Understand this story
On November 6, 2026, TMTPost reported that Moxin Technology released MoWorld2.0, a 4D world model built natively on Huawei Ascend. The report stated that the model is trained on Ascend computing power, and its inference cost can be reduced to as low as 30% of that of GPU solutions of the same scale. The article also noted that the deployment of world models still faces challenges such as the high cost of acquiring real physical data, the difficulty of balancing real-time performance and cost, and the need to verify the consistency between generated worlds and the real world. There are currently no subsequent developments.
AI-generated from public reports
Latest development · Report
World models become AI's next battleground, but deployment still faces three hurdlesDevelopments
1 confirmed developments
- Source:TMTPost · AGI
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 6
TMTPost · AGI
World models become AI's next battleground, but deployment still faces three hurdles
A TMTPost article points out that the core proposition of AI is shifting from generation to understanding and interaction, and world models are regarded as the key technological foundation for embodied intelligence. The article reviews the layouts of World Labs, AMD, NVIDIA, as well as domestic players such as BAAI and Kunlun Wanwei, and uses Magic Core Technology's MoWorld 2.0 as an example to introduce its progress in training on Ascend computing power, with inference costs reduced to as low as 30% of GPU solutions of the same scale. The article also proposes that the industry still faces three hurdles: the high cost of acquiring real physical data, the difficulty of balancing real-time performance and cost, and the consistency between generated worlds and the real world still awaiting verification.