OWC revealed the Stack AI device promising to solve local LLM processing limitations by expanding GPU memory via Thunderbolt. The industry currently relies on expensive cloud services like OpenAI and Anthropic, pushing users toward local solutions despite high hardware costs and privacy concerns. Early adopters will need to evaluate cost-effectiveness.

Resembling a Mac mini, the Stack AI sits beneath a Mac Studio and uses Thunderbolt 5 connectivity to extend VRAM with onboard flash memory. Unlike an eGPU, this external memory enhancement focuses specifically on handling larger language models. Initial support covers Windows and Linux, with Mac compatibility coming later.

The device addresses a critical memory limitation for AI researchers and businesses unable to afford high-end Macs priced at $5,099 for 128GB models. With OWC's solution, users could purchase Macs with desired processing power without maximum memory investments. OWC will showcase the device at Computex Taipei starting June 2.