~8% of enterprises · declining
The Optimizer
$1,565/day → $786/day · 69 days
Active remediation, FinOps embedded in engineering. Recommendations get triaged, owners get assigned, savings get tracked. No migration. No rebuild. Pure hygiene.
Field Manual · State of Cloud 2026
This field manual walks you through what $2.22M of real, annualized cloud spend across 13 enterprises in India, Southeast Asia, the Middle East and Latin America revealed — as a trail with four waypoints. At each one you'll pick up a skill, tick off actions, and move on. You'll finish with a plan, not a PDF hangover.
Your route · four waypoints
Every finding in the research maps to something you can do. So the trail is organized by capability, not by chapter. Walk it in order — each waypoint builds on the last.
Optimizer, Drifter or Scaler — every cloud bill is on one of three paths. Find yours.
≈ 6 min · 1 self-test 0287% of waste is forgotten, not chosen. Ten fixes recover 80% of it.
≈ 7 min · 10 checkboxes 03Teams detect waste 30× faster than they fix it. Put a name on the list.
≈ 5 min · 4 checkboxes 04Regions, metadata, resilience — make them decisions, not accidents.
≈ 6 min · 3 dig-deepersWaypoint 01 · From chapter 3
The "average enterprise cloud bill" is a meaningless number. In the same quarter of the same cohort, one customer halved spend while another quadrupled it. The story is in the spread.
Objective
Place your cloud bill on one of three trajectories — and decide whether it's the one you actually chose.
Every enterprise in the cohort — 13 firms, 25 production accounts, 32 regions — fell into one of three trajectories. Not three maturity levels. Three directions of travel:
~8% of enterprises · declining
$1,565/day → $786/day · 69 days
Active remediation, FinOps embedded in engineering. Recommendations get triaged, owners get assigned, savings get tracked. No migration. No rebuild. Pure hygiene.
~50% of enterprises · flat-rising
$5,787/wk → $6,361/wk · 8 weeks
Detection in place, action absent. Cost grows quietly, 5–10% a quarter, from accumulated unfixed waste. No new launch. Just drift. The default state of enterprise cloud.
~42% of enterprises · 2–4× growth
$1,429/wk → $5,918/wk · 6 weeks
Growth-stage spend. Every new team gets its own workspace, region, NAT gateway, snapshot policy. The growth is legitimate. The waste embedded in it is not.
Most enterprises are Drifters, and drift compounds. At 5–10% a quarter, a flat workload's bill grows 20–40% a year with zero new product shipped. That's the quiet budget overrun the flashy AI line items get blamed for.
Optimizers shed waste fast. Scalers accumulate it fast. Drifters stay stuck in between — which is why averages lie.
| Archetype | Share | Observed move | Window |
|---|---|---|---|
| Optimizer | ~8% | −50% daily spend | 69 days |
| Drifter | ~50% | +10% weekly run-rate | 8 weeks |
| Scaler | ~42% | 4.1× weekly run-rate | 6 weeks |
All ten of the report's pre-registered hypotheses were confirmed by the telemetry — including "customers detect waste 10× faster than they act." The real ratio turned out to be ~30×. The conventional wisdom about cloud is broadly correct; what's missing is the will to act on it.
Trail stop · self-test
Tap the card that sounds most like your last two quarters
Your read · ~8% of the cohort
Trajectory: declining · rare air
Your read · ~50% of the cohort
Trajectory: flat-rising · 5–10% quarterly drift
Your read · ~42% of the cohort
Trajectory: 2–4× growth · waste compounding inside it
Waypoint 02 · From chapter 4
Vendor narratives push big strategic moves — Reserved Instances, Savings Plans, Spot. Those account for just 6% of flagged items. The real waste is mundane: things nobody remembers creating.
Objective
Locate the 87% of waste that's forgotten, not strategic — and recover ~80% of it with ten fixes.
Across 25,225 live resources, 16.9¢ of every cloud dollar was provably recoverable waste, sitting in plain sight — before any architectural change, renegotiation, or repatriation. And 87.4% of the flagged items were orphans: resources with no owner, no purpose, and a running meter.
Why do orphans win? Three reasons, all human. Provisioning is frictionless while de-provisioning is scary. Accounts are organized by team or environment — almost never by lifecycle, so nobody owns the cleanup. And snapshots compound silently: every dev snapshots before a risky change, and almost nobody deletes afterwards.
| Step | Figure | Detail |
|---|---|---|
| Annualized cohort spend | $2,222,930 | $420,554 measured over a 69-day window, annualized |
| Proven recoverable savings | $375,217 | Only high-confidence recommendations counted |
| Waste-to-spend ratio | 16.9% | $375,217 ÷ $2,222,930 |
Excluded on purpose: architectural rebuilds, repatriation, commitment renegotiation, storage-tier moves without access evidence, and application-layer efficiency. Include those and the true recoverable figure is likely 30–40% of spend. For a $10M/yr cloud bill, 16.9% alone is $1.69M on the table; across a ~$680B global market, north of $100B a year. In rupee terms, a single mid-sized enterprise can be sitting on ₹50 crore — sharper still, because the bill is USD-denominated and the revenue defending it is not.
Trail stop · the main checklist
None of these are strategic decisions to revisit. They're hygiene to install — most take about five minutes each. Your progress saves automatically; print this page and it becomes a blank worksheet.
If your platform team can't recite this list from memory, you have a 16.9% problem.
Waypoint 03 · From chapter 5
Detection has been solved. Execution has not. Cloud teams identify waste roughly 30× faster than they act on it — and the reasons are human, not technical.
Objective
Turn a wall of recommendations into a worked list with a named owner — the only thing Optimizers do differently.
Here's the funnel the telemetry exposed, across the whole cohort:
Why the gap exists — all human, none technical. Owner ambiguity: the account belongs to platform, the workload to the app team, the cost to finance; nobody owns the recommendation. Risk asymmetry: deleting a snapshot has a small chance of breaking something visible and personal; keeping it has a 100% chance of costing money, diffusely. No apply-path: most FinOps platforms show recommendations; almost none execute them safely. The platform stops at the dashboard — the work happens, or doesn't, in the cloud console.
Waypoint 04 · From chapters 3 & 6
Multi-cloud, region spread, metadata sprawl, resilience posture — in the cohort, roughly half of these "strategies" turned out to be accidents nobody went back to revisit.
Objective
Audit where you actually run — providers, regions, metadata — and make each part deliberate.
Start with the provider picture: AWS takes two-thirds of every cloud dollar; Azure punches above its weight on fewer accounts; GCP is the long tail concentrated in product and SaaS. And "multi-cloud strategy" is mostly talk — exactly one enterprise in thirteen ran production across all three hyperscalers. Six of thirteen are single-cloud AWS, four single-cloud Azure, two single-cloud GCP. The rest of the conversation is failover plans nobody tested.
Then look under the compute layer. The visible workload is the tip; the metadata is the iceberg — 22 CloudWatch alarms per EC2 instance, 3 IAM roles per workload, 2.3 snapshots per volume per year. Metadata has no lifecycle, so the bill grows linearly with it even when workload is flat.
| Footprint pattern | Example (anonymized) | Regions | $ / region / mo |
|---|---|---|---|
| Global sprawl, one tenant | Global sporting goods · Azure | 32 | ~$60 |
| Concentrated production | Global CPG · Azure | 3 | ~$5,900 |
| Single-region monolith | Consumer internet · AWS | 2 | ~$12,500 |
| Spread without intent | E-commerce · AWS | 19 | ~$560 |
Two of these reflect deliberate architecture. Two reflect accident. Thirty-two regions at ~$60 each is not a strategy — it's entropy with a billing account.
Consolidators run 3–5 large subscriptions (~$15–20K/month each) — governance is tractable. Fragmenters run 20–35 small ones ($500–1,500/month each) — governance scales linearly with team count and cleanup becomes hopeless. One customer ran 33 distinct subscriptions in a single tenant; another ran 3 for comparable spend. The most visible symptom: Databricks workspace proliferation, each workspace spinning up its own NAT gateway and Premium SSD scratch volumes.
Every AWS enterprise in the cohort had at least one Single-AZ production database — 36+ flagged in one customer alone. The fix is trivial, the cost a small uptick in storage, the benefit the difference between a 5-minute outage and a multi-hour one when an AZ fails. The reason is never "we made a deliberate choice." It's always "we ran the migration script in 2021 and never went back."
One more thing before the debrief — the emerging-markets lens sharpens all of this. Cloud bills are USD-denominated; local revenue usually isn't, so a 10% currency swing is a 10% cost increase with zero new resources. That's a treasury risk, not a tech line item — and it's why EM CFOs increasingly avoid 3-year RI commitments, why sovereign cloud priced in local currency commands a premium, and why enterprises that skipped the legacy-migration era are, on average, more architecturally coherent than their developed-market peers.
Debrief · mission complete
Four waypoints, one field kit: a named trajectory, a ten-item fix list, an action loop with an owner, and a footprint you chose on purpose.
Waypoint 01
Optimizer ~8%, Drifter ~50%, Scaler ~42%. Most enterprises drift 5–10% a quarter without intent — the slope is now a number someone owns.
Waypoint 02
16.9% provably recoverable (likely 30–40% truly); 87% of it is orphans nobody chose. Ten fixes recover ~80% of the total.
Waypoint 03
Detection outruns action 30×; under 1% of signals get formally fixed. The blockers are human — so the fix is an owner, a cadence, and pre-approved blast radii.
Waypoint 04
Genuine multi-cloud is 1 in 13. Metadata is the iceberg, fragmentation is governance debt, and Single-AZ prod databases are theater — all now on your audit list.
The question every CTO should ask
"Which trajectory are we on — and is it the one we chose?"
If you can answer that in one sentence, with data, you got what this manual came to give you. If you can't yet, the trail is right there above you — and the checklist saves your place. One more thing to watch: the GPU shift replays every anti-pattern in this manual with three more zeros. A forgotten CPU weekend costs $37; a forgotten 8×H100 weekend costs $4,704. The teams that install these habits in 2026 will look like Optimizers in 2027. There is no third option.
Your next move
ZopDev builds the apply-path — the part of FinOps the whole industry skipped. Bring your trajectory, your checklist score, and your footprint audit; we'll help you work the list the way the Optimizer did.