🎯 Wide expert parallelism

The items that name it 2

• [AINews] Modal’s $1B ARR, $16B Series D makes it NYC’s newest AI Infra Decacorn, 3x usage and 3x valuation in 5 months, 15x in 1 yr
  article2026-10-15It is written here as Wide expert parallelism (wideEP).

  wideEP

• Long Live the Short King: Why 4-hi HBM Wins
  article2026-09-13

  In MoE models, places few experts on each GPU in one scale-up world so each expert handles larger token batches, at the cost of more all-to-all communication.