[Java] Spring Boot 3.5.0-M3: The AI & Security Milestone

Java's persistence as the language of the enterprise is often credited to its evolutionary nature, and nowhere is this more evident than in the Spring ecosystem. With the release of Spring Boot 3.5.0-M3 in March 2026, the framework has crossed a major threshold. It is no longer just a platform for microservices; it has transformed into a comprehensive AI Orchestration Engine while simultaneously hardening itself against the looming threat of quantum-scale decryption.

The 3.5 release cycle is defined by two primary objectives: making AI a first-class citizen in the Java developer's toolkit and providing a production-ready path for Post-Quantum Cryptography (PQC). Combined with massive improvements in GraalVM native images and the second generation of Project Loom optimizations, 3.5.0-M3 is the most significant Spring Boot release in recent memory.

Native AI Orchestration with Spring AI

The standout feature of 3.5.0-M3 is the deep integration of Spring AI directly into the core starter templates. This allows developers to treat Large Language Models (LLMs) and Vector Databases with the same "Dependency Injection" philosophy that Spring is famous for. You can now define a ChatClient bean and have it automatically wired with your choice of provider—OpenAI, Anthropic, or a local Ollama instance—with zero boilerplate code.

Beyond simple API calls, Spring Boot 3.5 introduces Auto-Configured Vector Stores. Whether you're using pgvector, Pinecone, or Milvus, Spring Boot now manages the schema generation and connection pooling out of the box. This enables Retrieval-Augmented Generation (RAG) to be implemented in a standard Spring Data repository pattern, making AI-driven features accessible to any Java developer without requiring a PhD in data science.

Post-Quantum Cryptography (PQC) Support

As the "Quantum Day" (Q-Day) approaches, the enterprise is racing to update its cryptographic standards. Spring Boot 3.5.0-M3 introduces native support for NIST-standardized PQC algorithms, including ML-KEM (Kyber) and ML-DSA (Dilithium). By integrating these into the Spring Security layer, Boot ensures that your TLS 1.3 and JWT signatures are resistant to future quantum attacks.

This is implemented via a new Hybrid Cryptography Provider. Recognizing that PQC is still maturing, Spring Boot allows for hybrid key exchanges where both a traditional ECC (Elliptic Curve) and a PQC key are used. If one is found vulnerable, the other still provides protection. This "belt-and-suspenders" approach is the new requirement for financial and governmental Java applications in 2026.

Project Loom: Second-Gen Virtual Thread Optimizations

While Virtual Threads (Project Loom) were introduced in Java 21, Spring Boot 3.5.0-M3 represents their full maturity. The Loom-Native Web Stack has been rewritten to eliminate almost all "pinned" thread issues that plagued early implementations. In high-concurrency scenarios, Spring Boot 3.5 can now handle 10x more concurrent connections on the same hardware compared to the 3.2 release.

This is particularly important for Agentic AI workloads, which often involve long-running, IO-bound requests to external LLMs. Instead of blocking a limited thread pool, each AI request now runs in its own lightweight virtual thread. This enables massive horizontal scale without the complexity of reactive programming (WebFlux), allowing developers to stay with the simpler, imperative Spring MVC model.

GraalVM: 40% Faster Startup, 30% Less Memory

The push toward Cloud-Native Java continues with major enhancements to the GraalVM AOT (Ahead-Of-Time) compiler. Spring Boot 3.5.0-M3 includes a new Profile-Guided Optimization (PGO) engine that is now part of the standard Maven/Gradle build plugins. By training the native image on real-world workloads during the build process, GraalVM can now produce binaries that are indistinguishable from JIT (Just-In-Time) performance.

For serverless environments like AWS Lambda or Google Cloud Run, this means cold-start times have dropped to under 100ms for typical Spring Boot microservices. The memory footprint has also been slashed by 30%, allowing organizations to fit more containers on the same Kubernetes nodes, directly impacting the cloud bill.

Observability: The Micrometer AI-Pulse Integration

In 2026, monitoring is about more than just CPU and RAM. Spring Boot 3.5 introduces AI-Pulse metrics via Micrometer. These metrics track LLM latency, token usage, and "hallucination probability" scores in real-time. By feeding these metrics into Databricks Lakewatch or Prometheus, SRE teams can get an immediate view of the health of their AI agents.

The new Observation API also includes "Trace-to-Prompt" mapping. When a user experiences an error, developers can trace the request all the way back to the specific prompt and model response that caused the failure. This level of transparency is essential for debugging non-deterministic AI systems in production.

Conclusion: Java's AI Renaissance

Spring Boot 3.5.0-M3 is a clear statement of intent: Java is not going anywhere. By embracing the AI revolution and the PQC security mandate, the Spring team has ensured that Java remains the safest, most scalable choice for the next generation of software. Whether you're building a simple REST API or a complex autonomous agent swarm, 3.5 provides the structure, security, and speed needed for the 2026 era.

As we move toward the final GA (General Availability) release later this year, developers are encouraged to start testing the M3 milestone. The bridge to the future of Java has been built, and it’s painted Spring Green.