When cloud spend starts eating margin: the operator's playbook
Cloud stopped being an IT line item. It's now a margin question, and most leadership teams can't say with confidence what's actually driving the number up.
Cloud used to be an IT line item. Now it's a margin question, and increasingly a resilience question, which makes it a leadership question whether or not the CFO has caught up to that yet.
The pattern is consistent across the businesses we look at. Cloud bills keep rising. The technology estate keeps getting more complex. And when you ask leadership to draw a straight line from that spend to performance, agility, or commercial return, most can't do it cleanly.
It rarely happens because of one bad decision. It happens in small, individually reasonable steps. A tool gets approved for one team. A platform arrives bundled into a vendor relationship nobody scrutinized closely. A development environment that was supposed to be temporary is still running two years later because shutting it down was never anyone's job. Contracts roll forward because renewing takes less effort than renegotiating. None of these decisions looks bad in isolation. Together, they're how a six-figure cloud bill becomes a seven-figure one without anyone deciding it should.
Two questions surface the problem fast. Is cloud run-rate growing faster than revenue, and if so, can anyone explain why in specific terms rather than "the business is scaling"? And if your primary provider had a bad day tomorrow, do you know exactly which customer-facing systems would go down, or would you find out in real time along with your customers?
Resilience Stopped Being a Theoretical Question
The last two years gave every leadership team a real-world education in what concentrated dependency actually costs. The CrowdStrike outage in 2024 grounded flights and knocked hospital systems offline without a single cloud provider's core infrastructure failing. Azure, Google Cloud, and AWS have each had outages since that took down services well beyond their own platforms, including plenty of businesses that had never given a moment's thought to how much of their operation sat behind a single provider's identity or DNS layer.
None of that means hyperscale cloud is unreliable. It means complex systems fail occasionally, and the more concentrated your dependency, the more that failure costs you when it happens.
The operators handling this well aren't chasing multi-cloud for its own sake, and they're right not to. Running everything twice, on two providers, is expensive, and it doesn't fix the underlying problem if you haven't first worked out which systems actually need that protection. The better approach is selective: identify where a provider failure would genuinely hurt the business, customer-facing systems, core data, identity, and build real redundancy only there. Everywhere else, simplify. That's a far more useful posture than treating resilience as an abstract architecture debate that never quite makes it onto the leadership agenda.
Where the Money Actually Goes
Three things account for most of the margin leakage we see. Workloads running on premium services and maximum redundancy that don't need either, because nobody ever went back and matched infrastructure tier to what the workload actually requires. Tool and vendor overlap, often inherited through acquisitions, where two or three platforms do the same job and nobody has had the argument about which one to kill. And spend that's invisible in the only terms that matter to a business, cost per workload, per transaction, per customer, rather than a single number on a monthly invoice nobody can decompose.
AI infrastructure is quickly becoming a fourth. Compute needs for a customer-facing inference workload and a batch training job are nothing alike, and treating them the same on the procurement side is an expensive habit a lot of businesses picked up in the rush to get AI initiatives moving.
What This Actually Requires
Most businesses don't need a cloud reset. They need someone with the authority to ask why spend is what it is, and the visibility to get a real answer rather than a vendor's summary. That's a governance gap more often than a technology one. Fix who owns the question and how often it gets asked, and the technical fixes tend to follow quickly. Leave it unowned, and the bill keeps climbing regardless of how good your engineering team is.
Why is cloud spend a margin problem and not just an IT cost anymore?
Cloud bills keep climbing while the technology estate gets more complex, and most leadership teams can't draw a clean line from that spend to any commercial return. It rarely comes from one bad call. It builds up through small, individually reasonable decisions, like a tool approved for one team or a temporary environment still running two years later, until a six-figure bill becomes a seven-figure one without anyone deciding it should.
Do businesses need multi-cloud redundancy to protect against outages?
Not across the board. Running everything twice on two providers is expensive and doesn't fix the underlying problem if you haven't worked out which systems actually need that protection. The better approach is selective: identify where a provider failure would genuinely hurt the business, customer-facing systems, core data, identity, and build real redundancy only there, simplifying everywhere else.
What actually drives most of the wasted cloud and IT spend?
Three things account for most of it: workloads running on premium tiers or full redundancy they don't need, tool and vendor overlap left over from acquisitions that nobody has resolved, and spend that's invisible in terms that matter, like cost per workload or per customer, because it only ever shows up as one number on an invoice. AI infrastructure is becoming a fourth, since inference and training workloads get billed the same way despite having very different compute needs.