IBM completed its acquisition of DataStax on 1 November 2025, and reviewing the Academy now means reviewing something in transition. The site carries IBM branding, the courses are still free, and there is a notice explaining that Apache Cassandra certifications are temporarily unavailable while the exam migrates to IBM's certification platform. No date is given. I want to be fair about this, because acquisitions genuinely do disrupt credentialing infrastructure and IBM has a large and functional certification apparatus that Cassandra exams will presumably land in eventually.
But temporarily with no date attached has a way of becoming a long time, and anyone who needs a Cassandra credential for a job application right now is stuck. That is a real cost to learners and it belongs at the top of an honest review. The training itself is the reason to be here, and the data modelling content is the best part. Cassandra punishes relational instincts more thoroughly than almost any other database.
You do not normalise. You do not design a schema and then work out your queries. You start from the queries you need to serve, you design a table per query, you denormalise aggressively, and you accept writing the same data several times because writes are cheap and distributed joins do not exist. Every experienced relational developer's first Cassandra schema is wrong in the same way, and the Academy's courses attack this directly, walking through partition key selection, the consequences of a partition that grows unbounded, and why a query pattern you did not plan for may require a new table rather than a new index.
This is genuinely hard conceptual material taught well, and it transfers to other wide-column and distributed stores. The distributed systems content is also honest. Tunable consistency, what quorum actually guarantees, the practical meaning of eventual consistency for application design, how repair works and why you cannot ignore it, and what a multi-datacentre deployment costs you in latency. Cassandra's architecture is an interesting teaching vehicle for distributed systems tradeoffs regardless of whether you deploy it, and the courses use it that way.
The newer material connects Cassandra and Astra DB to AI work, covering vector search and Langflow, the low-code framework for building AI applications that was one of IBM's stated reasons for the acquisition. This content is current and reasonably practical, though it has the slightly breathless quality that AI-adjacent vendor material usually has. My substantive criticism of the teaching is that it understates how hard Cassandra is to operate. The training will get you to a working cluster and a sensible data model.
It does not adequately prepare you for compaction strategies going wrong under an unexpected workload, for the operational reality of repairs on a large cluster, for JVM tuning, or for the fact that Cassandra rewards organisations with dedicated database engineers and punishes those without them. There is a reason DataStax built a managed service, and the training's enthusiasm for self-managed deployment sits slightly awkwardly against that. The course catalogue is also muddled. Numbered courses from different eras, DS and AC prefixes, and older material still in circulation make it harder than it should be to work out the right sequence.
Three point three. Excellent core teaching, real free access, and a credential currently in a holding pattern nobody will put a date on.