Database and AI Trends: Supabase Raises $500M, Google's Gemma 4 Shrinks, and Apache Livy Graduates

Supabase raised $500 million in its latest funding round and announced plans to adapt Vitess for better horizontal scaling of Postgres. The database platform teased a new offering called Multigres that aims to solve one of Postgres's long standing limitations: sharding at scale. Meanwhile Google introduced Gemma 4 12B, a small language model that nearly matches the benchmarks of the 26B version and can run on a standard laptop. Separately Apache Livy, the REST service for Apache Spark, has graduated to a top level project after nine years in incubation.

The Supabase funding round, reported by The Stack, positions the company to compete more directly with cloud database giants. Its Multigres project adapts Vitess, the sharding system originally built for MySQL at YouTube, to work with Postgres. This could allow Supabase users to horizontally scale read and write workloads without abandoning Postgres compatibility.

Gemma 4 12B brings local AI to laptops

Google's Gemma 4 12B model closes the gap between small and large language models. According to The New Stack, it nearly matches the benchmark performance of the 26B parameter version while consuming a fraction of the memory and compute. That means developers can run inference on a local laptop without cloud dependencies, a shift that has implications for edge computing, privacy, and cost sensitive deployments.

• Gemma 4 12B runs on consumer grade laptops with no GPU required for basic tasks.
• Its benchmark scores approach those of models double its size on common NLP and coding tasks.
• Local execution reduces latency and eliminates data egress costs for AI applications.

Apache Livy graduates after nine years

Apache Livy, the REST interface for Apache Spark, has graduated to a top level project at the Apache Software Foundation. The project spent nine years in incubation but has seen increased behind the scenes activity, per The Stack. Livy allows users to submit Spark jobs from any application using a simple REST API, making it easier to integrate Spark into web services and microservice architectures without a tight Java binding.

GitLab Field CTO Bryan Ross said in a recent interview that “AI is a multiplier” and that organizations with strong guardrails already in place have yielded early wins. He noted that regulated industries such as banking have moved faster than expected on AI adoption because they already had mature data governance and compliance frameworks. That perspective aligns with the broader trend: infrastructure that makes AI practical, whether through smaller models, better database sharding, or simpler job submission, is where the market is focusing.

Coming next: Supabase plans to open source Multigres components later this year. Google is expected to release a larger Gemma 4 model family. The Apache Livy project will likely see a surge in contributions as it transitions to top level governance.