AI clusters

Homa, a transport protocol designed as a TCP replacement for data centers, is being presented as the end of TCP for AI clusters. The claim is that traditional TCP cannot handle the demanding traffic patterns of AI training workloads, and Homa's approach offers lower latency and better performance for machine learning infrastructure.

Reports circulating in tech circles claim that Alibaba's Qwen 3.8 Flash Next, a 125-billion-parameter model, can run at roughly 100 trillion tokens per second on a single consumer RTX 4090 GPU — a throughput previously associated with multi-node H100 clusters. Enthusiasts are discussing what this means for local AI inference and the collapsing cost barrier between consumer and data-center hardware.

A week of releases across the Linux ecosystem: OpenMandriva ROME 26.09 and Nitrux 7.0 shipped, Ubuntu 26.10 entered beta, and updated versions of Firefox 157, Wine 11.19 and Git 2.56 landed. In project news, the COSMIC desktop has stopped accepting AI-generated pull requests, a decision drawing attention among developers. The roundup bundles the main open-source headlines from the past week.

AI safety, real-world robotics and digital identification are being covered together as a cluster of emerging technology issues. The discussion links concerns about controlling advanced AI with the growing deployment of robots outside labs and the spread of digital ID systems, raising questions about regulation, security and how these technologies will affect everyday life.