Technology / Business

Add as a preferred source on Google

AMD and Spectro Cloud Offer Dell-Based AI Coding System to Companies

AMD models annual costs about 75% below an assumed $2.04 million cloud bill, but its disclosed inputs do not fully explain its projected $513,500 total.

Updated 2 min read
Technicians inspect a GPU server chassis inside a hardware integration bay in Round Rock, Texas. / TokenPost.ai (macro)
Technicians inspect a GPU server chassis inside a hardware integration bay in Round Rock, Texas. / TokenPost.ai (macro)

Advanced Micro Devices (AMD) and Spectro Cloud announced a Dell-based AI coding system Nov. 10 that runs some developer requests on company-owned hardware, potentially reducing spending on external models.

The AMD Instinct Coder 355 puts AMD Instinct MI355X graphics processors and EPYC 9575F processors in Dell PowerEdge XE9785 servers. Spectro Cloud’s PaletteAI Inference Launchpad runs the GLM-5.3 model locally. AMD says one server node can support up to 85 developers, including 50 working at once.

AMD’s cost example assumes 85 developers each spend $2,000 a month on cloud AI coding services, or $2.04 million a year. It projects annual costs of $513,500 for the new configuration, about 75% less than that assumed bill. The estimate is based on AMD’s chosen usage and pricing assumptions, rather than measured customer savings.

The published assumptions include a $590,000 server list price, with software and licensing included, spread over three years. They also include $204,000 a year for external model access and $50,000 for data center operations. Those inputs add up to about $450,667 annually, roughly $62,833 below AMD’s projected total. Its published breakdown does not account for the difference.

Companies can set rules for when requests remain on their servers and when they may go to approved external models, including Anthropic Claude, OpenAI GPT and Google Gemini. The approach shifts some work away from metered external services, while leaving buyers responsible for the server, its operation and any outside model use.

Processing a request locally can keep its code and prompt within the company’s environment. Sending it to an external model requires passing along the content needed to handle that request. Companies can also set usage budgets and quotas for developers.

AMD, Spectro Cloud and Supermicro announced an earlier plan Aug. 5 that called for an initial configuration using Supermicro systems, subject to final validation. Spectro Cloud said the Dell-based system can be ordered through the company or its reseller and distribution partners.

Related articles

View all