We quote a 35 percent average infrastructure cost reduction across our hybrid and cloud-native migration programs, and every time we say it aloud in a steering committee, someone — usually the finance representative, occasionally a rightly skeptical infrastructure manager — asks where the number comes from. Good. They should ask. "Cloud saves money" has been repeated so uncritically for a decade that it has curdled into marketing noise, and plenty of organizations have migrated to Office 365 and Azure only to discover their costs moved sideways into different line items rather than down.
So this post is the anatomy lesson. We are going to itemize where the 35 percent actually comes from, using a composite of the enterprise and government programs behind that average — estates in the hundreds-of-servers range, 15,000-plus mailboxes migrated, SharePoint farms retired, DR contracts renegotiated. And, because the honest version is the only version worth writing, we will spend just as long on what does not shrink, what grows, and the accounting tricks that let vendors claim 50 percent while your CFO sees 5.
One definition up front: when we say hybrid coexistence, we mean the architecture pattern — Exchange hybrid, AD Connect, workloads split deliberately between on-premises and cloud during a multi-year transition — not a permanent fifty-fifty limbo. The savings arrive as the on-premises side empties out. Hybrid is the ramp, not the destination.
Where the money actually comes from
Across the programs we have audited after the fact, the reduction decomposes into five buckets, in descending order of size.
| Bucket | Typical share of the total saving | What actually happens |
|---|---|---|
| Retired server estate | ~30% | Exchange, SharePoint, and their satellite servers leave the floor |
| Storage | ~25% | Tier-1 SAN capacity and its refresh cycle stop growing, then shrink |
| Hardware refresh avoidance | ~20% | The 2019–2020 refresh that never gets purchased |
| DR simplification | ~15% | Replicated infrastructure and DR contracts scoped down |
| Licensing consolidation | ~10% | CAL stacks and third-party tools folded into suite licensing |
The retired estate is the visible part. A 15,000-mailbox Exchange 2010 organization is not two servers; it is DAG members across two sites, CAS arrays, edge transport, load balancers, an archiving platform, a journaling target, and the backup infrastructure gravitationally bound to all of it. One government program retired fourteen Exchange servers and a five-server SharePoint 2010 farm; the power, cooling, hypervisor capacity, OS licenses, and monitoring agents went with them. On a virtualized estate the saving shows up as reclaimed Hyper-V capacity — which is real money only if you actually consolidate hosts or defer their replacement, a point we will return to.
Storage is the sleeper. Mailbox databases and SharePoint content databases are the two hungriest tenants on most tier-1 arrays, and they are exactly what leaves. One enterprise client was staring at a seven-figure SAN expansion driven almost entirely by mailbox growth; the migration didn't reduce their storage bill so much as delete a purchase order. Backup storage compounds this — when the mailboxes go, so do their nightly fulls and their offsite copies.
Refresh avoidance is where timing matters enormously. The 35 percent programs were, without exception, timed against a hardware refresh horizon. Migrate eighteen months before the refresh and the avoided capital is genuine saving. Migrate eighteen months after buying new hardware and you have pre-paid for capacity you are about to abandon — the same project, executed at the wrong point in the cycle, produces half the number.
DR deserves its own paragraph because it is the bucket clients forget. Every on-premises workload with an availability requirement drags a shadow of itself: replicated storage, standby compute, a DR site contract, annual test weekends. Exchange Online's SLA replaces the entire Exchange DR shadow. One client's DR facility contract was renegotiated down by a third once mail and collaboration left the recovery scope — a line item nobody had listed in the business case because it lived in a different budget.
Licensing is the smallest and slipperiest bucket. Exchange and SharePoint server licenses, their CALs, the third-party archiving and anti-spam tools, the SSL inspection appliances scoped to mail flow — consolidated into per-user subscription licensing. Whether this saves money depends entirely on what you were paying before and what suite level you land on, which is why it is the bucket where we have also seen negatives.
What doesn't shrink
Now the other column of the ledger, because a 35 percent average contains programs at 25 and programs at 45, and the difference is mostly these.
People costs don't drop — they redirect. Nobody on the programs behind our average was made redundant by Office 365. The Exchange administrators became tenant administrators, identity engineers, and — in the best outcomes — the people who finally had time for the security backlog. If a business case books headcount reduction, challenge it. If it books avoided future headcount as the estate grows, that one usually survives scrutiny.
Network costs go up. Egress to the internet becomes the production mail path. Firewalls sized for a web-browsing workforce get resized for Outlook, OneDrive sync, and Skype for Business media. Every program funded circuit upgrades from the savings; one funded ExpressRoute. This is the most predictable "cost that moved" and it should be in the business case from day one.
The subscription is forever. On-premises infrastructure was lumpy capital that could be sweated for seven years in a bad budget cycle. Subscription licensing is smooth, permanent OpEx that arrives every month whether or not you had a good quarter. Finance teams understand this trade perfectly well — but the model comparison must run over five-plus years, not three, because year-four-and-five subscription costs are exactly where aggressive vendor models quietly stop counting.
The hybrid tail costs money until you cut it. AD Connect servers, the hybrid Exchange server retained for recipient management, ADFS farms where federation was chosen — that plumbing needs patching, monitoring, and certificates forever. Estates that treat "mostly migrated" as done leak savings indefinitely. The last mile — decommissioning, not migrating — is where several points of the 35 live, and it is the mile most programs never budget.
The accounting honesty test
Three questions expose most inflated savings claims, and we invite clients to aim them at us:
- Savings against what baseline? "Versus the cost of upgrading everything on-premises to current versions" inflates the number; almost nobody was going to do that anyway. We baseline against actual current run cost plus committed forward spend — the refresh that was genuinely budgeted, not the platonic ideal upgrade.
- Are reclaimed resources actually reclaimed? Freed hypervisor capacity saves nothing until hosts are consolidated or a purchase is deferred. Freed rack units save nothing in a datacenter you own outright and can't sublet. We only count buckets with a cash or committed-cost consequence.
- Over what period, and who owns the out-years? Our 35 percent is a five-year run-rate comparison, all buckets netted, including the costs that grew. It is an average across programs that were timed and executed to produce it — timed against refresh cycles, decommissioned all the way, network costs included. The same migration done with worse timing and a permanent hybrid tail lands closer to 15. The number is earned, not automatic.
What this means for a 2018 planning cycle
If you are building next year's budget right now, the practical sequence is: inventory the forward spend first (refresh horizons, DR contract renewals, SAN expansion triggers), because that calendar tells you when a migration produces maximum avoidance. Then model five years of both worlds honestly, including the network upgrades and the subscription out-years. Then — only then — argue about migration paths and tools. The programs that hit 35-plus were the ones where finance and infrastructure built the model together before any vendor was in the room.
If you're facing this
We have built and then audited enough of these business cases to know where the bodies are buried on both sides of the ledger. If you have a cloud migration business case that needs a skeptical second pass — or an estate where you suspect the savings are real but can't prove it to finance — get in touch. The assessment that produces an honest number usually costs a rounding error of the decision it informs.