AI Infrastructure news and analysis on Web Pulse: 31963 curated articles tracking the latest AI Infrastructure developments, tools, and research, updated continuously from vetted sources.
Marvell Technology surged approximately 30% in a single day after Nvidia CEO Jensen Huang identified the company as a potential "next trillion-dollar" player, highlighting the intense market focus on AI infrastructure. T…
AISOP launched an open protocol that defines AI agent workflows using Mermaid or JSON flow graphs, enabling multi-step programs with branching, parallel execution, and error handling in a single portable JSON format. The…
Zilliz launched the public preview of Vector Lakebase, a semantic-centric data platform that unifies open storage and elastic compute for AI workloads. The platform supports three workload modes—real-time retrieval, iter…
A team running a 70B Llama 3 fine-tune at 200 requests per second cut median time-to-first-token from 380 ms to 140 ms on the same hardware by implementing speculative decoding. The technique addresses memory-bound GPU u…
The European Commission has unveiled its tech sovereignty package aimed at reducing Europe's reliance on U.S. cloud services, chips, AI infrastructure and digital tools. Former MEP Marietje Schaake and POLITICO’s Laurens…
FerryAPI has introduced a four-layer cost attribution stack for SaaS teams that call multiple LLM providers, enabling per-request tracking by tenant, feature, assistant, thread, and model. The system uses an OpenAI-compa…
LGND AI won the 2026 Snowflake Startup Challenge, beating out competitors Airrived and Twine Security. The company builds Large Earth Models trained on 800 petabytes of satellite imagery to make geospatial data queryable…
Bernstein initiated coverage of TeraWulf and Cipher Mining with Outperform ratings and price targets of $36 and $32, respectively, citing the companies' transition from Bitcoin miners to "power landlords" for AI data cen…
Basecamp launched a command-line interface and agent-native tools that allow AI agents like Claude and Cursor to perform any user action within the platform, including writing documents, managing to-dos, and answering ch…
A developer has released LLMhop, a stateless HTTP router that directs OpenAI-compatible API requests to the appropriate LLM inference backend based on the model name in the request body. The single-binary tool, built in …
AWS offers over 15 purpose-built database engines across eight categories—including relational, key-value, in-memory, document, graph, wide column, time-series, and data warehouse—to match specific workload requirements …
New York’s Legislature is moving to ban new data-center construction for one year, citing concerns over water use, power grid strain, and pollution that critics call unfounded. The moratorium threatens to deny Empire Sta…
A developer has published a guide for building a Model Context Protocol (MCP) server from scratch using the bare JSON-RPC 2.0 protocol, bypassing common frameworks to give engineers full control over the wire format. The…
Docker published a post this week arguing that AI agents require execution isolation, tool access control, and runtime monitoring, while stating that permission prompts alone are insufficient as a security model. The dev…
Vercel launched Sandbox Drives in private beta, offering persistent, attachable storage with a lifecycle independent from any sandbox. Users can create a drive once and mount it at a configurable path across multiple san…
A laptop now handles 78% of one user's AI tasks locally, with only complex work routed to the cloud. The system, which classifies tasks as easy or hard and processes them accordingly, achieved 25% more throughput and cut…
Modular has introduced a five-stage composable routing system for large language model inference, replacing traditional fixed algorithms like round-robin and consistent hashing. The system, detailed in the third part of …
The modern cloud stack was built over two decades for human developers, with per-seat pricing, dashboard-based control surfaces, and provisioning measured in minutes — assumptions that break down when AI agents are the o…
An app built with AI costs little to create but hundreds of dollars per month once real users arrive, as each production need—authentication, file uploads, email delivery, error tracking, and abuse protection—requires a …
NVIDIA used multi-tier on-policy distillation to train its Llama-3.1-Nemotron-Ultra-253B model, enabling it to outperform larger systems on reasoning tasks without additional compute. The technique addresses distribution…