See it as a diagram
Everything below, as a diagram you can edit. Describe yours and see it in seconds.
No account needed · Editable canvas, not a picture
Draw the seams, that is the whole point
Connectivity: Direct Connect, ExpressRoute, and Cloud Interconnect for private links, or site-to-site VPN between a VPC, a VNet, and a VPC. Label each crossing edge with the mechanism and the direction. Peering between two clouds is not symmetric in practice, and the asymmetry is usually the interesting part.
Identity: this is the seam that gets omitted most often and asked about first. Entra ID as the identity provider federating into an AWS IAM role over SAML or OIDC, or workload identity federation into Google Cloud, is an architectural dependency as real as a network link. Draw it, and label the trust direction.
Data: for every replication or export edge, put the direction, the cadence, and the mechanism on the label. Nightly export to Cloud Storage is a different system from change data capture in near real time, and both get drawn as a plain arrow if nobody writes on it. DNS deserves one line too: which zone is authoritative, and what resolves across the boundary.
Say why each workload sits where it does
Multi-cloud is almost never a preference. It is data residency, an acquisition, a managed service only one vendor has, a negotiated commitment, or an exit plan. A diagram that shows the topology without the reason invites the same question at every review.
One line of annotation per provider zone fixes it: analytics on Google Cloud because of BigQuery, identity on Azure because the organisation is Microsoft 365, the customer application on AWS because that is where it was built. Annotations on zones are cheap, and they convert a topology picture into an architecture decision record people can argue with.
The same discipline applies to what is deliberately not duplicated. If observability is centralised in one provider, say so on the zone, because a reader will otherwise assume a second stack exists.
One canvas, three icon libraries
Datadef carries 2,098 icon files in one index: 439 AWS, 624 Azure, 226 Google Cloud, all 41 labelled Kubernetes resource marks, 373 tech-stack logos and 32 data tool marks. A multi-cloud description resolves every service name in one pass, so building the picture does not mean three downloads and a manual paste job. Browse the whole library at the icon index.
One caution specific to multi-cloud, and it is measurable. When three vendor sets share one index, an abbreviation stops being unambiguous. Querying that index for a bare alb returns two Azure load balancer marks, because Azure registered alb as an alias and AWS did not; a bare iam returns Azure IP Address Manager before anything from AWS. Neither answer is a bug in the matcher, it is what happens when an acronym belongs to two vendors. On a multi-cloud canvas, write the full product name, and check the one node where two clouds offer the same three letters.
Provider zones, labelled crossing edges, and uniform node sizing come out of the generation step, and the canvas stays editable for the annotations only you can write. Export to PNG or JPEG for the board deck, or keep a live embed in the platform wiki.
Terms differ per vendor, and a diagram carrying all three sets has to satisfy all three. The common denominators are the ones each vendor states plainly: use the marks as published rather than modified, do not use them to represent your own product, and do not imply endorsement. Microsoft adds the most specific version, asking that icons are not cropped, flipped, rotated or reshaped. The details are in Azure and Google Cloud icon terms and using AWS icons in presentations.
FAQ
How do I keep a multi-cloud diagram readable?
What should the arrows between clouds say?
Can I mix AWS, Azure, and Google Cloud icons in one diagram?
Should each cloud get its own diagram instead?
Which icon do I use for a cross-cloud VPN?