Google Cloud Previews Sub-File Caching to Slash Cross-Cloud Object Storage Egress

Google Cloud has released preview capabilities for its Borderless Lakehouse architecture, introducing intelligent cross-cloud caching alongside BigQuery cross-cloud connections. The new caching layer operates at sub-file block granularity for columnar formats like Apache Parquet, transparently retrieving and caching only the specific column chunks and dictionary pages required by queries hitting remote Amazon S3 and Azure Storage buckets. The accompanying cross-cloud connections feature allows BigQuery compute workers in Google Cloud regions to query raw external object paths (including Parquet, CSV, and JSON) without requiring pre-configured Iceberg catalogs, while enforcing default Google-managed encryption at rest and strict tenant isolation. For enterprise data and DevOps engineers, this release addresses the persistent financial and operational pain points of multi-cloud data estates. Historically, querying data stored in external object storage meant absorbing steep variable egress fees, dealing with network jitter, or engineering complex replication jobs that resulted in data staleness and duplicate storage costs. By pairing Apache Iceberg and Parquet columnar compression with local sub-file block caching, initial benchmarks demonstrate that organizations can reduce remote cross-cloud data transfers to under 5% of total scanned volume on repeat analytical and AI workloads. This development reflects a decisive industry shift away from monolithic data consolidation toward federated, in-place query architectures. As large language models and analytical agents demand unified context spanning disparate object stores, database silos, and SaaS applications, cloud providers are increasingly competing on open table interoperability (such as the Apache Iceberg REST catalog specification) rather than proprietary storage lock-in. Google's integration of local edge caching with Partner Cross-Cloud Interconnect private links aligns compute directly with distributed storage topologies across hyperscalers. In practice, infrastructure teams should evaluate their current cross-cloud ETL pipelines to identify read-heavy datasets residing in S3 or Azure Blob that can transition to zero-copy federation. While sub-file caching drastically suppresses continuous egress overhead, architects must still account for the baseline interconnect link fees and ensure query governance models map cleanly across AWS IAM and Google Cloud IAM roles. Regional data residency must also be audited, as the cache and query execution anchor strictly to the destination Google Cloud region.