tools-only Software Development Costs and Maintenance: 2022–2025 Trends and AI Impact
Introduction
Software development costs encompass not only the initial build of a system but also the ongoing expenses to maintain, update, and support that software over its lifecycle. In fact, maintenance has long been recognized as a majority cost component in the software lifecycle. Historically, organizations have found that maintaining and enhancing software after its release consumes a significant share of total expenditure – often well above the cost of initial development . This report examines general software development cost distribution with a focus on software maintenance costs as a percentage of total cost, highlighting recent data from 2022–2025. It also explores how the advent and adoption of AI tools in software development are influencing these cost dynamics. Both traditional development approaches and emerging AI-driven practices are considered, drawing on recent research studies, industry reports, and authoritative publications. A summary table is included to compare historical and current maintenance cost percentages, providing a clear overview of trends.
Traditional Cost Breakdown: Development vs. Maintenance
For decades, software engineering research has noted that maintenance (bug fixes, updates, improvements, and support after initial deployment) tends to outweigh initial development in terms of effort and cost. Over the full lifetime of a typical software product, maintenance is often estimated to comprise around 75% or more of the total cost . In other words, for every dollar spent building the software, three or more may be spent keeping it running and relevant over time. In some cases – especially for long-lived systems or projects where new features and fixes continue for years – maintenance can consume an even larger share. There are reports that maintenance costs have reached as high as 90% of total software expenses in certain projects , leaving initial development as only a small fraction of the overall cost.
Historically, the balance between new development and maintenance has shifted upward in favor of maintenance. Studies over past decades observed that maintenance’s share of the cost was growing. One analysis indicated that the maintenance portion increased by roughly 50% over two decades, reaching on the order of ~90% of total lifecycle costs by 2010 . This means that where maintenance might have been perhaps ~60% of costs in the 1990s, it had grown to nearly 90% in some estimations by the early 2010s. The reasons are multifold: as software systems age, they accumulate “technical debt” (design compromises, outdated technology, etc.) that makes changes more laborious, and the longer a system is in use, the more updates and fixes it will require. Additionally, early project estimates often underestimate maintenance needs, leading to budgets being stretched or projects halted when maintenance overruns occur .
It’s also useful to consider how organizations plan for maintenance in practice. A common rule-of-thumb in industry is to budget a certain percentage of the initial development cost per year for ongoing maintenance. Traditionally, many companies allocate on the order of 15–25% of the development cost annually for maintenance and support of the software. For example, guidance for mobile apps suggests reserving about 20%–25% of the original development budget to cover maintenance needs . In the SaaS (Software-as-a-Service) domain, recent industry breakdowns similarly recommend that roughly 20% of the total development cost be budgeted for ongoing maintenance of the product . This means if an application cost $1 million to build, one might expect to spend around $200k per year on updates, bug fixes, minor enhancements, and infrastructure to keep it running. Over a span of 5 years, those maintenance expenditures could equal or exceed the initial build cost.
The table below summarizes historical and current estimates of software maintenance cost as a percentage of total lifecycle cost, illustrating how the balance has changed (or remained high) over time:
| Timeframe/Context | Maintenance Cost % of Total |
|---|---|
| 1980s (Historical) | ~50% (around half of total costs on maintenance) – early observations (estimates) |
| 1990s | ~60% (growing maintenance burden) – estimated |
| 2010 (Pre-AI era) | ~90% (in some cases, very high share) |
| 2020s (Current) | ~75–80% (commonly cited share for most projects) |
| Typical annual maintenance budget | ~20% of initial development cost (per year) |
| Future (Projected with AI assistance) | TBD – Potential reduction in development effort; maintenance % could adjust (see discussion) |
(Table: Historical vs. current software maintenance cost percentages. Early decades saw maintenance consuming roughly half of costs, rising to an estimated 60%+ by the 1990s. By 2010 some analyses put maintenance as high as ~90% of total lifecycle cost. In the 2020s, a typical project still devotes around 75% or more of total costs to maintenance , and it’s standard to budget ~20% of the development cost per year for maintenance. The impact of AI tools on these percentages is still emerging.)
Recent Data and Trends (2022–2025)
In the period 2022–2025, software maintenance continues to command a large share of resources, although modern development practices (Agile, DevOps, cloud services) have somewhat changed how these costs are managed. Rather than a distinct “maintenance phase” at the end of a project, many organizations now practice continuous delivery, where software is continuously updated. This blurs the line between new development and maintenance, but it doesn’t eliminate the cost – it simply spreads it as ongoing iterative work. Teams still find that a substantial portion of their effort goes into refining and supporting existing code versus creating wholly new systems.
Industry surveys and market research in recent years reinforce how significant maintenance spending is. According to a March 2025 industry report, the global software support and maintenance market (which includes technical support, updates, and upgrades for software) was valued around $650 billion in 2022 and is forecast to reach about $1 trillion by 2030 . This reflects a healthy growth (~6% CAGR), indicating that companies are investing heavily in keeping software running and up-to-date. The drivers cited include the increasing complexity of software and the need for frequent updates (for security, compliance, etc.), which all necessitate ongoing maintenance efforts . Notably, as more businesses have moved to cloud-based and subscription models, maintenance is often built into those subscription fees – but it remains a real cost, either borne by the provider or passed on to customers.
Within enterprise IT departments, the budget split between “keeping the lights on” (maintaining existing systems) and new initiatives has historically been lopsided toward maintenance. It’s not uncommon to see 60–80% of IT budgets devoted to maintaining current software and infrastructure, with only the remainder available for new development or innovation. While specific percentages vary by organization, the overarching trend is that maintenance dominates spending. Recent data suggests this is still the case in the 2020s. For example, one source notes that over the product lifetime, maintenance typically exceeds 75% of costs – indicating that even today, companies should expect the majority of software TCO (Total Cost of Ownership) to come after the initial release.
At the same time, neglecting maintenance can have high hidden costs. Security vulnerabilities, system outages, and technical debt accumulation are risks when maintenance is underfunded . High-profile failures (like major system breaches or downtime incidents) in recent years have been attributed to organizations deferring essential maintenance. This has reinforced the understanding that adequate maintenance investment is not optional if software is to remain secure and reliable. Thus, many businesses have been increasing their maintenance budgets or outsourcing maintenance to specialized firms, even if it means a large ongoing expense, because the cost of failure can be worse.
In summary, 2022–2025 data continues to show maintenance as a hefty slice of software cost structure – often on the order of 70–80% of total lifecycle costs for a typical application. The absolute spending on maintenance is growing globally with the expansion of software usage . The percentages as a share of total cost have not drastically dropped in practice, though modern architectures (e.g. microservices, SaaS) aim to make maintenance more efficient. Any significant change to this pattern would likely come from a breakthrough in how we develop or maintain software – which is where the influence of AI tools comes into play.
AI-Driven Development: Effects on Costs and Maintenance
The advent of AI and machine learning tools in software development is a recent trend that promises to reshape productivity and potentially the cost balance between development and maintenance. AI-powered code assistants (such as GitHub Copilot, Amazon CodeWhisperer, and others) have been adopted by many developers since around 2021–2023. These tools use generative AI to suggest code, automate writing boilerplate, and even generate entire functions or modules based on prompts. The big question is: how do these AI tools impact software costs, especially maintenance costs?
On one hand, AI tools can greatly accelerate development. Early studies have shown substantial productivity boosts for developers using AI coding assistants. For example, in a controlled experiment by GitHub, developers using Copilot were able to complete a coding task 55% faster than those without the AI assistance . This kind of speed-up implies that the initial development phase can be completed in significantly less time, potentially lowering the upfront cost of software creation or allowing more features to be built in the same time frame. In addition to speed, AI can help developers stay in “flow” and reduce mental effort on repetitive tasks (with many reporting less frustration and more focus when using AI assistance) . All these benefits suggest that AI-driven development could reduce the labor cost component of writing new code, shifting some effort away from humans to the AI. If a project requires fewer developer-hours for initial build, the percentage of total cost attributed to development might shrink relative to maintenance (or at least not grow as fast), assuming maintenance effort stays constant.
However, the story does not end there. AI’s impact on maintenance costs is two-sided. While AI can generate code quickly, concerns have arisen about the quality and maintainability of that AI-generated code. Recent research indicates that code written with the help of AI may introduce more issues that need refactoring or fixing later. A 2023 study by developer analytics firm GitClear, which analyzed over 150 million lines of code, found some concerning trends since AI coding tools became popular . Notably, they observed a sharp rise in “code churn” – the percentage of code that gets written and then discarded or rewritten shortly thereafter. The study projects that code churn will double in 2024 compared to pre-AI baselines . This suggests that while AI helps add code quickly, much of that code might not be right the first time and ends up being reworked. In essence, AI can accelerate the production of code, but if that code is suboptimal, developers (or the AI) will spend additional cycles fixing or improving it – which is a maintenance activity.
The GitClear findings introduced the notion of “AI-induced technical debt.” As GitClear’s founder put it, “AI code assistants excel at adding code quickly, but they can cause ‘AI-induced tech debt’” . Technical debt refers to the implicit cost of future rework when quick-and-dirty solutions are chosen now. AI may inadvertently encourage more of these quick additions that later need cleanup. From a cost perspective, this could increase maintenance effort down the line. Developers might have to spend more time reviewing AI-written code, adding tests to ensure it’s correct, and refactoring duplicated or sloppy code that the AI introduced . In DevOps teams focused on long-term code health, this has raised flags that the initial gains in speed could be offset by higher maintenance workloads to achieve the desired quality and stability.
That said, AI is not only being used for code generation. There are also AI-driven tools for testing, code review, bug detection, and even automated bug fixing. For instance, modern static analysis tools and security scanners employ machine learning to find bugs or vulnerabilities faster, potentially preventing some maintenance issues. AI-based test generation can create unit tests automatically, catching regressions before code is deployed. These uses of AI can reduce maintenance costs by identifying and addressing problems early or by handling routine maintenance tasks automatically. The full benefit here is still emerging, but we can envision AI helpers that, for example, automatically suggest a fix for a bug (saving a developer hours of debugging) or optimize some code for better performance or lower cloud costs (a kind of self-maintenance). As of 2025, such capabilities are in early stages, but tech companies are actively developing AI that can assist in the maintenance phase of software, not just initial development.
AI vs. Traditional Approach – Net Impact: In a traditional development approach (without AI), developers might spend longer to write the code, but each line is written with human intent and perhaps more consideration, potentially resulting in code that, while slower to produce, might need slightly less rework later. In an AI-assisted approach, the pendulum swings: development is faster, but it may produce more artifacts that require maintenance (rework) sooner. It’s not yet clear which approach yields a lower total cost over the software’s life – that may depend on how effectively the AI is guided and how disciplined the team’s engineering practices are. A highly skilled team can leverage AI to get the best of both worlds (fast development and maintainable output) by carefully validating AI contributions. In contrast, teams that treat AI output as “fire-and-forget” may find themselves paying the price in a later maintenance phase.
Productivity Gains and Cost Projections with AI
While concrete long-term data is still being gathered, experts are optimistic that AI will improve overall software economics. McKinsey and other analysts have estimated huge potential value from AI in software engineering. For example, Gartner has projected that by 2028, the systematic adoption of AI code assistants will yield at least a 36% increase in developer productivity (compound) compared to current baselines . Other industry experts have suggested productivity boosts on the order of 50% or more as generative AI tools mature . These gains imply that a lot of developer time could be freed up from brute-force coding. In theory, if developers are 30–50% more productive thanks to AI, projects might be completed faster or with smaller teams – reducing the direct cost of development.
If development costs shrink or timelines shorten, one effect could be that maintenance becomes a relatively larger slice of the pie (since it remains necessary work). However, proponents argue that with AI, engineers can reinvest time saved into preventative maintenance activities: improving code quality, adding more tests, refactoring code proactively – all of which could lower maintenance costs in the long run by preventing issues. In other words, AI could enable a shift from reactive maintenance to more preventative care of code. It’s also worth noting that as AI handles repetitive tasks, developers can focus on higher-value design and architectural decisions, potentially leading to better-structured, more resilient software that is easier to maintain over time .
At present (mid-2020s), we are just beginning to see the influence of AI on maintenance metrics. Some early adopters report that using AI for code reviews or bug fixes has reduced the time their teams spend on routine maintenance. On the flip side, as discussed, others have noted new types of maintenance work (like curating AI-generated code). Quantitatively, broad industry percentages of maintenance cost have not yet dramatically changed because of AI – maintenance is still likely in the range of 70% of total costs for many organizations – but there is potential for change in the coming years. Projections are still speculative: one scenario is that by the end of this decade, if AI tools become truly integrated, we might see the maintenance share of costs start to decline (e.g., perhaps maintenance could drop to, say, ~60% of total cost) because the total effort to build and maintain software would decrease overall. Another scenario is that maintenance stays high as a percentage, but the absolute costs go down because both development and maintenance tasks are made more efficient by AI.
The key point is that AI is reshaping how work is done: faster production, but with careful attention needed to maintain quality. As one DevOps commentary put it, “the speed benefits during initial development could be overshadowed by increased complexity during deployment and future iterations” if AI-generated technical debt isn’t managed . In effect, AI doesn’t eliminate maintenance – it changes its nature. There may be fewer mundane tasks for humans (since AI can handle some), but maintenance may become more about oversight: reviewing AI outputs, addressing higher-level design issues, and handling integration of components that AI churns out.
Conclusion and Outlook
In summary, software maintenance costs remain a dominant factor in total cost of ownership, traditionally consuming the majority of resources over a software system’s life. Recent data up through 2025 confirms that organizations often spend on the order of 75% (or more) of total costs on maintaining and evolving software, as opposed to initial creation . This has not drastically fallen in recent years – if anything, the complexity and criticality of software have kept maintenance investments high. Strategies like budgeting ~20% of development cost per year for maintenance are common sense practices to avoid underestimating these needs .
The rise of AI in software development offers both opportunities and challenges in this cost equation. AI-assisted development can reduce upfront development time and cost, potentially changing how much effort goes into the “new development” portion of a project. This could free resources to either do more within the same budget or possibly to reduce cost overall. At the same time, early evidence suggests that without care, AI-generated code can introduce higher maintenance workloads (through technical debt and necessary rework) . Thus, the net effect of AI on software cost distribution is still unfolding. It is plausible that as AI tools improve, we will see maintenance costs (as a percentage) stabilize or even decrease slightly, since some maintenance tasks might be automated and software quality (with AI’s help) could improve. Indeed, the long-term vision is that AI might handle routine maintenance – for example, auto-generating patches or optimizations – thereby lowering the human effort needed post-release.
For now, organizations should prioritize findings from recent studies: leverage AI to speed up development and catch straightforward issues, but also invest in good engineering practices to mitigate any downsides (like AI-induced bugs or duplicated code). Authoritative analyses (from firms like McKinsey and Gartner) are optimistic that AI will drive significant productivity gains , translating into cost savings, but they also emphasize the need for process changes and quality focus alongside tool adoption . In the next few years (2025 and beyond), we will likely gather more data on maintenance vs development costs in AI-rich environments. For the time being, software maintenance remains a large cost center – often the largest – and any reduction in that burden through AI or other means will have a notable impact on software economics. Teams that successfully harness AI to write cleaner code and automate fixes could see a real decline in maintenance effort, whereas teams that use AI purely to crank out more code might find the maintenance percentage staying as high as ever, or even rising.
Sources:
Boehm, B. et al. (classic software engineering economics data on maintenance vs development).
Pressman, R. Software Engineering: A Practitioner’s Approach (noted high maintenance cost percentages).
Wikipedia (Software maintenance) – “Over the lifetime of the product, software maintenance is estimated to comprise 75 percent or more of the total development cost.”
JPLoft Tech Blog (2025) – noting maintenance costs “up to 90%” in some cases and typical initial maintenance ~20–25% of dev cost .
Netguru (2025) – SaaS cost breakdown, ~20% of development cost should be budgeted for maintenance .
Verified Market Reports (2025) – global software support & maintenance market $650B in 2022, growing to $1T by 2030 .
GitHub Copilot Research (2022) – AI assistance led to 55% faster task completion .
DevOps.com (Feb 2025) – GitClear study on AI-related code quality issues (rising churn, tech debt) .
Gartner prediction (2023) – AI code assistants to boost dev productivity ~36% by 2028 .
Slalom/StackOverflow insights – experts foresee up to ~50%+ improvement in dev efficiency with GenAI .
Exiger (2025) – “Crumbling Bridges: The Failed Economics of Software Maintenance” (importance of maintenance investment) .
McKinsey (2025) – AI-enabled software product development lifecycle (discusses quality and speed improvements from AI) .