The Hugging Face Blog announcement highlights a significant infrastructure friction point: egress fees often dictate where AI teams run compute, effectively creating storage-based vendor lock-in. While eliminating read-side egress costs could provide more flexibility in scheduling jobs across reserved capacity, the practical impact may be narrower than suggested. In our view, the real test will be whether this integration meaningfully changes behavior for large-scale training runs, where the volume of checkpoint writes—still subject to egress—often rivals reads.
The partnership appears strategically aimed at making Hugging Face's storage a neutral hub in a fragmented cloud market, but its success likely depends on whether teams prioritize this flexibility over the performance and simplicity of keeping compute and storage within a single cloud provider's ecosystem.