The company plans to deploy modular GPU units across 40+ global sites to provide on-demand AI inference capacity for small and mid-sized businesses.