Alibaba Targets 20GW AI Capacity by 2032 with New Chips
Alibaba has outlined a full-stack AI strategy at its Apsara Conference, aiming for 20GW of data center capacity by 2032 through new proprietary chips and a roadmap for 10-trillion-parameter models. The update includes the confirmation that Qwen 4 is in training, alongside significant advancements in autonomous self-improvement techniques and next-generation multimodal audio and speech models.
According to Macau Business, Alibaba has confirmed that its next-generation Qwen 4 model is currently in active training, with a roadmap targeting 5 to 10 trillion parameters for subsequent versions like Qwen 5. The company also highlighted progress in recursive self-improvement, noting that its latest models autonomously optimized chip design workflows to reduce area by 42% while simultaneously upgrading its multimodal audio and speech capabilities for lower latency.
Source: Macau BusinessAlibaba unveils its next-generation AI strategy, including new proprietary chips and a roadmap for models with up to 10 trillion parameters.
Source: DagangNews
