AMD's Instinct Coder 355 puts Claude Code and Cursor on local GLM-5.3
A bid to pull AI coding off metered frontier APIs: a turnkey on-prem appliance, confidential inference on Google's TPUs, and a ready-to-code local box.

Copy markdown
Your coding agents, pointed at a local model
AMD's Instinct Coder 355 — with Supermicro and Spectro Cloud — is a turnkey on-prem box running GLM-5.3 that plugs straight into Claude Code, Cursor, and VS Code, with PaletteAI keeping requests local and routing to Claude, GPT, or Gemini only when policy allows. AMD models up to 75% lower TCO for 85 developers and a ~5-month payback versus per-seat API bills, though that figure is modeled, not measured.
Confidential inference lands on Google's TPUs
AMD's own release confirms its EPYC CPUs now host Google's new TPU 8tx for confidential inference, using SEV to extend hardware-attested isolation to the accelerators — a reversal from April's AMD-free TPU line. MarkTechPost reports a companion TPU 8i Max that doubles inference compute, so serving models on regulated data with attested isolation is now on Google's roadmap.
A ready-to-code local AI box, zero setup
AMD and Microsoft's Project Zenith ships Ryzen AI Halo machines with a tuned Windows 11 dev environment preloaded — VS Code, Git, GitHub CLI, Copilot CLI, WSL, and Python/Node/.NET — plus up to 192GB of unified memory for running LLMs and agents locally. It's on sale now at Micro Center, and you can replicate the exact setup on an existing Ryzen AI Halo through Microsoft's Windows Developer Config on GitHub.