AI Infrastructure news and analysis on Web Pulse: 13265 curated articles tracking the latest AI Infrastructure developments, tools, and research, updated continuously from vetted sources.
NVIDIA's closed-source cuda-checkpoint tool, which freezes and restores CUDA processes, suffers from slow PCIe transfers. Engineers reverse-engineered the tool to understand why checkpoint transfers fail to saturate PCIe…
A technical guide explains how to build production AI agents with database-backed context retrieval and long-term memory, addressing the failure of demo agents at scale by architecting three memory layers: working memory…
OpenAI found roughly 30% of SWE-Bench Pro tasks broken, while Cognition announced SWE-1.7 rivaling GPT-5.5, shifting competition from model performance to benchmark control. A new runtime security layer emerged with tool…
XAI released Grok 4.5 on 8 July 2026, priced at $2 per million input tokens and $6 per million output tokens, making it roughly five times cheaper than competitors like Claude Fable 5 and GPT-5.5. The model ranks fourth …
Meta released Muse Spark 1.1, a multimodal reasoning model for agentic work, alongside the public preview of the Meta Model API. INT21 tested the model in a Swarm research system with 29 agents processing 40.5 million to…
Trigger.dev, an open-source platform for serverless asynchronous tasks, migrated its observability infrastructure from Postgres to ClickHouse to handle the growing volume of telemetry from long-running AI workflows. The …
AMD has published an analysis of network traffic patterns in AI training clusters, identifying congestion hotspots and harmonic-like oscillations caused by synchronized collective communication operations. The report exa…
Fastly announced a prototype to secure Model Context Protocol (MCP) servers, addressing AI-specific attack vectors, production-scale performance, and authentication complexity for a major global publisher. The solution c…
OpenAI released GPT 5.6 in three models—Sol, Terra, and Luna—now available on Vercel's AI Gateway. The models offer improved agentic capabilities in coding, biology, and cybersecurity with greater token efficiency. AI Ga…
Meta released Muse Spark 1.1, a multimodal reasoning model with a 1M token context window, now available on Vercel's AI Gateway. The model supports text, image, video, PDF, and audio inputs, and is designed for agentic t…
Claude Fable 5 can be used as an orchestrator in multi-agent workflows, delegating execution to cheaper models like Opus or Sonnet, cutting token costs by 10x without quality loss. The approach involves using Fable 5 for…
Context rot degrades AI agent output quality when a model's context window reaches 70-80% capacity, causing attention dilution and signal-to-noise loss. Anthropic's Claude Code addresses this with an auto-compact feature…
A new multi-agent workflow pattern uses Fable 5 as an architect for planning and Grok 4.5 as a construction crew for execution, cutting costs by shifting 90% of token spend to cheaper inference. The approach built a 50-d…
A new analysis distinguishes personal AI agents built for individual use from production AI agents designed for multiple users, highlighting that Markdown-based workflows fail at scale due to structural limitations in ha…
Palantir CEO Alex Karp argues that the AI bubble is deflating, with value shifting from data-providing companies to dominant platforms that engineered dependency through loss-leading API pricing and contract structures. …
Apple plans to release a new Apple TV 4K in 2026 with an A17 Pro chip, Wi-Fi 7, and Bluetooth 6, but the launch is delayed until after Siri AI features debut in iOS 27. The device will retain the same design as the curre…
AI agents that can both read and send email create a 'Lethal Trifecta' of security risks from untrusted content, sensitive data, and external communication. MailKite, a company building agent email infrastructure, offers…
Hugging Face announced that its Transformers modeling backend for vLLM now matches or exceeds native vLLM speed for compatible architectures, reducing the need for separate hand-optimized serving ports. The update allows…
Skillburst launches a platform that lets non-engineers use AI workflows built by their technical teams inside tools like Claude, Cursor, and ChatGPT without writing code. The startup addresses the problem of workflows be…
Zhipu AI, a Beijing-based developer of large language models, priced its share placement at HK$1,588 per share to raise approximately $4 billion, testing investor appetite for Chinese AI stocks after its stock surged nea…