@misc{indiciae54b9a035bc0c,
  title = {Beyond Capacity: Scalable MoE LLM Inference via High-Bandwidth Flash with Direct GPU and HBM Paths},
  author = {Seeyeon Kim and Juhyeong Jin and Joo-Young Kim},
  year = {2026},
  url = {https://arxiv.org/abs/2608.14333},
  note = {Source identifier: 2608.14333}
}