AI infrastructure booms; open-weight rift grows

Anthropic / Claude ecosystem

Anthropic's Opus 5 Sets New Benchmark High on ARC-AGI-3 Intelligence Test

Claude Opus 5 achieved a new high score on the ARC-AGI-3 benchmark, designed to measure autonomous reasoning in unfamiliar environments. The result indicates significant improvements in planning and general intelligence compared to previous models.

Frontier model providers

Google Launches Gemini Spark for AI Pro Users in the U.S.

Google has begun rolling out Gemini Spark, a new AI assistant, to Google AI Pro subscribers in the United States. The release introduces enhanced task management and cross-app integration features.

Cloudflare and OpenAI Pilot Program to Enhance AI Search with Network Insights

Cloudflare and OpenAI have initiated a joint pilot project that leverages Cloudflare's global network data to improve AI search efficiency. The collaboration aims to boost the speed and relevance of AI-powered search queries.

AI developer tooling & infrastructure

Ollama Releases v0.32.4-rc0 with Bug Fixes for Local AI Inference

Ollama has released version 0.32.4-rc0, a patch focusing on stability improvements and bug fixes for local large language model serving. The update underscores the rapid iteration in open-source AI tooling.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

OpenAI and Anthropic Decline to Join Open-Weight Regulation Opposition, Deepening Industry Rift

OpenAI and Anthropic have refused to sign the open-weight AI models letter that now counts 50 supporters, highlighting a split in the AI industry over model openness. Their abstention signals a preference for stricter regulation, especially as China advances its distillation capabilities.

Canada Launches Public Consultation on AI Transparency Framework

The Canadian government has opened a public consultation to develop a comprehensive AI transparency framework as part of its AI for All strategy. Input is sought on practical measures for AI accountability and disclosure.

South Korean Lawmaker Proposes Role-Based AI Content Labeling Bill

Assemblyman Ahn Cheol-soo of South Korea has introduced a bill that splits AI labeling obligations between developers (metadata) and service operators (visible labels), aiming to reduce legal uncertainty while improving transparency. The amendment updates the existing AI framework act.

Industry & market moves

AWS Funds Lean-Focused Research to Bring Mathematical Guarantees to AI

Amazon Web Services is investing in the Lean Focused Research Organization to advance the Lean programming language for formal verification of software, including AI systems. The goal is to enable mathematically proven trust in agentic AI behaviors.

South Korean internet giant Naver has secured a $10 billion investment from Nvidia and Brookfield to build a gigawatt-scale AI factory, transforming from a domestic player into a global AI infrastructure provider. The deal highlights cross-border funding for sovereign AI capabilities.

SK Group and NVIDIA Expand Partnership to Build 2-GW AI Factories and Next-Gen Memory

SK Group and NVIDIA have deepened their strategic partnership, planning to build 2 gigawatts of AI factory capacity and co-develop next-generation AI memory technologies as part of a $500 billion-plus initiative. The collaboration spans multiple subsidiaries including SK hynix.

AI product & feature launches

AMD Launches Helios Rack Delivering 2.9 ExaFLOPS for Cloud-Free AI Clusters

AMD has unveiled the Helios Rack, its first fully integrated AI infrastructure system, capable of 2.9 exaFLOPS. Designed for on-premises deployments, it enables large-scale AI training and inference without reliance on cloud providers.

Fastly Partners with Experian to Verify AI Agent Identity at the Edge

Fastly has joined Experian's Agent Trust ecosystem, enabling edge-based identity verification and authorization for autonomous AI agents. The integration comes as over half of web traffic is now automated, raising new trust challenges.

KAIST and NVIDIA to Build 'Physical AI Brain' Predicting Human Motion

KAIST and NVIDIA are collaborating to create a Human Motion Foundation Model, a general-purpose AI that predicts and understands human movement for applications in wearables, humanoids, and digital healthcare. The project is part of a broader $300M research initiative.

China Unveils Yuanyan-1, an AI-Powered Autonomous Electron Microscope

China has introduced Yuanyan-1, an intelligent transmission electron microscope that uses AI to autonomously perform high-precision imaging, potentially revolutionizing materials science and biology research. The system replaces manual operation with automated workflows.

Research with immediate practical relevance

Google's Gemma Tech Enables 28.9M-Parameter LLM on ESP32 Microcontrollers

Google has demonstrated a 28.9 million parameter language model running entirely on off-the-shelf ESP32 microcontrollers, enabling offline, air-gapped inference on commodity IoT chips. This breakthrough paves the way for edge AI in low-power devices.

China Achieves Breakthrough in Brain-Computer Interface with Large-Scale EEG Data Collection

Chinese researchers have developed a system to synchronously collect EEG data from thousands of participants across regions, enabling massively larger datasets for training AI on brain signals. This advance could accelerate neurotechnology and AI integration.