Why Do Enterprise Cloud Costs Increase After Migration, and How Can IT Teams Prevent It?
Most cloud business cases are written in the future tense and reviewed in the past tense. The plan promises lower run costs, the cutover goes reasonably well, and two quarters later someone in finance puts the invoice next to the original model and asks what happened. The first instinct is to blame the provider or the migration partner. The real answer is usually less dramatic and more useful. The organization moved its workloads but kept the habits of a data center: fixed budgets, procurement as gatekeeper, capacity sized for a peak that arrives twice a year.
The numbers say this is widespread, not a handful of careless teams. Flexera's 2026 State of the Cloud Report, based on 753 cloud decision-makers, found estimated wasted spend on IaaS and PaaS climbing to 29 percent after five years of decline, while 17 percent of respondents overshot their public cloud budgets in the past year. This article looks at why that happens, and at what an IT team should change before the next migration wave rather than after it.
Why Enterprise Cloud Costs Rise After Migration
Enterprise cloud costs usually rise after migration because the organization changes where workloads run but not how it plans, buys and owns them. Five things drive most of the increase. There is a double running period nobody budgeted. Servers arrive at their old size. Data now bills every time it moves. Compliance duties turn into storage and logging charges. And consumption has no single owner, which AI workloads now make harder to forecast. Prevention starts before cutover: decide placement workload by workload, budget the overlap, trace the data path, and have cost allocation running from the first day.
Where cloud spending stands in 2026
Start with scale. Gartner put worldwide end user spending on public cloud at $723.4 billion in 2025, up 21.5 percent from $595.7 billion the year before. Its later forecast has growth of 21.3 percent in 2026, with AI integration as the main driver. India is growing faster. Gartner expects Indian end user spending on public cloud to reach $17.5 billion in 2026, a 28.1 percent rise from $13.7 billion in 2025.
A quick note on why market size figures disagree. Gartner counts end user spending, which includes software delivered as a service. Synergy Research Group tracks a narrower slice, the infrastructure providers sell, and put full year 2024 cloud infrastructure services at $330.4 billion. A headline of $330 billion and another of $723 billion can both be correct. They measure different things, so check which one a vendor is quoting before it goes into a board paper.
Waste is the number that matters more to this topic. Five years ago Flexera's respondents estimated 30 percent of cloud spend was wasted and reported public cloud running over budget by an average of 24 percent. So the improvement was real but modest, and it has now reversed. It is a survey estimate, not an audit, but the direction is what deserves attention. Consider the awkward part: 63 percent of organizations now have a dedicated FinOps team and 71 percent run a Cloud Center of Excellence. Process maturity went up and waste went up with it. Flexera links the uptick to the cost complexity of AI and newer IaaS and PaaS services.
The FinOps Foundation's data points the same way. In its 2026 survey, 98 percent of respondents said they manage AI spend, up from 31 percent in 2024. And Gartner is blunt about where this leads it predicts that by 2028 one in four organizations will be significantly dissatisfied with their cloud adoption, citing unrealistic expectations, suboptimal implementation or uncontrolled costs.
My reading of these figures is that the cloud is not failing. Cost outcomes are being decided by operating discipline, and discipline has not kept pace with the number of services a team can switch on.
Six reasons the bill rises after cutover
1. The double running period nobody budgeted
Migrations rarely switch off one estate and switch on another. Waves overlap, data center contracts and licenses expire on their own calendar, and the cloud meter starts on day one. The FinOps Framework's 2026 update makes this explicit: its new executive strategy capability asks leaders to set guardrails with full awareness of overlapping legacy, migration and decommissioning costs. Cloud cost guidance from Cloudaware lists the double run period alongside labor, data movement and integration as the biggest migration cost drivers, and notes that savings tend to appear only after workloads stabilize and temporary overhead is gone. If the business case assumed a clean handover, the bill can look wrong for several quarters even when nothing is broken.
2. Lift and shift moves the inefficiency and adds a meter
On premises servers are sized for peak load plus growth. They cost the same whether they idle or work, so nobody tracked utilization closely. Move them unchanged onto instances billed by the hour and that headroom becomes a recurring charge. This is my analysis rather than a survey finding, but it fits the waste figures above. It also explains why the FinOps Framework now stresses that workload placement decisions should come earlier, before deployment commitments are made, including whether a given technology category suits the workload at all.
3. Data now bills when it moves
Moving data into a cloud is generally free. Moving it out, or between regions, is not. Cloud consultancy Blazeclan notes that hybrid designs also pay for each API call from on premises systems into the cloud, and that many companies carry six figure egress bills they never modeled. Phased migrations are the worst case, because chatty applications get split across environments for months. And the pattern is becoming structural. Gartner forecasts that by 2030 over 60 percent of enterprises will run intensive AI model activity in one cloud while keeping their data in another, up from under 10 percent today. Data movement is turning from an edge case into a design assumption.
4. Compliance quietly becomes storage and ingestion
This one is specific to India and rarely appears in global cost articles. Rule 6 of the DPDP Rules requires data fiduciaries to retain logs and personal data for one year to support detection, investigation and remediation, unless another law requires a different period. CERT-In's directions separately require ICT system logs to be kept for at least 180 days within India. Many cloud logging and monitoring services price by volume ingested and retained (a general pricing pattern, so check your own contracts). An estate that used to write logs to local disks can suddenly produce a telemetry bill nobody planned. One DPDP implementation playbook advises confirming SIEM retention of at least 365 days and budgeting storage as a real line item.
5. The bill has no owner
On premises, procurement was a natural throttle. In the cloud any engineer can create cost in minutes. Flexera reports that 85 percent of organizations still call managing cloud spend a top challenge. The team meant to fix it is usually small: among organizations spending over $100 million, the average FinOps team is only 8 to 10 practitioners plus contractors. Small teams cannot chase every untagged resource, which is why ownership has to be pushed out to the teams that create the spend.
Tooling is getting better here. FOCUS 1.3 adds transparency into shared cost allocation for containers and databases, a separate dataset for contract commitments, and metadata that signals data freshness. Native FOCUS exports are available from more than 11 providers, including AWS, Microsoft Azure, Google Cloud, Oracle and Alibaba Cloud. A common billing language does not create ownership, but it removes an excuse.
6. New services and AI arrive faster than the guardrails
Flexera notes that generative AI adds dynamic, hard to predict workloads and new pricing patterns that complicate forecasting. The FinOps Foundation reports AI cost management as the top skill teams want to add. Looking ahead, Gartner predicts 50 percent of cloud compute resources will go to AI workloads by 2029, up from less than 10 percent. That is a forecast, but it tells you where the forecasting problem is heading.
A case worth studying: 37signals
Public numbers on cloud cost are rare, which makes the 37signals story useful. The company behind Basecamp and HEY ran up cloud charges of $3,201,564 in 2022. Those bills grew even though its workloads were relatively stable and predictable. After moving to its own hardware, its cloud bill fell to about $1.3 million, all of it AWS S3 storage under a contract that had not yet expired, a saving of almost $2 million a year. Its CTO projects total savings from the combined exit at well over $10 million across five years.
Read it carefully before repeating it in a meeting. 37signals has around 60 employees, the workloads are steady, the figures are the company's own, and it carries its own hardware and operations effort. It is not proof that the cloud is a mistake. What transfers is the question: for each workload with steady demand, has anyone recomputed whether running it in the public cloud is still the best economics? Most enterprises have never asked it after the migration ended.
The rules and standards now shaping the cost equation
India: DPDP, SEBI and RBI
The DPDP Rules were notified by MeitY in mid November 2025. The Board provisions applied immediately, consent manager obligations follow at twelve months (around November 2026), and substantive data fiduciary obligations follow at eighteen months (around May 2027). The timeline was reaffirmed recently: on 12 August 2026 the government told the Lok Sabha that core obligations take effect within 18 months of the November 2025 notification. Penalties can reach ₹250 crore. The cost link is direct. You cannot minimize data movement, tier storage sensibly or define retention until you know where personal data actually sits.
For regulated entities the constraints are tighter. SEBI's cloud framework points regulated entities to MeitY empanelled providers, keeps data storage and processing within their data centers, and has the entity retain ownership of data, logs and encryption keys. The RBI's IT outsourcing directions came into effect on 1 October 2023 and require an outsourcing policy with an exit strategy for different scenarios that keeps the business running during and after exit. In practice this means the cheapest region or provider is not always an available option, and an exit plan is a costed deliverable rather than a paragraph in a policy.
Standards: FinOps Framework 2026 and FOCUS
The FinOps Framework 2026, published in March, added Executive Strategy Alignment as a new capability and technology category guidance covering public cloud, SaaS, data center, data cloud platforms and AI. FOCUS 1.3 was ratified on 5 December 2025, and validator support for FOCUS 1.4 is due later in Q3 2026. For an IT head the practical use is simple: ask every provider and major SaaS vendor for FOCUS formatted billing data, and build reporting on it from the start.
A pricing signal from Europe
The EU Data Act does not bind Indian enterprises unless they serve EU customers, but it is shaping vendor behavior globally. From 12 January 2027, providers in the EU may not charge switching fees, including data egress charges. Google dropped exit egress fees in January 2024 and AWS followed in March 2024, yet a UK competition authority review found uptake low to moderate, suggesting fees were never the only barrier and that contract terms and export duties decide whether a customer can leave. Also note that existing contracts do not update automatically, so clauses must be reviewed or the old cost structure stays in force. My suggestion: use these programs as a benchmark in your next renewal conversation, whether or not the law applies to you.
How IT teams can prevent the increase
Think in two phases. Five decisions belong before cutover, and four habits belong after it.
Before cutover
Decide placement workload by workload. Steady, predictable systems and bursty, seasonal ones have different economics. A blanket "everything to cloud" mandate is how the first migration bill goes wrong. Bringing placement into architecture review, as the FinOps Framework now advises, is cheaper than fixing it later.
Budget the overlap. Put the double running period in the business case with an end date and a named owner for decommissioning. Legacy contracts and licenses should have their own exit calendar.
Trace the data path. Draw where each dataset is read and written, and price cross region traffic, egress and hybrid links before waves are scheduled. Keep chatty components together, even if that changes the wave plan.
Cost the compliance layer. Include log retention, encryption key management, data residency and DR in the estimate. If you handle personal data of people in India, use the DPDP retention rule as a design input, not an afterthought.
Read the exit terms. Check egress, export formats and commitment penalties now. For regulated entities the RBI exit strategy requirement makes this mandatory anyway.
After cutover
Allocate from day one. Enforce tags and account structure with policy, not reminders. Unallocated spend cannot be owned, and unowned spend is where waste hides.
Rightsize with engineering in the first quarter. Give teams their own numbers and a target, switch off non production environments outside working hours, and buy commitments only once a stable usage baseline exists.
Measure unit cost, not the total bill. Cost per transaction, per customer or per branch tells you whether spending is healthy. Flexera reports 64 percent of organizations now show value delivered to business units, and that shift from savings to value is the right direction.
Put guardrails ahead of AI. Budgets, quotas and anomaly alerts should exist before a model goes to production. The FinOps community is showing growing interest in pre deployment controls that address cost and policy before spend materializes.
What happens next
These are forecasts, not facts. Gartner expects continued double digit growth in public cloud, as the figures above show. It also predicts more than half of organizations will not get the results they expect from multicloud deployments by 2029, because connecting and managing different environments is hard. Regulatory dates cluster in the next twelve months: the EU switching deadline in January 2027 and the DPDP obligations around May 2027. FinOps itself is widening: 90 percent of respondents manage or plan to manage SaaS spend, up from 65 percent in 2025.
My expectation is that the conversation moves from "how much did the migration save" to "what does each workload cost per unit of business output, wherever it runs." Enterprises that can answer that for cloud, SaaS, AI and their own data centers will negotiate better and surprise their CFO less.
NS3TechSolutions’ Approach to Enterprise Cloud Cost Optimization
Cloud cost problems sit on an awkward border. Finance sees the invoice, engineering sees the architecture, and the network and security teams see the traffic and logs that drive a surprising share of both. NS3TechSolutions works across that border, with IT infrastructure, enterprise networking, network security, cloud, managed services, SOC and NOC operations, and IT consulting in one place.
In practice that shapes the work in three ways. Before migration, NS3 can help an IT team assess workload placement and map the data paths and network links that will carry production traffic, the costs most often missed in a business case. During and after migration, security and monitoring design matters: retention, log routing and SOC coverage need to meet regulatory expectations without paying for the same telemetry twice. Once the estate is live, managed services and a NOC give the review rhythm described above an owner, so rightsizing, commitment reviews and anomaly checks happen on a schedule and not only after a budget scare. For workloads that fit better on premises or in a hybrid setup, infrastructure deployment expertise matters as much as cloud skills. The aim is not to argue for or against public cloud. It is to make sure each workload lands where its economics and compliance obligations make sense.
Practical checklist to save and share
Does every migration wave have a decommissioning date and a named owner?
Is the double running period in the approved budget?
Has each workload been classified as steady or bursty before placement?
Do we have a data flow map showing cross region, egress and hybrid link traffic?
Are log retention periods set to meet DPDP (one year) and CERT-In (180 days in India)?
Do we know where personal data sits, by system and region?
Are tags and account structure enforced by policy?
Is at least one cost metric tied to business output, such as cost per transaction?
Are commitments reviewed against actual usage on a fixed schedule?
Do AI projects have budgets, quotas and alerts before production?
Have we reviewed egress terms, export formats and exit clauses in every major contract?
Can we export billing data in FOCUS format from our main providers?
Frequently asked questions
Q. Why is our cloud bill higher than the on premises cost it replaced?
A. Usually because of overlap costs, servers migrated at their old size, data transfer charges, compliance driven storage and logging, and spend with no clear owner. Fixed on premises costs were paid whether or not they were used, while cloud charges scale with every resource created.
Q. How much cloud spend is wasted?
A. Flexera's 2026 survey estimates 29 percent of IaaS and PaaS spend is wasted. That is a self reported estimate from decision makers, so treat it as a benchmark to compare against your own audit, not a fixed rate.
Q. Is FinOps enough to stop cost increases?
A. Not by itself. 63 percent of organizations already have FinOps teams and waste still rose. FinOps works when it is paired with placement decisions before migration, engineering ownership and guardrails.
Q. Do egress fees matter for Indian enterprises?
A. Yes, for data heavy and multicloud designs. The EU Data Act removes switching charges only for EU customers from 12 January 2027, but Google and AWS already run free egress programs for customers leaving. Negotiate egress and exit terms explicitly.
Q. Does the DPDP Act raise cloud costs?
A. It can, mainly through retention. Rule 6 requires one year retention of logs and personal data for detection and investigation, which increases storage and monitoring volume. Tiered storage and a clear retention policy keep this manageable.
Q. When does moving a workload back from the cloud make sense?
A. When demand is steady and predictable, egress and storage dominate the bill, or residency and exit requirements favor control. The 37signals numbers above are one example, from a small company with its own operations skills, so model your own workloads before deciding.
Q. Which KPIs should a CIO track?
A. Unit cost per business output, share of unallocated spend, estimated waste, commitment coverage, forecast accuracy against budget, and the cost of retained logs.
Q. How should we budget for AI workloads?
A. Treat them as their own scope. Set quotas and alerts before production, forecast in ranges instead of single figures, and review usage weekly at first. Unpredictable usage patterns are among the top concerns Flexera reports for cloud AI initiatives.
The takeaway
A cloud bill is a record of decisions. Some were made on purpose. Most of the increase enterprises see after migration traces back to decisions that were postponed: where a workload belongs, who owns its cost, how long its data must be kept, what it costs to leave. Postponed decisions do not disappear. They compound at the provider's rate. The teams that stay ahead of their budgets in 2027 will not be the ones with the cleverest discounts. They will be the ones who treated cost as a design requirement on the first day of the migration and kept a person, not a dashboard, accountable for it.