AI training data storage has always been a major challenge at the infrastructure layer. Managing the upload of massive small files not only burns money but also involves dealing with high on-chain transaction fees, and the operational process is painfully complicated.

Gata, as an execution infrastructure for AI, recently integrated Walrus's Quilt batch storage solution, which is seen as the key to solving these issues. In simple terms, Quilt automatically packages a bunch of small files into a single storage unit—how effective is this? The storage overhead for 10KB-level files is directly reduced by 420 times. This number may sound incredible, but it indeed results from reducing redundant storage metadata and gas fees on the SUI chain.

Even more impressive, Quilt supports individual access to files within the package, allowing quick retrieval of target datasets without unpacking. This is perfect for the high-frequency data reads required in AI training. Behind this is Walrus's Red Stuff encoding technology, which ensures high data availability while significantly lowering storage costs.

In other words, Gata can now focus more on innovation at the AI execution layer, and developers can access cheaper, more efficient AI training infrastructure. The significance of this collaboration is to gradually loosen the bottleneck at the infrastructure layer.