TSMC Records Signal AI Boom, Yet Investors Hesitate
TSMC delivered a fifth straight record quarter at $40.2B revenue, lifting its full-year growth forecast above 40% and guiding $60-64B in 2026 capex,...

Created by Yuzhou He
Global AI research, product, startup, and infrastructure news with no regional bias
Explore the latest content tracked by Global AI Pulse
TSMC delivered a fifth straight record quarter at $40.2B revenue, lifting its full-year growth forecast above 40% and guiding $60-64B in 2026 capex,...
AI agents dominate July's trending repos, shifting from research papers to practical tools.
Claude Code now uses an unreleased Bun v1.4.0 preview written in Rust rather than the public v1.3.14 release. The change marks Anthropic's direct control over the runtime powering its AI coding agent.
Biren is advancing near-packaged optics to link up to 1,024 AI accelerators in supernodes, overcoming the physical limits of copper connections that...
EchoScribe delivers fully local transcription and summarization using Whisper models, keeping sensitive meetings and interviews off the cloud...
Autonomous coding agents plan, execute, and self-correct entire features end-to-end, unlike reactive copilots limited to line suggestions.
Generic Kubernetes multi-cluster advice breaks down for AI clouds, causing idle GPUs, security gaps, and costly provisioning delays.
Open-source platforms now let users build LLM apps, RAG systems, and agents through visual canvases and natural language prompts instead of code.
-...
Enterprise AI agents dynamically select tools and adapt execution at runtime, unlike static service accounts.
Agentic AI workloads demand high-core CPUs for orchestration and multi-task coordination, unlike GPU-centric training, with projections of 4x core...
Investors are rotating from AI infrastructure plays to companies leveraging AI at scale and cybersecurity providers, signaling a maturing market. IBM...
Chat2Scenic introduces the first iterative RAG framework that turns regulatory texts into executable DSL scenario scripts for autonomous driving...
Two new techniques demonstrate how frozen or small models can gain capability and slash costs without retraining.
GPT-5.6 Sol is emerging as a versatile system, not just a language model, with breakthroughs spanning multiple domains.
LLMs can synthesize reliable clinical notes from existing unimodal dermatology datasets to train multimodal models that outperform state-of-the-art foundation models on cross-modal retrieval and zero-shot tasks across 15 datasets.