A certificate authority you run yourself turns a Windows VM into a trust anchor for everything else on the network: VPN, internal HTTPS, RADIUS, code signing and machine certificates. The catch is that a private CA is only as good as its key protection and its template design, and both are usually configured wrong on the first attempt. This walks through standing one up on a Windows Server VM and the decisions that matter beyond the wizard.
Standalone Root, Enterprise Issuing
The standard layout is a two-tier PKI: an offline standalone root that signs exactly one subordinate, and an online enterprise issuing CA joined to the domain. The root’s private key should not live on the server that issues certificates. Generate it, sign the subordinate’s request, then shut the root down or take it off the network entirely — its only future job is signing a new subordinate in five or ten years.
The issuing CA is the machine that is actually online, and it is the one whose key you protect with an HSM or at minimum a documented offline backup. On a single Windows VM, the pragmatic compromise is a standalone root used once and a domain-joined enterprise CA for issuance. Do not skip the tier split to save a VM; a compromised single-tier root means reissuing every certificate on the network.
Key Length, Hash and Validity Choices
Use RSA 3072 or 4096 for the root, and either RSA 2048 or ECDSA P-384 for the issuing CA. SHA-256 is the minimum hash; SHA-1 is refused by current browsers and most modern TLS stacks. Root validity of 15 to 20 years and issuing CA validity of 5 to 10 years is normal, with the constraint that the issuing CA must always expire before its parent.
End-entity certificates should be short-lived — 90 days to one year for internal TLS, which limits the blast radius of a leaked key and forces the automation to actually work. The one exception is machine certificates used for 802.1X or VPN, where re-enrollment during an outage is painful; those often justify a two-year lifetime.
Templates Are Where the Security Lives
Duplicate the built-in “Web Server” template rather than editing it, and set: subject name supplied in the request, SAN required, key usage limited to digital signature and key encipherment, and application policies restricted to server authentication only. Leaving the default “any purpose” or allowing client authentication on a web server template is the most common over-permission mistake in an internal PKI.
Set enrollment permissions explicitly at the template level and remove Domain Computers auto-enroll from templates that do not need it. Auto-enrollment is convenient for machine certificates and dangerous for anything that grants elevated trust. Publish only the templates you intend to be usable — an unpublished template cannot be requested, which is a cheap way to reduce the attack surface.
CRL and AIA Distribution Points
Every certificate the CA issues embeds URLs for the CRL and for the CA certificate itself. If those URLs point to a hostname that clients cannot resolve, revocation checking fails and some clients refuse the connection outright. Use a name that is resolvable from every network segment — typically an internal DNS record on the domain, not the CA’s NetBIOS name.
For a small environment, host the CRL on an internal IIS site over HTTP, publish it on a schedule shorter than the CRL validity, and set the CRL overlap period generously. Delta CRLs are unnecessary at small scale. Test by opening the CDP URL from a client machine that is not the CA, which is the only test that proves the distribution point works.
Publishing the Root into Trust
Distribute the root CA certificate through Group Policy under Computer Configuration → Policies → Windows Settings → Security Settings → Public Key Policies → Trusted Root Certification Authorities. Domain-joined machines get it automatically; non-domain machines — including Linux servers and mobile devices — need it deployed by their own mechanism, and forgetting them is the usual reason a certificate “works everywhere except on the NAS.”
Publish the root and the issuing CA certificate into the NTAuth store as well if you plan to use smart cards or certificate-based logon. Without the NTAuth entry, the certificates chain correctly for TLS but logon is rejected with a “no trust” error that is difficult to diagnose because everything else about the certificate looks fine.
Auditing and Revocation Under Load
Enable auditing on the CA so that issued, denied and revoked requests are all logged, and review the log weekly rather than only when something breaks. A request denied because a template permission is wrong looks identical to a request that was never made if you are not logging denials. Set the CA database to a known location on a volume with room to grow; at a few hundred certificates the Jet database is small, but it never shrinks on its own.
Revocation is the part of a PKI that most often goes untested until it matters. Revoke a test certificate, publish the CRL, and confirm from a client that the connection is actually refused. If the client accepts the revoked certificate, your CRL distribution point is broken and revocation is effectively disabled across the whole estate, for every certificate you have ever issued.
Operating a Private CA Without Surprises
Back up the CA database, the CA certificate and the private key together, and store the key backup offline with the passphrase recorded separately. Test a restore once, on a throwaway VM, before you need it. A CA backup you have never restored is a hope, not a backup. Monitor the CA’s own certificate expiry with an alert at 90 days — an expired issuing CA silently invalidates every certificate it ever signed.
Our Windows VPS comparison lists which plans allow you to keep a VM offline or isolated enough to serve as a root CA host, and which are domain-joined by default. For the wider provisioning picture — disk layout, RAM for the CA database, and backup targets — see our Windows VM resource planning notes.



