovereignty requirements.
- Artificial Intelligence
Jalapeño inference chip powers OpenAI's full stack for abundant intelligence
OpenAI's custom Jalapeño inference chip beats commercial rivals on throughput per kilowatt, reshaping its compute stack and developer ecosystem.
- Technology
IBM Granite 4.2: Three New Local LLMs Target Enterprise Reasoning
IBM unveils Granite 4.2 (3B, 8B, 30B) with 128k context, tool-use training, and enterprise-grade reasoning, reshaping local AI deployments with IBM Granite 4.2.
- Artificial Intelligence
Netflix's LLM Serving Platform
Netflix details its in-house LLM serving platform with Triton and vLLM for improved AI performance and efficiency in content recommendation.
- Artificial Intelligence
The AI Agent Protocols Imperative: Why Vinton Cerf’s Retirement Signals a New Era for Autonomous Systems
The primary focus of his final predictions is on the need for formal AI agent protocols to ensure reliable interoperability among autonomous agents.