AMD Instinct Coder 355 brings local-first AI coding to large teams

Spectro Cloud and AMD announced on November 10, during KubeCon North America 2026, the AMD Instinct Coder 355, a turnkey appliance that brings AI-assisted coding inside the enterprise data center, without relying on the cloud for everyday tasks.

The appliance combines Dell PowerEdge XE9785 servers with AMD Instinct MI355X GPUs and Spectro Cloud's PaletteAI Inference Launchpad software. The pitch is local-first: open models run locally, and frontier models are only used when the task justifies it and company policy allows it.

What's inside the appliance

  • Dell PowerEdge XE9785 server
  • 2 AMD EPYC 9575F 64-core processors
  • 8 AMD Instinct MI355X GPUs, with 288 GB of HBM3E memory and up to 8 TB/s of bandwidth each
  • 3 TB of DDR5 memory
  • Local GLM-5.3 model, with routing to Claude, GPT and Gemini when needed
  • Compatibility with Claude Code, Cursor and Visual Studio Code

According to the companies, a single node serves up to 85 developers, with around 50 using it concurrently. Spectro Cloud says the solution can cut AI coding token costs by up to 75% compared with frontier cloud models, and AMD says local performance can reach up to 98% of those models' benchmark performance. Both figures come from the vendors and have no published independent verification so far.

Why local-first matters

When inference happens inside the company, source code, prompts and data never have to leave the corporate environment. In the AMD Instinct Coder 355, PaletteAI's intelligent routing keeps routine tasks on the local model and only sends to external services what the team approves, helping control costs and comply with data policies.

The Coder 355 expands the line started by the Coder 325, launched in August 2026 for smaller teams. The new version targets larger engineering teams, with bigger open models, longer context windows and more developers and agents in parallel.

Availability

The AMD Instinct Coder 355 can already be ordered through Spectro Cloud and its reseller and distributor partners. During KubeCon, the solution was demonstrated live at Spectro Cloud booth #125.

Image credit: official Spectro Cloud press image, published with the press release on BusinessWire.

Sources

Transparency: this article was produced with the assistance of artificial intelligence, based on cross-checking public posts and internet sources. Please consult the original sources to confirm the details.

Muse