Lenovo announced on November 10 its Wanquan hybrid token mother factory in Zhangjiakou, targeting a 15000P heterogeneous compute cluster by 2027
The mother factory is designed for an annual token output above 20 quadrillion, a fivefold gain in tokens per second, a 70 percent reduction in blended cost per million tokens and a PUE held under 1.1, while the Changsha project targets 1200 hours of finished footage per month
On November 10, Lenovo held the Wanquan hybrid token factory brand launch in Zhangjiakou and announced that the first Wanquan hybrid token mother factory in China is landing there. Zhangjiakou is a node in the national east data west computing Beijing-Tianjin-Hebei hub, the country top wind and solar green power city and its only national level renewable energy demonstration zone, as well as a benchmark dual node and a pilot for coordinated computing and electricity, which suits the green development needs of artificial intelligence compute.
At the event Lenovo described a three layer hybrid system for enterprise intelligence. The bottom layer is the Askcloud artificial intelligence infrastructure, the middle layer is the Wanquan hybrid token factory carrying the Lenovo intelligent cloud and intelligent computing center capabilities, and the top layer is the Qingtian and Baiying agent solutions and services. According to the company, the factory supports an annual token output above 20 quadrillion, a fivefold increase in tokens produced per second, a yield rate above 99.9 percent, a 70 percent reduction in blended cost per million tokens and a PUE held under 1.1.
For the national rollout, Zhangjiakou will be the core with a heavy industry transformation plant in northeast China, a smart manufacturing plant in north China, an education oriented deployment in northwest China, short drama generation in central China and advanced manufacturing in east China, with Changsha next to build a new media token factory and artificial intelligence film and television innovation center with Mango TV. The long term goal is a 15000P heterogeneous compute cluster and an intelligent computing center of roughly 100MW by 2027, annual token output above 20 quadrillion, peak token output of 200 million per second and maximum concurrent users above 100 thousand.