Is Spring Boot still worth it in 2026 — or is it just the Java default you inherit?
Elena Vasquez
September 18, 2026
I have inherited more Spring Boot apps than I have started. That is the honest sample. The question “is it still worth it” is usually asked by someone who already has a pom.xml the size of a novella, or by someone who wants permission to write the next service in Go. I have done both. I still reach for Boot when the problem is a Java shop’s problem. I do not reach for it to prove I am modern.
Worth it means: I can hire, I can find a library for SAML and for batch, I can explain the request path to a junior, and I can still sleep when the JVM is the runtime the company already knows how to heap-dump. Default you inherit means: nobody chose it this year, the actuators are on, and the new service is Boot because the last one was Boot. Both can be true at once. Here is how I tell them apart in 2026.
What Boot still is, after the hype moved to native and to Go
Spring Boot is an opinionated way to start a Spring application: starters, auto-configuration, an embedded Tomcat or Netty, Actuator, a culture of “add a dependency and a property.” Spring Framework underneath is the actual product — DI, transactions, MVC, WebFlux, Security. Boot is the packaging. People conflate them. I still say Boot because that is the artifact I generate and the parent POM I fight.
In 2026 the platform is not stuck in 2015 XML. Virtual threads (Project Loom) made the old “we must go WebFlux for concurrency” argument quieter for a lot of I/O-bound services. Spring Boot 3.2+ and 3.4-era apps on Java 21 are a different machine than the Boot 2 / Java 8 estate I still see in banks. If your “Spring is slow and blocking” slide is from 2018, update the slide before you update the stack.
GraalVM native image is real enough that I have shipped a Boot native binary for a CLI-shaped worker. Startup was the win. The lose was the build, the reachability metadata, and a library that used reflection like a hobby. I would do native again for a scale-to-zero function. I would not do it for a long-lived API that already takes 30 seconds to warm a cache. The default I inherit is a JVM that has been up for weeks. Native is a choice. Treat it like one.

When I would still start a service in Boot tomorrow
The company is a Java shop. Hiring is Java. The security review knows Spring Security. The batch jobs are already Spring Batch. Starting the next bounded context in Micronaut to save 40MB of RSS is a tax I will not pay unless I can name the bill.
The problem is enterprise-shaped: OIDC, an LDAP leftover, a message listener on JMS or Kafka with a transaction that must include the database, a report that is ugly but paid-for. Spring’s gravity is those integrations. I have rewritten one of them in Go. I rewrote the happy path. I did not rewrite the SAML exception. We put Boot back in front of the identity piece and I stopped being clever.
The team already thinks in annotations and application.yaml. Teaching hexagon-and-manual-wiring in a week is not a moral victory if the feature is a webhook. I will introduce narrower modules and fewer god @Service classes. I will not introduce a new runtime to avoid @Transactional.
I like Boot for a modular monolith: several Gradle modules, one process, Spring’s context can still get messy but it is one deploy. That matches how I think about paying for a split on purpose. Boot does not require microservices. The internet just paired them in a talk.
When it is the default I inherited and I should stop adding to it
A 200MB heap to serve a JSON transform that a 20MB Go binary does on a cron. I have seen this. The right move is not “Quarkus will save us.” The right move is to ask whether this process should be Java at all. If the answer is no, I extract the worker. I leave the Boot app that owns the domain.
Auto-configuration I do not understand. Actuator exposed too wide. A parent POM that imports the universe. That is not “Spring is bad.” That is an estate. I treat it like lead paint: stabilize, do not remodel the whole house because a conference said Micronaut is lighter.
WebFlux in a team that cannot reason about operators, used because a blog said reactive is the future. I have un-reactor’d more than I have reactor’d. Virtual threads plus blocking JDBC is the 2026 default I would pick for most CRUD. Reactive stays for the proxy, the gateway, the place we already have a Netty culture.
Quarkus, Micronaut, Helidon, “just use Go”
Quarkus is the one I take seriously as a Boot alternative in the same language. Developer mode is excellent. Native story is a first-class pitch. If I were starting a Kubernetes-native Java service on a team that wants Java and wants fast startup, I would trial Quarkus. I would not migrate a 300-class Boot app to Quarkus for the tweet. Extensions are not a 1:1 of Spring’s long tail. Count the Spring-only libraries you actually use — Cloud Config, Batch, a weird IBM MQ starter — before you fall in love with a supersonic subatomic slide.
Micronaut’s compile-time DI is the right idea. I have enjoyed it in small services. The ecosystem is thinner. I use it when I want Java or Kotlin without Spring’s reflection tax and I do not need the long tail. I do not use it as a personality.
Go is the right answer when the service is a sharp edge: a proxy, a proto worker, a thing whose value is a small binary and a boring deploy. I have said the same about Rails versus Go for product apps. Java versus Go is the same shape. Do not “just use Go” for the module that embeds a rules engine the business already bought in Java.
Kotlin + Boot is the version of this stack I actually like writing. I would start a new Boot service in Kotlin if the team will accept it. I would not force a rewrite of 800 Java files to feel contemporary.

The parts of Boot I insist on, and the parts I disable
I insist on Actuator for health and metrics — and I lock it down. I insist on a structured logging story (Logback plus JSON in prod). I insist on tests that boot a slice of the context, not a mock of every bean (I have used @SpringBootTest too much and @WebMvcTest too little; I now pick the smallest slice that fails for the right reason). I insist on Flyway or Liquibase, not “Hibernate will update the schema.”
I am cautious with Spring Data JPA as the default repository for every noun. I like it for simple aggregates. I write SQL (jOOQ, or Spring JDBC, or a native query I can read) when the page is a report. Hibernate will happily N+1 you. I have been the person who found 400 queries behind one page. That is not a reason to leave Java. It is a reason to read the SQL.
I disable the kitchen sink. I do not add Spring Cloud to a two-service shop. I do not add a config server for three properties. I do not add Eureka when Kubernetes already has a Service. Boot’s gravity will pull in a 2016 Netflix stack if you let a starter’s readme choose your architecture.
So: worth it, or inheritance?
If you are choosing a stack for a Java organization in 2026, Boot is still worth it. It is the default because the default is the hiring market, the library market, and a request path a thousand people can maintain. That is a feature. Boring is a feature.
If you are choosing a stack for a greenfield that is a small API with no Java estate, Boot is often inheritance-from-the-industry rather than a fit. Go, Rails, even Quarkus on a team that wants JVM-but-lighter — those are real choices. “We know Spring” is a valid reason. “The tutorial used start.spring.io” is a weaker one.
If you already have Boot, the 2026 work I would fund is not a rewrite. It is Java 21, virtual threads where they help, a dependency audit, Actuator hygiene, and SQL you can see. Then extract the worker that should not be a JVM. That is how I treat the default I inherited: make it honest, do not set it on fire to feel young.
I will keep writing Boot. I will keep writing the other things when the bill says so. The question is not whether Spring had a good decade. It did. The question is whether the next process you start deserves that decade, or deserves a smaller runtime and a smaller lie about what “enterprise” means. I only call it worth it when I can name which of those I am choosing.