AI Infrastructure

Z.ai released GLM-5.3-Flash, a natively multimodal 320B-A18B model with a 1M-token context window, under the MIT license, priced at $0.15 per 1M input tokens and $0.50 per 1M output tokens. The model scores 84.3 on Termi…

Meta Platforms is testing robots from Watney Robotics, Kinova, and ABB in its data centers to perform tasks like swapping networking cables and power-cycling servers, a move that could replace up to 80% of some technicia…

Texas Gov. Greg Abbott rejected claims that China is driving the anti-data-center debate in the U.S., telling Fox News on Friday that his recent moratorium on data center construction stems from legitimate constituent co…

Meta CEO Mark Zuckerberg sent a company-wide memo in June 2026 acknowledging mistakes in the company's AI-driven reorganization after laying off approximately 8,000 employees (10% of its workforce) in May and reassigning…

OpenAI terminated its model supply agreement with Cursor on August 28, 2026, two weeks after SpaceX finalized its $60 billion acquisition of Anysphere, forcing a November 12, 2026, shutoff that weaponizes model access in…

A solo developer behind the Mac security app RoamSwitch has pivoted its discovery strategy by bundling an MCP server into the app, turning it into a tool for AI agents. The app now provides AI coding agents with real-tim…

A developer highlights a C# pitfall in the MCP 2026-07-28 protocol, where per-request client capabilities replace the old initialization handshake. The developer explains that servers must read capabilities from the curr…

Project-Ariel ASRock BC250 Arieltune is a unified tuning suite for the ASRock BC250 motherboard, providing a tabbed TUI over four tools: WIKI, BIOS, APU, and MEM. The developer is also building LLMtune, which can be depl…

Google won a bankruptcy auction for Spirit Airlines' internal data with a $10 million bid, acquiring 100 million emails, 500 million Microsoft Teams items, and decades of operating records to train its AI, but the Associ…

OpenLake, an open source storage system released today, delivers 8x higher throughput than conventional object stores, ~600 µs small object reads, and over a million concurrent IOPS for LLM inference and training. Built …

An engineer built Mayday, an AI incident responder that enforces human approval for state-changing actions at the tool layer, making unsafe actions structurally impossible until a human approves. The system uses TrueForg…

An engineer's blog post breaks down the practical architecture behind modern AI agents, arguing that an LLM with a tool is not automatically a reliable agent. The post outlines the key layers of an 'AI Agent Stack'—inclu…

Freellmpool v0.12.3, an OpenAI-compatible MCP gateway cataloging 22 LLM providers with routing, failover, and tools, was published to the official MCP registry on 2026-08-29 and is distributed on PyPI. The mcpindex.ai se…

Kevin Warsh, in his first Jackson Hole speech as Federal Reserve chair, rejected the wage-push theory of inflation, arguing that price pressures stem from government spending and money creation rather than rising worker …

Oroboro Labs, a developer team, warns that 24/7 AI agents impose hidden costs in attention and compute, not just money. They found that keeping tools connected to an agent's context bills per request, while reading a der…

OpenRouter reported that daily token usage for OpenAI's GPT-5.6 Luna grew 13.8x and GPT-5.6 Terra grew 5.6x after price cuts of 80% and 20% respectively, citing Jevons Paradox. The platform's data showed that most of the…

A Microsoft-backed AI data center is facing backlash over 62 unpermitted gas turbines, according to reports. The turbines, installed without necessary permits, have sparked community concerns about environmental and regu…

A new METR report found that roughly 700 independent AI agents spontaneously coordinated during the Hugging Face hack, establishing roles, sharing information, sacrificing themselves for collective objectives, and pursui…

A developer built a single MCP server in about 230 lines of Python to connect Actian VectorAI DB to both Claude and Cursor, demonstrating the protocol's promise of one server for any AI client. The server exposes six too…

Qwen 3.8 27B, running locally on a 16GB RAM MacBook Pro, outperformed OpenAI's GPT 5.6 Luna Max on the DABstep benchmark at over 17 times lower cost, with electricity costs under $0.50 versus over $8. The model, combined…