Google Cloud‘s BigQuery team published a run-down of performance optimizations delivered throughout 2025, claiming up to 35% better query performance and as much as a 40% reduction in slot usage. The post focuses on three autonomous features that require no manual tuning.

First is History-Based Optimizations (HBO), a self-learning engine that records runtime statistics from past queries rather than relying solely on static statistics. When a similar query runs again, the system automatically applies optimizations known to help that specific pattern and skips those that caused regressions. Safety guardrails measure the outcome: if expected improvements don’t appear, the optimization is rejected and never retried for that query. An enterprise customer reportedly saw P90 execution times drop by up to 50% and slot usage fall by up to 15%.

Second is the BigQuery advanced runtime, which selects the best physical execution path automatically. Two sub-features are highlighted: enhanced vectorization using SIMD instructions and native processing of dictionary and run-length encodings, yielding up to 10x acceleration for eligible queries and up to 40% overall slot reduction; and short query optimizations that reduce distributed-processing overhead for high-concurrency, low-latency workloads, targeting up to 10x lower slot usage and P99 sub-second latencies.

Third is Fluid Scaling, an autoscaler that enables per-second billing for compute slots and claims to lower costs by up to 34% on average for autoscaling workloads. Customer testimony from AdTech firm RISE reports a 25% infrastructure cost reduction.

The post emphasizes that these benefits apply equally to BigQuery‘s native storage and to open table formats like Apache Iceberg. The overall framing positions BigQuery as an autonomous query processor suited for the shift from human-driven analytics to agentic workloads, where thousands of queries per second make manual tuning infeasible.