
A Dreadnode proof‑of‑concept demonstrates AI malware that runs locally on on‑device LLMs (Phi‑3‑mini via ONNX), autonomously exploiting misconfigured Windows services to escalate privileges—flagging a nascent threat limited today to high‑end workstations and CoPilot+ PCs but with serious security implications as on‑device AI spreads. Exo Labs built a “frankencluster” pairing an NVIDIA DGX Spark with an Apple Mac Studio to split prefill and decoding work, accelerating Llama‑3.1 8B inference by 2.8× vs. a Mac Studio baseline and showing how heterogeneous homebrew clusters can reclaim compute sovereignty. And AI startup Poolside announced a modular 2 GW West Texas training campus (starting with a 250 MW phase using 40,000 NVIDIA GB300 GPUs), underscoring an escalating infrastructure and power race that could reshape who controls large‑scale AI training.
Comments
Want to join the conversation?