In January 2026, a single trading session erased roughly $300 billion in market value from software companies. Analysts called it the 'SaaSpocalypse.' But that's not the real story. The real story is quieter — and it's happening in boardrooms, not on NASDAQ.
After 15 years building enterprise software for organizations including the Dubai Royal Family, Leonardo SpA, NEOM, and dozens of mid-market companies across Europe and the MENA region — through my company Lasting Dynamics — I can tell you that the economics of custom software have changed more in the last two years than in the previous fifteen combined. Enterprise software that used to cost €2–3 million and take 18 months to deliver is now being built in 5–6 months for €300–500K. This isn't a pitch. These are real numbers from real projects running right now.
The question for every CTO and CFO in 2026 is no longer 'can we afford to build custom software?' It has become: 'can we afford to keep paying for SaaS that doesn't fully serve us?'
Three Forces That Broke the Old Equation
The traditional logic was simple and compelling: SaaS = fast deployment, predictable costs, zero maintenance overhead, someone else's problem. For two decades, this logic was mostly right. Then three things happened simultaneously — and broke it.
1. The SaaS AI Tax: +20–37% at Every Renewal
Starting in 2024, every major SaaS vendor began bundling AI features into existing product tiers — whether you asked for them or not. Tropic, a procurement intelligence firm, coined the term 'AI Tax' to describe the 20–37% price uplift vendors impose at renewal. Zylo's 2026 pricing report confirms SaaS inflation is running at nearly 5× the general market inflation rate — vendor prices rising 11–14% annually against GDP inflation of approximately 2.7%. Among the top 500 SaaS companies, there were 339 pricing and packaging changes in 2024–2025 alone.
2. You're Paying for Half a Platform Nobody Uses
The average enterprise manages 291 SaaS applications as of 2026, up from 110 in 2020. And according to BetterCloud, 51% of purchased licenses go unused. In 61% of organizations, unplanned SaaS cost increases forced cancellation of projects or initiatives in the past 12 months. You are paying full AI-surcharge price for a platform half your team isn't using — and the half that does use it consumes perhaps 20% of its features.
3. AI Collapsed the Cost of Building Custom
This is the structural shift that changes everything. GitHub's research across 4,800 developers shows AI coding tools deliver 55% faster task completion. McKinsey puts time-to-market reductions at 30%. Independent research shows AI reduces time spent on testing, documentation, and code review by 26–80% depending on the function. The custom software development market is now growing at a 22.6% CAGR — because the economics have flipped. The compounded effect: a project that required a team of 15 engineers for 18 months in 2022 now requires a team of 5–6 for 5–6 months. Same scope. A fraction of the cost.
Michele's Take
Three years ago at Lasting Dynamics, a mid-complexity enterprise workflow platform took us 14 months and a team of 12. Last quarter, we delivered an equivalent system for a European logistics company in 5 months with a team of 5 — using AI-assisted development throughout. The output quality was higher. Test coverage was better. The cost to the client was €380K instead of €1.8M. The math has genuinely changed — and we see this compression on virtually every enterprise project we take on today.
The €400K Software That Used to Cost €2M
Let me make this concrete. Here is the typical cost breakdown for a medium-complexity enterprise platform — a workflow automation system with role-based access, multi-tenant architecture, API integrations with 4–5 external systems, and an analytics dashboard. The kind of platform a mid-size logistics, pharmaceutical, or financial services company would commission.
| Component | 2022 Cost (EUR) | 2026 Cost (EUR) | Reduction |
|---|---|---|---|
| Architecture & discovery | €120K–180K | €35K–55K | ↓ ~70% |
| Backend development | €600K–900K | €120K–180K | ↓ ~75% |
| Frontend & UX | €200K–350K | €60K–90K | ↓ ~70% |
| Testing & QA | €150K–250K | €30K–50K | ↓ ~80% |
| DevOps & infrastructure setup | €100K–180K | €25K–40K | ↓ ~75% |
| Documentation & project management | €180K–240K | €40K–60K | ↓ ~75% |
| Total (mid estimate) | €1.7M–2.35M | €310K–475K | ↓ ~75% |
The reduction is real, and the driver is not cheaper developers. European senior engineer rates have actually increased to €90–120/hour. The driver is velocity: the same engineer now accomplishes in 3 hours what previously took 10. AI coding assistants, AI-generated test suites, AI documentation, and AI-assisted architecture review have restructured the effort distribution of every software project.
“We built our AI system on OpenAI and dropped our contract for Salesforce CRM. We saved approximately $2 million annually on that platform alone — and that was just the beginning.”
— Sebastian Siemiatkowski · CEO, Klarna (2025)
Klarna is no longer an outlier. According to Velsof's 2026 research, 35% of enterprise teams have already replaced at least one SaaS product with custom-built software, and 78% plan to build more custom tools this year. The break-even analysis — once reliably 3–5 years — has shifted dramatically. A custom build replacing €60K/year in SaaS costs now typically breaks even in 14–19 months (APIPilot, 2026). For a 200-seat deployment, every year beyond month 19 is pure savings.
The Break-Even Has Moved
The healthcare example from Velsof's research is instructive: a 340-staff organization paying €97,920/year in SaaS subscriptions (covering only 60% of actual workflow needs) commissioned a custom build at €185K with €37K annual maintenance. Break-even: month 19. Three-year savings: €108K. By year 5, the savings compound to over €300K. This is increasingly typical, not exceptional.
The Real Prize Isn't Cost — It's Your Data
The cost inversion gets the headlines, but it's not the deepest advantage. When you build proprietary software, you own every byte of usage data your team generates — and that data becomes a compounding asset. Imagine replacing Slack and Jira with a single system where tickets, conversations, and decisions converge in one data model. An AI layer can then actually learn from how your organization works: it spots bottlenecks, proposes and ships automations on its own, installs corrective guardrails the moment a process drifts from policy, monitors that work is progressing as intended, and surfaces the right suggestion to the right person at the right time. No SaaS vendor can give you that — your data is fragmented across their platforms and trapped inside their model, not yours. With custom software, the system gets smarter the more your team uses it. That feedback loop is impossible to rent.
SaaS Fatigue Is Now a Budget Crisis
It's no longer just friction. For many enterprises, unplanned SaaS costs are actively crowding out strategic initiatives. The numbers tell a clear story:
- 61% of organizations cut projects or cancelled initiatives in 2025 because of unplanned SaaS cost increases (BetterCloud 2026)
- The average enterprise's 291 SaaS tools have extensive functional overlap — most organizations have 3–5 tools doing variations of the same thing
- The true TCO of enterprise SaaS is 2.5x–4x the advertised subscription price, after implementation, integration maintenance, training, and workflow workarounds
- Workers lose an average of 51 minutes per week to tool fragmentation and context-switching across disconnected platforms — over 44 hours per employee per year
- AI-feature surcharges have driven a 34% average cost increase since 2024 — for features most enterprises have not adopted or validated
The most damaging entry on that list is the last. SaaS vendors are raising prices to fund their AI infrastructure costs — and passing those costs directly to enterprise customers who may not be using the AI features at all. You are subsidizing vendor R&D on your operating budget.
I'm Not Saying to Throw Everything Away
Let me be clear, because nuance matters. SaaS is not dead. It won't disappear. There are entire categories where SaaS will remain the right decision indefinitely — email infrastructure, payment processing, HR administration for small-to-mid teams, basic analytics. These are commodity functions where vendor scale benefits you. The question is not SaaS vs. custom as a binary. It's about knowing where each model creates value — and where it destroys it.
The framework I use when advising enterprises is three questions:
- Does this function create competitive differentiation? If yes, build custom. If it's generic overhead, buy SaaS.
- Is your per-seat cost scaling faster than the value you're extracting? If yes, model the custom build at 36 months. The answer often surprises CFOs.
- Are you using more than 60% of the platform's feature set? If you're consistently below that threshold, you're paying the vendor's R&D tax on features your team ignores.
| Use Case | SaaS | Custom Build |
|---|---|---|
| Email, calendar, meetings | ✅ Always SaaS | — |
| Payment processing | ✅ Always SaaS | — |
| Core operational workflows | ⚠️ Evaluate at year 2–3 | ✅ Build if differentiating |
| Customer-facing product features | ❌ Rarely right | ✅ Build |
| Proprietary data pipelines & AI | ❌ Never | ✅ Build |
| Compliance-critical processes | ⚠️ Sector-dependent | ✅ Usually build |
Three Signals It's Time to Build Instead of Buy
- Your next SaaS renewal includes an AI surcharge you haven't validated. Before you sign, commission a 36-month cost model comparing renewal vs. custom build. In 2026, that model frequently favors building — especially above 100 seats.
- Your team builds workarounds around the tool. Every Zapier automation, every Excel export, every manual step that bridges two SaaS platforms is a signal: the software doesn't fit your workflow. Custom software starts from your actual workflow — not the vendor's assumptions about what your workflow should be.
- You're paying per seat for more than 200 users. Per-seat pricing is linear. Custom software scales with infrastructure — meaning the marginal cost of adding users approaches zero. The crossover point is lower than most CFOs assume.
Where This Is Heading
Gartner predicts that by the end of 2026, 75% of developers will be orchestrators — defining what to build, validating what AI generates, and owning production reliability rather than writing every line by hand. The marginal cost of software will keep falling. The organizations that understand this now and start building strategic software assets — instead of renting commoditized platforms at a 2.5–4x markup to subsidize vendor AI research — will carry a structural cost and capability advantage into the next decade. At Lasting Dynamics, we're seeing this shift accelerate on every enterprise engagement. It's not a prediction anymore. It's a measurement.
The question your organization needs to answer in 2026 is not 'can we afford to build custom software?' Modern AI-assisted development has made that question obsolete. The real question is: which parts of your software stack are creating competitive advantage — and which ones are you renting from vendors at a 2.5–4x markup to fund their AI research?
That calculation — made honestly, with real 36-month numbers — will shift the build-vs-buy decision for most mid-to-large enterprises in ways nobody anticipated when SaaS was the safe default. The safe default has changed.