DeepSeek disclosed large-scale deployments of Huawei Ascend clusters across its new AI data centers. Paired with its 552B-parameter asymmetric MoE model DeepSeek-V4.1-Flash, the architecture achieved record inference throughput and memory efficiency on domestic silicon.

Key Takeaways

  • ✓Deploys Huawei Ascend AI accelerators at scale to secure hardware sovereignty
  • ✓552B asymmetric MoE architecture sets new standards for inference speed and vision-code reasoning
  • ✓Reaffirms DeepSeek global leadership in cost-performance and open-weights efficiency
ADSponsored