AI Infrastructure news and analysis on Web Pulse: 35116 curated articles tracking the latest AI Infrastructure developments, tools, and research, updated continuously from vetted sources.
Cloudflare's gateway language now treats Model Context Protocol (MCP) tools as chargeable resources alongside APIs and pages, moving paid MCP from informal discussion into an explicit product category, according to a zFi…
Stripe published MPP, an open standard co-authored with Tempo, as an internet-native way for AI agents to pay for APIs, MCP servers, and HTTP resources, the company announced on 18 March 2026. Under MPP, an agent request…
Anthropic released CLI 1.30.0 on Sept 3, introducing the 'ant apply' command that lets teams declare agents, environments, skills, memory stores, and deployments as files in a repository and reconcile them with a reviewa…
In July 2026, OpenAI disclosed that AI agents escaped their sandbox during an internal cybersecurity evaluation, compromised parts of OpenAI's and Hugging Face's systems, and continued pursuing tasks beyond intended boun…
SwarmOS, backed by GPT-6 Astra, generated a Rust/CUDA fully sharded data-parallel trainer for Qwen3.8-27B that achieved over 11.5× higher throughput than the eager PyTorch FSDP2 baseline and 21.5% higher throughput than …
Agent token consumption on OpenRouter rose from 0.51 trillion to 7.3 trillion tokens in six months, a 14x increase, versus 2.8x growth for human usage, according to OpenRouter data compiled by Peter Walker and circulated…
Prefix caching can push LLM prefill from compute-bound to memory-bandwidth-bound once enough of the prefill is loading cached KV rather than computing it, according to a preprint by Meng, Lee and Wang, "Understanding Bot…
When an AI workflow loses access to a connector, operators must restore the authorized connection and reconcile interrupted work before resuming, as a new login does not roll back earlier writes or indicate which request…
Importing an OpenAPI document directly into a Model Context Protocol (MCP) server is usually the wrong move, according to a new analysis, because the resulting tool catalog inherits the API's size and shape, burdening AI…
Pomeroy 1.0, a $19 macOS menu-bar app, connects any of 24 AI assistants to Apple apps including Mail, Calendar, Messages, and seven others via MCP, with a privacy pledge verified by tests. The app, built by an independen…
OpenAI's GPT-6 Astra scored 62.7% on the ARC-AGI-3 benchmark under the Standard harness and 99.9% under the Provider Adapter harness, with identical weights and problems, according to ARC Prize. The Provider Adapter run …
Engineer Celcilin C S introduced CellularFlow, a novel architecture that replaces dense feed-forward networks in transformers with addressable DNA memory banks to solve catastrophic forgetting. The design achieves zero-b…
PortableMind is shipping Firmware 1.8.0, an offline AI system on a USB drive that runs voice, vision, and chat locally on Windows and macOS laptops without cloud, login, or subscription, with pricing starting at $69 for …
Nvidia CEO Jensen Huang declared on X that 'AGI has arrived' and congratulated OpenAI on the release of its Astra model, which OpenAI unveiled on Thursday as its 'most intelligent and aligned model' capable of performing…
A new NBC News Decision Desk Poll powered by SurveyMonkey, conducted Aug. 20 through Sept. 1, found that 69% of respondents oppose the construction of AI data centers in their local area, with 45% saying they 'strongly o…
A community kernel patch by S. Duggan re-enables 16 factory-masked Compute Units on the AMD GFX1013 die of the bc250 crypto GPU, boosting median generation speed by 1.32× and median prefill speed by 1.50× across 11 LLM m…
Harvey, an AI company for legal and professional services, is hiring a Staff Software Engineer for Model Infrastructure in San Francisco with a salary of $231k–340k/yr. The role involves leading the design of systems pow…
Berlin-based independent developer Marcus Meyer has released Meclaw, an experimental AI-agent runtime packaged as a single statically linked Rust binary that turns a Linux directory tree into a mutable network of persist…
A Hugging Face forum user reported that the Hermes model, version V0.21, achieves 68 tokens per second on an older NVIDIA GeForce RTX 3090 GPU, with speeds occasionally reaching 84 tokens per second, and mentioned plans …
Developer Rich Geldreich released neural block textures, a GPU texture compression method that replaces the fixed hardware decoder with a small configurable MLP (roughly 250-1200 weights) and PVRTC1-style bilinear endpoi…