State of Databases 2026

Databricks spent $1 billion on Neon. Snowflake grabbed Crunchy Data for a reported $250 million. Supabase went from a $2 billion to $5 billion valuation in four months. 2025 had one thesis: PostgreSQL wins.

The numbers back it up. PostgreSQL hit 55.6% usage in Stack Overflow's 2025 Developer Survey, making it the most used and most admired database for the second year running. It now powers Databricks' new Lakebase and Amazon Aurora DSQL; even Turso's $4.99/month edge databases run on Postgres. AI agents created over 80% of Neon's databases, not people.

PostgreSQL 18 shipped vector search through pgvector. Oracle rebranded its database "26ai" and bundled it free. SQL Server 2025 added DiskANN indexes. When three of the biggest database platforms ship the same feature in one year, the standalone vector database pitch gets a lot harder.

Redis learned that lesson the hard way. Redis 8.0 offered AGPLv3 as a license option, but the Valkey fork had already locked in AWS, Google, and Oracle, with benchmarks showing over 3x higher throughput. The damage from the 2024 license change was stark: the project went from 12 outside contributors handling 54% of commits to zero making more than five.

Security was brutal. Silk Typhoon breached the U.S. Treasury through a PostgreSQL injection chained with a BeyondTrust flaw. "MongoBleed" exposed credentials across 87,000 MongoDB instances. A researcher found 184 million passwords in an unprotected Elasticsearch database. The shakeout was real too. Fauna shut down in May, and the market is firmly rewarding consolidation. Cloud database spending hit $22.43 billion and is on pace for $62 billion by 2032.

Actions for 2026: Upgrade to PostgreSQL 18 for async I/O and UUIDv7. Try Valkey 8.1 as a Redis replacement. Patch MongoDB against MongoBleed if you're running affected versions.

Database 2025 Timeline

Database 2026 Watchlist

1. PostgreSQL 19

When: September 2026
Context: Development continues with rumored 64-bit Transaction ID (XID) to solve decades-old "transaction wraparound" issue—a critical maintenance burden for massive Postgres clusters. May also include built-in pgvector indexing merged into core. Adoption of Rust via pgrx will grow for high-performance extensions. PostgreSQL MD5 authentication, deprecated in version 18, faces removal as SCRAM-SHA-256 becomes mandatory.
Action: Track PostgreSQL 19 development. Plan upgrade from PostgreSQL 17 or earlier given PostgreSQL 13 is now EOL.

2. Valkey 9.0 and Redis Competition

When: Q1-Q2 2026
Context: Valkey 8.1 achieves 28% lower memory usage than Redis while cloud providers offer 33% lower pricing. Redis Software 7.2 (based on BSD-licensed Redis OSS 7.2) reaches EOL Feb 28, 2026, forcing migration decisions. Valkey patched security CVEs in 9.0 release candidates.
Action: Plan Valkey migration before Redis 7.2 EOL. Benchmark Valkey 8.1 against Redis 8.2 for your workload. Consider cloud provider pricing differences.

3. Snowflake Postgres GA

When: H1 2026
Context: Following Crunchy Data acquisition (estimated ~$250M), Snowflake Postgres is in private preview. Snowflake also acquired Datometry for SQL dialect conversion tools. Snowflake open-sourced pg_lake allowing PostgreSQL databases to query Iceberg tables in S3 directly, effectively turning every Postgres instance into a potential Lakehouse node.
Action: Sign up for Snowflake Postgres preview if using Snowflake for analytics. Test pg_lake for bridging operational Postgres with lakehouse analytics.

4. MySQL 8.4 LTS Transition

When: Throughout 2026
Context: MySQL 8.0 reaches EOL in April 2026 (ending with 8.0.46 per Oracle's release notes). MySQL 8.4 LTS expected to replace 8.0 as the Long-Term Support track. Oracle laid off dozens of MySQL engineers—up to 70 senior MySQL developers in September 2025, raising community concerns about Oracle's commitment.
Action: Plan MySQL 8.0 to 8.4 LTS migration. Consider Percona Server for MySQL or MariaDB as alternatives given Oracle's uncertain commitment.

5. Vector Database Consolidation

When: Throughout 2026
Context: With PostgreSQL 18, Oracle 26ai, and SQL Server 2025 all offering native vector indexing, standalone vector databases face uncertain futures. PostgresML, Hydra, and Voltron Data faced struggles or shutdowns—their "AI-on-Postgres" value propositions overtaken by the core Postgres project's own rapid evolution. Expect at least one major acquisition (e.g., cloud provider buying vector DB startup). Purpose-built vector databases will narrow to specialized high-performance use cases as integrated solutions mature. M&A speculation includes Cloudera and Couchbase joining forces, and SAP HANA may spin off or get more open.
Action: Test whether pgvector or integrated database vector capabilities can replace standalone vector databases. For billion-scale embeddings, benchmark Pinecone, Milvus, and Qdrant. Watch for acquisition announcements.

6. Apache Iceberg Ubiquity

When: Now
Context: Apache Iceberg won the table format war. Microsoft Fabric, Oracle 26ai, Snowflake, and Databricks all offer native Iceberg read/write. DuckDB 1.4.2 gained full Iceberg write support (INSERT, UPDATE, DELETE). This forces vendors to compete on compute engine performance rather than storage format lock-in. Snowflake open-sourced pg_lake allowing PostgreSQL databases to query Iceberg tables directly.
Action: Standardize on Iceberg for new data lake workloads. Plan migration from Delta Lake or Hudi. Use pg_lake or similar tools to bridge OLTP and analytics.

7. Post-Quantum Cryptography in Databases

When: Throughout 2026
Context: Microsoft made PQC algorithms (like ML-KEM) generally available across Windows and Azure platform APIs, with hybrid PQC key exchange coming to the Windows TLS stack—databases running on these platforms can inherit PQC-capable TLS connections. Oracle 26ai features quantum-resistant encryption. "Store Now, Decrypt Later" threats drive early adoption, and first operational mandates for Post-Quantum Cryptography appeared in 2025.
Action: Review PQC-ready database options for sensitive data. Check TLS configurations for quantum resistance. Databases on Windows/Azure can leverage platform-level PQC APIs for encrypted connections.

8. Aurora DSQL and Active-Active Distributed SQL

When: Throughout 2026
Context: Aurora DSQL's active-active architecture with multi-Region strong consistency may set a new expectation for OLTP databases. DynamoDB's multi-Region strong consistency with witness topology enables zero-RPO without full replica costs.
Action: Test Aurora DSQL for new globally distributed workloads. Consider DynamoDB MRSC witness topology for cost-effective global durability.

9. Open Table Formats and Streaming-Lakehouse

When: Throughout 2026
Context: Consolidation shifts from "which format" to "which catalog + governance + engine interoperability." Iceberg 1.10, Delta Lake 4.0, Hudi 1.0, and Paimon 1.0 all reached maturity in 2025. Apache Polaris positions itself as an Iceberg REST-based catalog layer. Flink 2.0 pushes streaming-lakehouse patterns into a unified stack.
Action: Standardize catalog layer (Polaris, Unity, Gravitino). Test Flink 2.x + Paimon for operational-to-analytical pipelines.

10. S3 Tables and Storage-First Database Patterns

When: Throughout 2026
Context: S3 Tables added cross-Region Iceberg replication and Intelligent-Tiering. S3 Vectors reached GA scaling to 2 billion vectors per index with ~100ms latencies. Together they normalize a "storage-first database" pattern shifting workload gravity from specialized services into object storage.
Action: Test S3 Vectors for cost-sensitive vector workloads not requiring sub-10ms latency. Consider S3 Tables replication for cross-Region lakehouse consistency.

11. AI Agent Database Workloads

When: Now
Context: Over 80% of Neon databases created by AI agents. Neon's Agent Plan offers custom resource limits for AI platforms. Gartner predicts AI agents will intermediate most B2B work by 2026—"Self-Driving" database capabilities moving from marketing hype to reality. Sub-second provisioning and scale-to-zero become essential. 74% of IT leaders expect budget increases in 2026 with AI readiness becoming core architectural mandate.
Action: Test Neon, Turso, or Supabase for AI agent workloads needing instant database provisioning. Add rate limiting and cost controls for agent-initiated database creation.

12. DuckDB and Embedded Analytics

When: Now
Context: DuckDB 1.4.0 "Andium" introduced first LTS release with AES-256-GCM encryption. With almost 25 million monthly downloads on PyPI. DuckLake launched as new lakehouse format. Turso's breakthrough with concurrent writes overcame SQLite's single-writer limitation.
Action: Try DuckDB for local analytics and data transformation pipelines. Consider Turso for edge and serverless applications requiring SQLite compatibility with write concurrency.

13. OpenSearch 3.x and Model Context Protocol

When: Now
Context: OpenSearch 3.0 delivered 9.5x faster performance with GPU-accelerated vector search and Model Context Protocol (MCP) support allowing AI agents to query search indexes. AWS announced GPU acceleration for vector operations at re:Invent 2025.
Action: Upgrade from OpenSearch 1.x (deprecated May 2025). Test MCP integration for AI agent applications. Consider GPU-accelerated indexing for large vector workloads.

14. Vector Search Quality and Correctness

When: Ongoing
Context: Using vector search for mission-critical tasks (medical, legal document retrieval) raises quality concerns—issues of drift, bias, and false positives in nearest-neighbor search need addressing. Research emerged in 2025 on "safe vector search": algorithms to quantify uncertainty of vector matches and hybrid approaches that cross-verify with symbolic searches.
Action: Use hybrid search (combining vectors with keyword/BM25) as default pattern. Add uncertainty quantification for high-stakes vector queries. Cross-verify vector results with traditional search for critical applications.

15. Cloudflare D1 and Edge SQL

When: 2026
Context: Cloudflare D1 (essentially a hosted SQLite with Durable Objects) should reach GA—enabling SQL at the edge. **Turso** disrupted the market with $4.99/month for unlimited databases (500 monthly active limit), making database-per-tenant architectures 100-1000x cheaper. Turso's serverless JavaScript driver uses pure HTTP communication via fetch(), working on Cloudflare Workers, Vercel Edge Functions, and Deno Deploy. Edge databases are becoming essential for low-latency AI applications.
Action: Try D1 and Turso for edge-first applications. Consider database-per-tenant architectures where latency matters.

16. Oracle-to-PostgreSQL Migration Wave

When: Now
Context: AWS Database Migration Service has helped customers migrate 1.5M+ databases. Drivers include zero licensing fees and estimated 30% lower TCO. Key tools: Ora2Pg (open source), AWS DMS (serverless), EDB Migration Toolkit (enterprise), pgLoader (up to 3TB/hour throughput).
Action: Plan Oracle-to-PostgreSQL migration for cost savings. Target PostgreSQL 18's async I/O performance improvements that close the gap with proprietary databases.

17. Skills Evolution for DBAs

When: Throughout 2026
Context: DBAs in 2026 need SQL tuning skills alongside vector index management and ML model deployment. Expect new certifications like "Azure Vector Data Engineer" or "Certified AI Database Specialist." Tools like PostGraphile and Prisma are improving GraphQL-to-SQL integrations—2026 might see a major DB provide a built-in GraphQL endpoint. Cockroach Labs reportedly postponed IPO plans, instead taking strategic investment from cloud partners.
Action: Upskill on vector search fundamentals, embedding models, and RAG architectures. Test GraphQL integrations for developer experience improvements.