Cloud PKI vs On-Premises CA: How Security Teams Should Decide in 2026
For most enterprises the question is no longer whether to run a public key infrastructure, but where to run it. Certificate volumes keep climbing, public certificate lifetimes are shrinking, and the people who once maintained an internal certificate authority by hand have usually moved on to other priorities. Schutz IT, which delivers cloud PKI for enterprise security and identity teams, sees the same decision land on the desk of almost every client: keep the on-premises CA, move to a hosted model, or combine the two. This guide sets out how to think about that choice in 2026.
Why the on-premises CA is under pressure
A traditional two-tier Microsoft AD CS deployment worked well when certificates were issued mainly for domain controllers, a handful of web servers and some VPN gateways. Today the same team is asked to cover Kubernetes workloads, service meshes, IoT devices, Wi-Fi authentication, code signing and customer-facing APIs. Each of those use cases adds certificates, templates and renewal paths, and every manual step is a chance for an expired certificate to take a production service offline.
Three pressures stand out. The first is certificate lifetime: browsers and industry bodies are moving public TLS certificates towards much shorter validity periods, and internal policy usually follows. The second is staffing, because the engineers who understand root key ceremonies, CRL distribution points and OCSP responders are hard to hire and harder to replace. The third is audit: a single offline root in a locked cabinet is a fine control until the person holding the smart card leaves the company.
What a hosted PKI actually changes
Moving PKI into a managed service does not remove the need for policy, it removes the plumbing. The provider operates the issuing CAs, the revocation infrastructure and the HSM-backed key storage, while your team keeps ownership of certificate profiles, enrolment rules and who may request what. In practice this shifts effort from keeping servers patched to defining a clean certificate policy, which is the work that actually reduces risk.
A hosted model also tends to shorten onboarding. Instead of building a new issuing CA for each business unit, you define a profile, point the enrolment protocol (ACME, SCEP, EST or an API) at the service and let automation handle issuance and renewal. Teams that previously waited weeks for a certificate template can now provision one in an afternoon, and short-lived certificates stop being a threat because renewal is no longer a human task.
A practical decision framework
Rather than debating cloud against on-premises in the abstract, score each certificate use case against four questions.
First, how many certificates does it need and how fast does that number change? Container and workload identities that churn by the hour are a poor fit for anything manual. Second, what does the compliance regime require? Some regulated environments mandate specific HSM certification levels or in-country key custody, and those requirements are worth confirming with your auditor before you shortlist any vendor. Third, how tightly is the use case bound to your directory? Domain-joined auto-enrolment is a strength of AD CS, and many organisations keep it for exactly that reason. Fourth, what is the recovery story if the CA or its provider is unavailable?
Most mid-sized and large organisations end up with a hybrid answer: a hosted service for workloads, devices and external-facing certificates, and a smaller on-premises footprint for the few cases that genuinely need it. That outcome is perfectly healthy, as long as one team owns the inventory across both.
Where identity and certificates meet
PKI rarely stands alone. The same trust anchors that secure service-to-service traffic also protect the token-signing keys, mutual TLS connections and API gateways behind customer logins. If you are reviewing a sound CIAM strategy this year, include certificate management in the scope, because an expired signing certificate on an identity provider locks out every customer at once, with no gradual degradation to warn you.
The practical step is to map each identity component to the certificate it depends on, record the owner and the expiry, and feed that list into the same discovery and renewal process used for the rest of the estate. Where customer identity runs on a separate platform from workforce identity, make sure the certificate inventory does not stop at the boundary.
Cost, risk and ownership
Cost comparisons between hosted and self-run PKI often go wrong because they count licences and servers but ignore people and incidents. An honest model includes the engineering hours spent on CA patching, CRL publication and template changes, the cost of an expiry-driven outage measured in lost revenue and emergency effort, and the audit time spent proving that key ceremonies were followed. When those lines are included, the gap between the two approaches usually narrows or reverses for organisations with more than a few thousand certificates.
Risk deserves the same discipline. A hosted provider concentrates operational risk in one place, which is why contract terms matter: ask about key custody and export rights, the exit process if you leave, the published availability targets for issuance and revocation, and how incidents are communicated. A self-run CA concentrates a different risk, namely dependence on a small number of internal experts. Neither is automatically safer; the safer option is the one whose failure modes you have written down and rehearsed.
Finally, settle ownership early. Certificate problems tend to fall between network, infrastructure, application and identity teams, and each assumes another is watching expiry dates. Naming a single accountable owner for the certificate inventory, with a clear escalation path to application teams, is often the cheapest improvement in the whole programme and costs nothing beyond a decision.
Migration without the drama
Teams that move successfully tend to follow the same sequence. Begin with discovery, because nobody has a complete certificate inventory on day one and the gaps are where outages hide. Next, define a small number of certificate profiles rather than recreating every legacy template, and retire the ones nobody can justify. Then migrate the highest-churn use cases first, since they gain the most from automation and carry the least dependency on legacy directory features. Finally, run the old and new CAs in parallel long enough to observe a full renewal cycle before decommissioning anything.
Keep a close eye on revocation throughout. Whatever model you choose, test that clients actually check revocation status and that your responders stay reachable during a failover, because revocation that exists on paper but is never enforced provides no protection. For a deeper look at how shrinking validity periods change your renewal planning, see this earlier note on shorter certificate lifetimes and what they mean for security teams.
Next steps
Start by listing every certificate-dependent system you own, assigning each an owner, and noting where renewal is still manual. That single spreadsheet will tell you more about your real risk than any vendor comparison. From there, pick one high-churn use case, move it to automated issuance, and measure how many renewals complete without human involvement. If you would like help scoping that first step, you can reach the Schutz IT team through the Schutz IT contact page.
Comments
Post a Comment