AI Infrastructure

Worldmodeldata, a Cambridge-based deeptech startup, raised £7 million in seed funding and appointed former Meta policy VP Chris Kelly to its board. The company aims to build the largest library of AI training data from v…

Cloudflare and AWS have both integrated the x402 open protocol into their edge networks, enabling agent-to-service micropayments with stablecoin settlement. The protocol revives HTTP's 402 status code for payment-gated r…

Amazon has dropped the price of the MSI Gaming RTX 5080 16G Inspire 3X OC GPU, a high-end graphics card with 16GB GDDR7 memory and a compact triple-fan cooler, making it an attractive option for gamers and creators seeki…

Hugging Face announced major updates to its Kernels project, introducing a new repository type on the Hub, improved security with trusted publishers and code signing, revamped CLIs, and expanded framework and backend sup…

The AI coding agent ecosystem is rapidly maturing, with new open-source tools like Orca for orchestrating parallel agents and AgentsView for cross-agent analytics emerging alongside research quantifying how code cleanlin…

AMD has released the Primus Tuning Agent, a tool that automatically searches for optimal training configurations for large language models by using a projection engine as a fast scoring oracle. In a case study with Mixtr…

AMD Quark, a quantization library optimized for AMD Instinct MI350 GPUs, enables MXFP4 quantization for Diffusers and xDiT FLUX.1-dev image generation, achieving up to 1.92× speedup over BF16 eager and 1.41× over BF16 to…

Verihubs, an Indonesian AI company, rebuilt its data warehouse on ClickHouse, achieving up to 98% faster queries and 50% lower cloud costs. The new architecture replaced daily Postgres batch pulls with real-time streamin…

Open-weight model GLM-5.2 matched Claude Opus at a fraction of the cost, global LLM token spending fell 20% from its May peak, and a GPT-5.5 Codex flaw caused quiet output degradation, signaling a market shift from model…

New research from Elastic reveals that one in three Australian businesses exceeded their AI budget last year, with nearly a third pausing or winding back deployments due to unjustified costs. Despite this, half plan to i…

Security researchers at MBG Security built a global network of honeypots with exposed AI inference and agent endpoints, observing attackers actively scanning for and exploiting vulnerabilities like CVE-2026-40217 to gain…

Anyshift's AI agent, Annie, now integrates with Elasticsearch to provide SRE teams with AI-powered context during incident investigations, enabling natural language queries of log data. The integration offers read-only a…

Tencent released Hy3, a 295B-parameter Mixture-of-Experts open model with 21B active parameters and a 3.8B MTP layer, following the Hy3 Preview launch in late April. Hy3 scored 2.67/4 on benchmarks, outperforming GLM-5.1…

AI agent observability is essential for monitoring long-running autonomous workflows that can run for hours without human supervision. The four core signals to track are execution traces, cost tracking, latency, and fail…

Frontier AI models like GPT-4o and Claude 3.5 Sonnet excel at novel, complex tasks but cost 50-200x more than cheap models like GPT-4o mini and Claude Haiku. Smart routing between tiers can cut inference costs by 90% whi…

Anthropic's Claude Sonnet 5 and Opus 4.8 models present a cost paradox: Sonnet 5 is cheaper per token but can cost more in agentic multi-step workflows due to higher error rates and retries, while Opus 4.8's higher per-t…

Andreessen Horowitz released a document arguing that the United States must lead global AI development by establishing the infrastructure through which all nations access the technology, effectively owning the platform. …

GLM 5.2, an open-weights model from Z.ai, has emerged as a genuine competitor to frontier models like Opus and GPT, but its slow inference speed and lack of vision and web search capabilities limit its practical use. The…

Anthropic released Claude Science in beta on June 30, 2026, a daemon-centered agent harness for scientific execution, durable artifacts, and verification. The system splits kernel working state from daemon authority, ena…

At the AI Engineer World's Fair in San Francisco, the central theme was that AI models are no longer the bottleneck; instead, human attention has become the scarce resource, as teams adopt multi-agent loops and software …