AI Infrastructure news and analysis on Web Pulse: 35381 curated articles tracking the latest AI Infrastructure developments, tools, and research, updated continuously from vetted sources.
OpenRouter has tracked the deepseek/deepseek-v4-flash-0731 endpoint at streamlake/fp8 since 2026-08-13, recording $0.032 in total spend across 66,386 queries. The B3IT border-input detector measured a total-variation lev…
A developer successfully wired a local MCP server into Claude Code, enabling the AI client to directly query a Chroma-based notes index via the today-i-ran-notes server. During testing, the tool accurately reported that …
Hugging Face released @huggingface/kernels, a JavaScript library for loading and running optimized WebGPU kernels from the Hugging Face Hub, along with an initial collection of 207 kernels published as versioned packages…
AMD's ATOM and vLLM-ATOM inference engines were optimized for high-interactivity LLM inference, targeting responsiveness for single users rather than batch throughput. The optimizations focus on removing fixed costs from…
Intel Gaudi 3's vLLM-Gaudi PagedAttention implementation underutilizes hardware for sliding window attention, but optimizations for serving Gemma 4 31B expanded usable KV cache by 3.5×, raised decode MBU from 37% to 59%,…
Weaviate introduces a late-interaction multi-vector retrieval approach that embeds entire PDF pages—including charts and tables—as single units, eliminating the need for OCR and text chunking in RAG pipelines. The method…
AWS announced agentic observability with Amazon OpenSearch Service MCP Apps on 25 August 2026, an extension to the Model Context Protocol that renders interactive dashboards inside AI chat windows, returning both a text …
NVIDIA put Groq 3 LPX into full production on 24 August, an inference accelerator designed for AI agents' long token streams, claiming roughly four times faster latency than the nearest alternative in an Artificial Analy…
Anthropic released Claude Fable 5.1 on September 1, 2026, cutting cache-read pricing by 75% to $0.25 per million tokens while keeping base input at $10 and output at $50 per million tokens, yielding an estimated 25% lowe…
Val Town, a platform for writing and hosting server-side JavaScript, found that the majority of new Pro users were referred by AI, mostly Claude, prompting the company to begin optimizing for answer engines. The company …
A new measurement of MCP servers shows tool-definition weight varies by more than 30x across popular servers, with Notion's 24 tools costing about 19,050 estimated tokens versus a one-tool fetch server at around 600 toke…
On September 1, 2026, Perplexity released Hybrid Compute for its Mac app, which splits tasks between the cloud and the user's Mac based on data sensitivity, with sensitive steps running locally via an on-device classifie…
Anthropic's Claude Fable 5.1, released September 1, 2026, strips its thinking blocks when a conversation is downgraded to an older model, causing the target model to re-plan without that reasoning and potentially raising…
Agent memory providers including Mem0, Zep, Letta, LangMem, Amazon Bedrock AgentCore Memory, Redis, Supermemory, and Cognee are converging on a common architecture that moves memory outside individual agents so multiple …
A developer built a hardware-in-the-loop setup for AI coding agents using a Raspberry Pi Compute Module 5 and an Orange Pi 3 LTS, enabling the agent to flash, boot, and debug real embedded devices. The project, called cm…
A developer's AI pipeline was fixed after a blog commenter, pm25coder, pointed out that provenance records could be gamed by consumers that record source IDs without reading content. The team implemented read receipts at…
A developer built an AI-powered call router using Telnyx's Edge Runtime and AI Inference binding, which understands caller intent and routes calls without external glue code. The system uses a StatefulActor to hold call …
Nvidia is expanding its IP licensing business by partnering with MediaTek, which will integrate Nvidia's NVLink Fusion interconnect into its datacenter XPUs, with Nvidia investing $3.5 billion in convertible bonds in Med…
A team of developers built EIR, a secure multi-agent healthcare platform on Google Cloud, for the All Things Agentic Hackathon. EIR coordinates specialized agents for patient access, recovery, scheduling, and hospital op…
Amazon Web Services (AWS) published a guide on connecting an AgentCore Runtime hosted Model Context Protocol (MCP) server to Amazon Quick, enabling AI agents to access external tools and data. The integration uses AgentC…