Vendor Watch

The case for paying for a Kafka distribution used to rest on two operational facts: someone had to run ZooKeeper, and long retention meant buying broker disk. Both have moved. Apache Kafka 4.0 is KRaft-only, with ZooKeeper mode removed rather than deprecated, and tiered storage under KIP-405 reached general availability in the 3.9 line, so cold segments can sit on object storage without a commercial add-on. None of that makes Confluent Platform redundant. ksqlDB, the commercial connector catalog, Control Center, and RBAC are still real products with no open drop-in, and estates built on them are not cheap to unwind. What has changed is where the value sits: the part of the subscription that was about running the cluster has shrunk, and the part above the broker has not. Inventory which side your estate actually uses before you price the renewal against the alternative.

Since Broadcom completed the VMware acquisition in late 2023, the changes that defined the first renewal cycle, perpetual licenses retired, the product catalog collapsed into a few subscription bundles, per-core pricing with 16-core-per-CPU minimums, have stopped being news and become the steady state. Widely reported renewal increases in the 3x to 10x range remain the anchor number teams bring to us. What has kept changing is the other side of the ledger: the conversion tooling into Proxmox VE, XCP-ng, and Hyper-V has matured to the point where the technical migration is rarely the blocker, estate discovery and the blocker inventory are. If your renewal lands in the next two quarters, run the estate through the RVTools analyzer before the negotiation, not after it.

IBM’s acquisition of HashiCorp closed in early 2025, and the question we get most is whether the BSL-era forks would keep momentum once the deal settled. For secrets management the answer so far is yes: OpenBao, the Linux Foundation fork of Vault, has held a steady release cadence and remains compatible with the core Vault API surface, KV engines, and auth methods most estates actually use. The practical exit cost is therefore not the server swap but the long tail of consumers, every app, pipeline, and operator that authenticates, plus the audit of dynamic-secrets integrations. Teams with Vault renewals approaching should scope that consumer inventory now; the migration itself is increasingly the easy half.

Oracle’s 2023 switch of Java SE licensing to a per-employee metric, priced on total headcount rather than installations, continues to generate widely reported audit and outreach activity. The pattern practitioners describe: a Java licensing conversation becomes the doorway to reviewing the whole Oracle relationship, database included, and teams without a credible exit story negotiate from weakness. You do not need to actually migrate to benefit from being able to: a scoped, costed PostgreSQL exit plan for even a slice of the estate measurably changes the conversation. Our Oracle-to-PostgreSQL guide covers what genuinely ports cleanly and what (PL/SQL depth, RAC dependencies) does not, which is exactly the split a negotiation plan needs.

CentOS 7 reached end of life on June 30, 2024, and the surveys and vendor telemetry that circulate keep finding meaningful fleets still running it, now two years without security updates. The uncomfortable part is that this is among the easiest migrations we cover: migrate2rocky and almalinux-deploy convert in place, most estates need no application changes, and the effort is dominated by testing discipline rather than technical difficulty. If unconverted CentOS 7 boxes survive in your estate, they are usually a symptom of missing ownership rather than missing tooling, worth raising precisely because the fix is cheap. Extended-support vendors exist for the genuinely stuck cases, but for everything else the conversion remains the rare migration that is faster than the meeting about it.

Since the Cloud Software Group takeover, Citrix has consolidated its catalog into platform bundles with widely reported minimum commitments and per-user floors that hit small and mid-size deployments hardest, the accounts the vendor has openly deprioritized. The honest caveat we repeat in every VDI guide: there is no drop-in open-source Citrix. Apache Guacamole and Kasm Workspaces replace the remote-access and workspace-streaming use cases well, but full VDI brokering with profile management is a rebuild, not a swap. That makes the decision framework unusual: the first question is not “which alternative” but “how much of what Citrix does do we still actually use”, and for a large share of estates the answer has shrunk to something a gateway covers.

Cisco closed its Splunk acquisition in March 2024, and the renewal pattern practitioners describe since is consolidation: Splunk spend folded into wider Cisco enterprise agreements, with the leverage dynamics that implies in both directions. The economics that drive exits have not changed, ingest-based pricing punishes exactly the log growth every estate experiences. What decides feasibility is the SPL surface: years of saved searches, dashboards, and detection content are the real migration, which is why our guide treats “move the pipeline to OpenTelemetry and Grafana/Loki” and “translate the content” as separate projects with separate go/no-go decisions. Teams that start by re-pointing new log sources at an open pipeline, while leaving history in place until retention expires, consistently report the smoothest exits.

When Proxmox VE 8.2 shipped a native ESXi import wizard in 2024, it removed the last excuse for treating VMware exits as exotic: point it at an ESXi host, select VMs, and let it stream disks directly, no intermediate export. Releases since have kept sanding down the rough edges practitioners reported around large disks and network remapping. The important nuance our guide insists on: the importer solves the plain-VM majority, which was never the hard part. RDMs, shared-bus clusters, FT-protected workloads, and USB-dongled license servers still need per-VM strategies, which is exactly the blocker inventory the RVTools analyzer produces. Run the inventory, subtract the blockers, and what remains is increasingly a scheduling exercise measured in storage bandwidth.

The BSL relicensing that triggered the OpenTofu fork is now history, HashiCorp sits inside IBM, but the practical takeaway keeps strengthening: OpenTofu reads Terraform state, consumes the same providers, and has shipped its own features (state encryption among them) that the upstream lacks. The migration pattern that works is anticlimactic: repoint the binary, run plan, verify a no-op diff, commit. Provider pinning and the occasional enterprise-only feature (Sentinel policies, some cloud-block integrations) are the honest exceptions, and they announce themselves in the first plan run. For teams still deferring: the exit cost here is about as low as this index ever records, which is itself the reason to do it before any renewal conversation, not after.

XenServer under Cloud Software Group has followed the same trajectory as its Citrix sibling: subscription licensing, per-socket pricing, and widely reported cost increases at renewal. What makes this case unusual is the exit: XCP-ng is a direct fork that shares the Xen hypervisor, storage formats, and management concepts, and Vates’ Xen Orchestra imports XenServer pools with minimal ceremony. It is the closest thing enterprise IT has to a drop-in replacement, which is why it carries the best exit-path score on our Lock-in Index. The migration is typically measured in days per pool, not weeks, and the payback is usually inside a single renewal cycle. For estates still paying XenServer subscriptions, this is the exit we recommend evaluating first purely on effort-to-savings ratio.