Google Adds Vector Search to MySQL Service, Advancing LLM Capabilities

Google has added vector search to its MySQL database service, making it the first to offer this capability ahead of Oracle, which has not yet added any large‑language‑model (LLM) features to MySQL.

Andi Gutmans, Vice President of Google Cloud Databases, said that over the past 12 years Google has rapidly innovated in vectors, now providing preview vector search in several Google Cloud databases, including Cloud SQL for MySQL, Memorystore for Redis, and Spanner.

Vectors are the fundamental building blocks of LLMs; since the launch of ChatGPT in 2022, LLMs have become a focal point for tech giants, governments, and media. LLMs represent words or language components as vector embeddings based on statistical similarity, and Google supports Word2Vec, an early NLP technique.

Percona’s Dave Stokes noted that Oracle’s engineering team has no current plans to add vector functionality to MySQL, criticizing Oracle for focusing resources on HeatWave and providing only minimal features for the community edition.

Other vendors are also adding vector search: PlanetScale announced the feature for its MySQL‑based distributed system last October, Redis now supports vector search in its released versions, and Couchbase introduced vector search in both its DBaaS Capella and Enterprise editions.

Last year, major databases such as Oracle, Cassandra, MongoDB, PostgreSQL, and SingleStore added vector search support, while specialized vector databases like Pinecone have proliferated.

Forrester Research’s Noel Yuhanna said vector search is becoming a standard feature of enterprise databases, with about 35% of enterprises considering vector databases and an expected rise to 50% within 18 months.

He emphasized that vector search is critical for generative AI applications such as similarity search for data, images, documents, customer intelligence, fraud detection, chatbots, and content personalization.

Yuhanna also noted that only roughly 22% of organizations are currently planning LLM/GenAI strategies for their databases, though this is expected to double in the next two to three years, and that widespread adoption of vector‑enabled databases will likely take several years.

Google is also bringing its GenAI models closer to the analytics environment by offering Gemini through Vertex AI for BigQuery users, aiming to provide multimodal and advanced reasoning capabilities.

Integrating Vertex AI, BigQuery, and BigLake can reduce data movement, improve governance and security, eliminate redundancy, and lower management costs.

The trend toward a “Lakehouse” architecture—combining unstructured and structured data on a single platform—is already being adopted by about a quarter of enterprises to lower costs and run BI, data science, AI/ML, and operational workloads together.