Nearly nine in ten organisations now run more than one cloud. Flexera's 2026 State of the Cloud report puts multi-cloud adoption at 89%, and the average sits at around 2.6 providers.

Almost none of them chose this. Multi-cloud is usually the result of an acquisition, a data residency requirement, a team that preferred one provider, or a workload that landed somewhere for reasons nobody has documented. It is an outcome rather than a strategy.

The industry's answer has been the same for a decade. One console. One view. Everything in one place. That market reached roughly sixteen billion dollars in 2025 and is growing at close to 28% a year, which is a lot of money for a promise that keeps not quite landing.

I work across AWS, GCP and Azure most weeks, and I have come round to thinking the promise fails for a reason nobody puts on a product page.

It is a vocabulary problem, not a display problem

Aggregating three consoles gives you three consoles on one screen.

The clouds do not merely look different. They think differently. A compute unit is an instance, a virtual machine, or an instance again but with different lifecycle semantics. Permissions are IAM policies attached to roles, or role assignments over scopes, or IAM bindings on resources, and the three models do not map onto each other cleanly enough for a translation to be lossless.

Tagging is the clearest example because it looks like it should be trivial. A tagging policy defined in AWS does not apply in Azure. Same concept, same intent, entirely separate enforcement, and the moment they drift your cost reporting quietly stops being comparable.

Billing is worse. Each provider has its own pricing model and its own discount mechanism, Reserved Instances against Savings Plans against Committed Use Discounts, and its own invoice taxonomy. Finance ends up reconciling three documents that describe similar things using incompatible words.

Put all of that behind one login and you have not unified anything. You have arranged three vocabularies side by side and still need to know which one you are reading.

The bill for the gap

This is not an aesthetic complaint. Flexera puts average cloud waste at 29% of spend, and notes it climbs with each additional provider added without centralised governance. That figure rose in 2026 for the first time in five years, attributed partly to AI workload sprawl spreading across providers.

Alongside it, roughly 80% of organisations report widening visibility gaps across their cloud operations. Not shrinking as tooling matures. Widening.

The providers themselves have started reaching across. Microsoft shipped a GCP connector in preview in late 2025, joining its existing AWS connector, to project resources from other clouds into Azure's management plane. That is a genuinely useful development and it is also, structurally, one vendor asking to be the place you stand. Which is fine until the answer you need lives in the cloud you did not pick as home.

Unify the question, not the screen

The shift that helped me was giving up on one view and going after one question instead.

What I actually want during an incident or a cost review is not a screen showing everything. It is to ask something in plain terms and have whatever holds the answer produce it. Which resources match this name. What is running in this region. Which of these is oversized. Where did this spend come from.

That is a routing problem, and it does not require the clouds to agree on anything. It requires a layer that knows AWS calls this thing one word, Azure another, and GCP a third, and that translates on the way in and normalises on the way out.

This is the part of Vörr I lean on most. Searching for resources by name, by category, by cloud, or by exposure returns a single normalised set regardless of which provider each one lives in, and I do not have to know in advance where to look. When the question spans clouds, which most interesting questions do, that difference is the whole day.

It also removes a failure I used to hit constantly. Searching the wrong console and finding nothing feels identical to searching the right console and finding nothing. One of those means the resource does not exist and the other means you looked in the wrong place, and at eleven at night they are indistinguishable.

What still does not unify, and probably never will

Being honest about the limits matters more here than in most topics, because the market for this is full of claims.

Identity does not unify. The three IAM models are genuinely different, and a layer that flattened them would be lying to you about your security posture. Recent analysis on multi-cloud identity governance makes the point that these expose privilege through different primitives, and the reasonable goal is consistent visibility rather than a single model. You will still write policy three times.

Provisioning does not unify either, not really. Terraform gets you one language and one workflow, which is worth a great deal, but the provider blocks underneath remain provider-shaped and so does the debugging.

Cost normalises only to a point. You can get everything into one currency and one time window. You cannot make a Savings Plan and a Committed Use Discount mean the same thing, and any tool that presents them as equivalent is producing a number that is tidy and slightly false.

What I would do about it

Pick a primary vocabulary for your team and write everything down in it. If most of your infrastructure is on one provider, use that provider's nouns as the house language and translate at the edges. Teams that speak three dialects spend a surprising amount of an incident establishing which one is being used.

Get tagging right before anything else, and enforce it per provider because you have no choice. Tags are the only join key you get across clouds, and every downstream question about cost, ownership or blast radius runs through them. A tag standard that exists in a document and not in policy is not a standard.

Then stop measuring your tooling by whether it shows you everything. Measure it by whether somebody can get an answer without first knowing which cloud to ask. Those are different properties, and only one of them helps at eleven at night.

Author note

I spent a while believing the single pane of glass was a tooling problem that somebody would eventually solve properly. I no longer think that. The clouds are not going to converge on shared vocabulary, because their differences are the product, and any layer that pretends otherwise is smoothing over something you will need to know about later.

What did change things for me was giving up on seeing everything at once and getting better at asking questions that do not care where the answer lives.

Mohan Gopi, Associate DevOps Engineer at Frigga Cloud Labs, working across AWS, GCP and Azure. If your team has landed somewhere different on this, particularly on cross-cloud tagging, I would like to hear it. LinkedIn.

The Single Pane of Glass Never Arrived