AMD has launched a turnkey box that brings AI coding assistants in-house. The AMD Instinct Coder 355, announced on November 10 at KubeCon North America in Salt Lake City, is an on-premises appliance built for enterprise development teams that want to run AI coding tools on infrastructure they own and control, rather than sending proprietary code to a cloud service.
The launch was a joint effort from AMD, Dell, Supermicro and Spectro Cloud. It is the second product in AMD's Instinct Coder line, following the Instinct Coder 325, which arrived in August 2026 and, according to Spectro Cloud, has seen strong demand from organizations running AI coding inference on their own hardware.
Eight MI355X GPUs under the hood
The appliance is built around a Dell PowerEdge XE9785 server packing eight AMD Instinct MI355X GPUs. Each GPU carries 288 GB of HBM3E memory with up to 8 TB/s of memory bandwidth, and the system pairs them with two AMD EPYC 9575F 64-core processors and 3 TB of DDR5 system RAM, according to Spectro Cloud's launch materials.
The larger GPU count is the point of the upgrade. Where the earlier Coder 325 served smaller teams, the MI355X hardware in the Coder 355 can run larger open-weight models, handle longer context windows, and serve more concurrent developers and AI coding agents at once. Spectro Cloud says one box can support about 50 developers working concurrently.
Local-first, with a cloud escape hatch
The appliance runs the GLM-5.3 model locally through Spectro Cloud's PaletteAI Inference Launchpad, the software layer that manages deployment, metering, quotas and governance. When a task is too hard for the local model, the software can route it out to a frontier service — Anthropic Claude, OpenAI GPT or Google Gemini — provided IT policy allows it. Teams can also bring their own models.
Spectro Cloud claims the box cuts AI coding token costs by up to 75 percent compared with cloud frontier models, and delivers up to 98 percent of frontier-model benchmark performance. Both figures come from the vendor; no third-party workload mix has been published. AMD describes the saving in its own materials as up to 75 percent lower total cost of ownership.
A crowded week for AI hardware
The Coder 355 arrives as on-premises AI hardware gains momentum. Enterprises are weighing the productivity gains of AI-assisted coding against concerns about sending proprietary code to third-party APIs, and vendors from every part of the stack are responding with boxes that keep the work inside the building. The appliance competes directly with cloud-based tools such as GitHub Copilot and Cursor, as well as other on-premises offerings from silicon vendors and systems integrators.
The AMD Instinct Coder 355 is available to order through Spectro Cloud and authorized resellers, with live demonstrations running at the company's KubeCon booth. No list pricing or lead times have been disclosed in the public announcement.



