Qualcomm plans to repurpose its High Bandwidth Compute data-center architecture into smartphones, personal computers and automobiles to boost on-device AI performance. The first-generation chips will debut in data centers in 2027, with broader commercialization targeted in 2028.
Qualcomm this week launched its High Bandwidth Compute architecture for data centers, marking its first foray into server-grade AI processors. The design stacks memory and compute vertically to enhance data flow and processing speed compared to traditional side-by-side chip layouts.
The company plans to integrate portions of its data-center chip design into smartphones, personal computers and vehicles to enable more advanced on-device AI functionality. Executives are in discussions with manufacturers to tailor the architecture for mobile power and thermal constraints.
High Bandwidth Compute moves memory modules closer to compute engines by stacking chips, reducing latency and increasing bandwidth. This approach aims to support "always-on" AI tasks on handheld devices without excessive battery drain.
Qualcomm expects to debut the first-generation data-center chips in 2027, with full commercialization of the architecture scheduled for 2028. The timeline for integrating this technology into consumer devices has not been finalized but is targeted after the initial server rollout.
