AlloyDB AI Functions: 23,000x Faster, 6,000x Cheaper

Google’s AlloyDB just dropped some serious performance and cost improvements for its AI functions. We’re talking up to 23,000x faster and 6,000x cheaper—enough to make even the laziest capybara sit up and take notice.

What Are AI Functions?

AI functions bring Gemini’s smarts directly into your SQL queries. Instead of building complex pipelines to extract insights from unstructured data, you can call ai.generate, ai.rank, ai.if, and ai.forecast right in your database. These are now Generally Available.

New functions include ai.summarize, ai.agg_summarize, and ai.analyze_sentiment. For example, ai.agg_summarize can condense thousands of product reviews into a single, coherent summary per product group.

Smart Batching: Up to 2,400x Faster

Smart Batching intelligently groups AI function calls, deduplicating prompt overhead. AlloyDB automatically determines the optimal batch size, handles retries, and can process 10,000 rows per second—a 2,400x improvement over row-by-row LLM calls. Available in preview for ai.if and ai.rank.

Optimized AI Functions: 23,000x Faster, 6,000x Cheaper

For even bigger gains, Optimized AI Functions train a lightweight proxy model on your data. This model runs locally, slashing costs to 1/10th of a cent per query and processing up to 100,000 rows per second. If accuracy dips, AlloyDB automatically falls back to the full LLM. Currently in preview for ai.if.

Getting Started

• New to AlloyDB? Grab a 30-day free trial.
• Enable AI functions with a few prerequisites and start using them in SQL.
• For maximum performance, follow the optimized functions guide.
• Scale throughput with smart batching or array-based functions.

Check out our Google Cloud review for more on AlloyDB and other GCP services.

Original announcement published on Google Cloud.