← All research articles
Research Article · Lean AMS & Opex

The Silent Factory in Every Factory: What Enterprise IT Really Costs, and Where the Savings Are

Sizing IT spend, the opportunity inside it, how to optimize IT OPEX, and why a greener footprint follows

20 Nov 2026 · 12 min read · terramavens
Abstract

Every factory, bank, hospital and retailer runs a second, quieter operation: the IT estate that keeps the visible one going. This terramavens study brings together published industry benchmarks to size what organizations spend on IT, show where the budget goes, and set out how much of it is recoverable. IT spend is a low single-digit share of revenue in most sectors and far higher in financial services. Industry research suggests the average company could optimize and reinvest about 30% of its IT spend, yet only about one organization in nine sustains its cost savings. We set out a five-step approach to optimizing IT OPEX and show why the same levers, such as retiring idle assets, right-sizing cloud and extending device life, also reduce the IT carbon footprint. Where the evidence is thin or comes from vendors, we say so.

Every factory has a second factory inside it. The visible one makes products or delivers services. The quieter one, made up of applications, servers, cloud accounts, networks, laptops and the people and vendors who run them, makes the first one possible. It rarely appears on a plant tour, and its costs are scattered across departmental budgets, vendor contracts and corporate cards. We call it the silent factory.

This study asks four practical questions. How much does the silent factory cost? Where does the money go? How much of it can be recovered? And how should an organization go about recovering it? A fifth point follows from the answers: the same actions that cut IT operating cost also cut the energy and emissions behind it, so a greener footprint arrives as a secondary benefit and not as a separate programme.

Our approach is to work only from published benchmarks and to be plain about how far each figure can be trusted. Where a number comes from a vendor survey or from respondents' own estimates, we say so.

Key takeaways

  • IT spend is a low single-digit share of revenue for most sectors (a median of 2.9% in a 2025 North American survey), and several times higher in banking and financial services.
  • About a third of enterprise IT budgets goes to people (30.8%), and spend is spread across so many owners that IT itself controls only a small share of some categories, such as SaaS.
  • Industry research suggests an average company can optimize and potentially reinvest about 30% of IT spend. Yet only 11% of organizations sustain their cost-optimization savings, so holding the gains matters as much as finding them.
  • The evidence for recoverable waste is strongest in cloud, software licences, support labour and ageing infrastructure. These areas overlap, so the opportunities should not be added together.
  • Retiring idle assets, right-sizing cloud and extending device life cut cost and carbon at the same time. The carbon benefit is a by-product of cost work and should be measured, not assumed.

1. How much is spent

The most widely used yardstick is IT spend as a percentage of revenue. The best-known benchmark is an annual survey of more than 750 IT executives in the US and Canada. Its 2025 edition reports an overall median of 2.9% of revenue. The range by sector is wide:

SectorIT spend as % of revenueNote
All sectorsMedian 2.9%2025 survey, US and Canada
Financial services4.4% to 11.4%25th to 75th percentile, published in 2021
Discrete manufacturing1.4% to 3.2%25th to 75th percentile, published in 2021
BankingDescribed as among the most IT-intensive sectors2025 survey, no figure published openly
Energy and utilities, process manufacturing, wholesale distributionDescribed as low to moderate, or comparatively low2025 survey, no figure published openly

The sector percentages for financial services and discrete manufacturing come from an earlier edition of the same benchmark. The full industry-by-industry tables are paid research, so we show only what is openly published. The same research finds that sector explains far more of the variation than company size does.

To make this concrete, here is simple arithmetic on those published ranges, offered as an illustration and not a forecast. A company with USD 1 billion in revenue and the 2.9% median spends about USD 29 million a year on IT. In financial services the 4.4% to 11.4% range implies USD 44 million to USD 114 million. In discrete manufacturing, 1.4% to 3.2% implies USD 14 million to USD 32 million.

At the global level, analyst forecasts put worldwide IT spending at about USD 5.44 trillion in 2025. The shares below are our own calculation from the published segment figures:

Segment2025 spend (USD trillion)Share of total
IT services1.6931.0%
Communications services1.2823.6%
Software1.2322.7%
Devices0.7614.0%
Data centre systems0.478.7%
Total5.44100%

These are market-level figures. They include consumer and public-sector spending and measure what vendors earn, not what a single company budgets, so they show the shape of the spend and not a benchmark for your own.

2. Where the money goes inside the enterprise

The same 2025 enterprise survey gives a view of how IT budgets are divided, and how that has changed over a decade:

Budget itemShare of IT budgetTrend
Personnel30.8%Down from 40.9% ten years ago
Security5.4%Up from 1.5% ten years ago
AI1.9%Tracked separately for the first time
In-house data centres9.0%Down from 10.2%
Cloud plus in-house data centres14.6%Combined
Capital spending13.0%Down from 21% in 2016

The openly published summary does not break out hardware, software, outsourcing or cloud on their own, so we do not show them. The direction is still clear: less money is tied up in owned equipment, more is paid as a running cost, and the budget is shifting towards security and AI.

The second feature of the silent factory is how little of it IT controls. A 2026 SaaS-management dataset reports that IT controls just 15% of SaaS spend and 13% of application ownership, with other departments responsible for almost three-quarters of ownership. That is one vendor's dataset, but it matches what many finance leaders see: costs sit in business-unit budgets, on corporate cards and inside vendor contracts, which is why they are hard to see as a whole.

3. How big is the opportunity?

The best-known estimate comes from consulting research, which finds that an average company has an opportunity to optimize and potentially reinvest 30% of its IT spend through improvements to IT productivity. Applied to the USD 29 million example above, that would be roughly USD 8.7 million a year. It is an average across companies, not a promise for any one of them.

The more useful question is where the waste sits. The table collects the evidence we could trace, area by area.

AreaWhat the evidence saysHow to read it
CloudIn annual cloud surveys, respondents estimated that 27% of cloud spend was wasted in 2025 and 29% in 2026, the first rise in five years, as AI workloads grewSelf-reported estimates from a survey, not audited waste
Software and subscriptionsA SaaS-management vendor finds organizations leave 46% of licences unused on average, about USD 19.8 million of wasted SaaS spend a yearVendor data, from large SaaS portfolios, so smaller firms will see less in absolute terms
Support labour and managed servicesAt two financial-services organizations analyzed in consulting research, 90% of the IT organization was on support and administrative work such as testing, deployment and maintenance. Application maintenance and infrastructure often make up 40% to 60% of the IT organization. For banks, the same research reports 5% to 10% savings within six to 12 months from reducing demand, and the same from renegotiating and consolidating vendorsThe 90% figure is two organizations. The savings ranges are for banks
Service desk and ticket volumeA 2024 benchmark of more than 9,000 organizations found 53% ticket deflection when generative-AI self-service is used. One large platform vendor reports that for its own IT service desk, 15% of incidents are currently eligible for its Level 1 AI specialist, and 46% of those are resolved by itVendor data. The two figures show how far outcomes differ, depending on how many tickets are eligible
Technical debtCIOs in a survey estimated tech debt at 20% to 40% of the value of their technology estate, and most put less than a fifth of their tech budget into paying it downA 2020 survey of 50 CIOs at financial-services and technology companies
Data centres and infrastructureCooling takes about 7% of electricity in efficient hyperscale data centres and over 30% in less-efficient enterprise ones. A 2015 study found 30% of nearly 4,000 sampled servers had done no useful work for six monthsThe server study is old and from vendor customers. Treat it as an indication to test, not a current benchmark

Two cautions apply. First, these areas overlap. A retired application removes a licence, a server and the tickets that supported it, so the opportunities cannot be added into one total. Second, finding savings is the easier half. Industry research finds that only 11% of organizations succeed in sustaining their cost-optimization savings. Savings that are not governed tend to creep back.

4. How to optimize IT OPEX

The evidence points to a sequence, and its order matters. Each step can start small.

  1. See the whole estate. Build one baseline from the data you already have: service tickets, licence and usage records, cloud bills, contracts, asset lists and the headcount behind each service. The aim is to give each cost a name, an owner and a unit, such as cost per ticket, per user or per workload. Without this step the later ones are guesses.
  2. Size the opportunity by lever. Use the table above as a checklist and estimate each lever from your own data, not from industry averages. Rank by value, effort and risk, and note where levers overlap so the same saving is not counted twice.
  3. Take the quick wins. Reclaim unused licences, shut down idle environments and servers, right-size over-provisioned cloud resources, and consolidate and renegotiate vendor contracts. Published bank studies report payback within six to 12 months for the demand and vendor levers, and break-even within two years for the programme as a whole.
  4. Remove the recurring work. Look at the tickets, alerts and requests that keep returning. Some can be removed at source by fixing the monitoring or configuration that causes them. Others are routine requests that can be automated end to end. A smaller set, the more complex diagnostic work, may suit AI agents that gather evidence and propose a fix for an engineer to approve. Realistic expectations matter: as the vendor figures above show, the share of work that is eligible can be much smaller than headline deflection rates suggest.
  5. Hold the gains. Make the savings stick with named owners, a regular review of cost and usage against the baseline, and rules that stop new waste at the point of purchase. This is the step the 11% figure says most organizations miss.

The same approach applies whether the work is done in-house or with a partner. For how we apply it to managed services and run cost, see our IT Opex Optimization page and the terra-diagnose diagnostic.

5. The greener footprint that follows

Enterprise technology is a modest but real part of the climate picture. ICT as a whole is estimated at roughly 1.5% to 4% of global greenhouse gas emissions. Consulting research puts enterprise technology at 350 to 400 megatonnes of CO2e, about 1% of the global total, and notes that it is a large share of Scope 2 emissions in some sectors: 45% in insurance and 36% in banking. Energy-agency estimates put data centres at about 415 TWh, around 1.5% of global electricity, in 2024, with a base case of about 945 TWh, just under 3%, by 2030.

The link to cost is direct. Anything that removes a resource that was not needed also removes the energy and embodied carbon behind it.

LeverCost effectCarbon effect
Retire idle servers and unused applicationsStops hosting, licence and support spendLess electricity and cooling. Cooling alone is 7% to over 30% of data centre electricity, depending on efficiency
Right-size cloud and remove idle capacityTargets the 27% to 29% estimated wasteFewer compute hours. This follows from using fewer resources, and none of the research cited here measures it
Buy fewer devices and keep them longerFewer purchasesAbout 80% of a notebook's emissions occur in manufacturing, and adding two years of life reduces emissions by about 30%
Move workloads to more efficient facilitiesLower energy billsThe average data centre power usage effectiveness (PUE) reported in a 2024 industry survey was 1.56, so there is room to improve
Source cleaner electricity and lower-carbon equipmentLittle or no cost savingCarbon only. This is where the green case needs its own business case

One more finding is useful: end-user devices generate 1.5 to 2.0 times more carbon than data centres, and 50% to 60% of device emissions can be addressed through sourcing changes, mainly buying fewer devices per person and extending the life of each. That is a place where cost and carbon move together.

For companies that report under India's BRSR or the EU's CSRD, the same inventory work supports the energy and Scope 2 and Scope 3 figures those frameworks ask for. See our posts on the BRSR and on global ESG frameworks.

6. Limitations

  • Industry coverage is partial. Only two sectors have openly published percentile ranges. The full industry tables are paid research.
  • Geography. The enterprise survey covers US and Canadian organizations. Spend patterns elsewhere, including India, may differ.
  • Vendor and survey sources. Several waste figures come from vendors who sell optimization tools, or from respondents' own estimates. They indicate scale, not audited savings.
  • Dates. Some sources are several years old, as flagged in the text. Benchmarks move, and figures here are as of November 2026.
  • Overlap. Opportunities in different areas overlap and cannot be summed.
  • Carbon. We did not find a reliable published figure for IT's share of a typical company's total emissions, so we do not give one. Measure it for your own organization.

7. Our view

The silent factory is large enough to matter and fragmented enough to hide. For most sectors it is a low single-digit share of revenue, and for a mid-sized enterprise that still means tens of millions a year. The published evidence says a meaningful part of that spend is recoverable, particularly in cloud, software, support labour and ageing infrastructure, but it also says most organizations struggle to keep the savings they find. The practical answer is to see the whole estate, size each lever from your own data, act on the quick wins, remove recurring work, and put governance around the result. Do that, and the greener footprint follows as a by-product of the cost work, with the carbon benefit to be measured and reported alongside the savings.