old: put storage next to your compute new: put compute next to your storage
Hugging Face co-founder Julien Chaumond says modern data infrastructure reverses traditional design by placing compute next to storage
Moving massive datasets to compute pipelines is increasingly inefficient
539435.1K
8:03 AM · Jun 15, 2026 · 5.1K Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS45LIKES1

Noé Flandre@NoeFlandre
@julien_c next: put your storage on hf buckets
11hViews 45Likes 1
RETWEETS2
Julien Chaumond@julien_c
old: put storage next to your compute new: put compute next to your storage
12hViews 5.1KLikes 39Bookmarks 3

Tech Actu@Tech_Actu_eu
@julien_c La data gravity rend ça presque inévitable : quand un checkpoint pèse des To, déplacer les poids coûte plus cher que le calcul. Rapprocher le compute du stockage, c'est arrêter de payer le réseau à chaque itération.
10hViews 30

Rob Phillips@iwasrobbed
@julien_c Storage becomes your compute
10hViews 19

Oleg kAI@oleg_kai
@julien_c true for inference-heavy. training still needs shared optimizer state on the wire. when every gradient step touches every shard, the bottleneck moves from compute-to-storage to storage-to-storage. fabric still wins for training.
6hViews 2