The company iterated and upgraded its proprietary AI silicon and compute servers, with dedicated models for LLM training and inference officially entering the mass production stage.

The new generation servers use the latest proprietary AI silicon, supporting efficient training of hundred-billion parameter LLMs, with single-machine training performance improved by 150% and energy efficiency optimized by 40%. The servers have formed a scalable product line, supporting flexible delivery by rack, cage, or zone.

The mass production of dedicated servers marks that the company has scalable supply capabilities in the computing hardware field, meeting the bulk procurement needs of large customers.