The old default is breaking
A publicly listed European logistics group runs 11 Salesforce orgs, 4 ServiceNow instances, and a Workday deployment that costs $8.4M per year in licenses alone. In 2025, the CIO counted 1,340 custom fields, 92 integration middleware flows, and 27 full-time staff dedicated to keeping the stack glued together. The SaaS was supposed to be the cheap option.
That was the calculation a decade ago. It no longer holds.
Why buy won the 2010s
Buying SaaS made sense when custom development cost $1,200 per function point and took 18 months per module. Salesforce, Workday, and ServiceNow turned capex into opex and moved the integration burden onto vendors. Gartner estimated in 2019 that enterprises ran 80% of new workloads on packaged cloud applications. The build option was reserved for true differentiation.
The hidden cost was configuration. The average Salesforce org we’ve audited carries 340 custom objects and 1,800 validation rules. At that point, the customer is building software — just inside someone else’s runtime, at someone else’s margin.
What AI generation changed
AI-native generation collapses the cost curve that made buy the default. A workflow that took a four-person team 14 weeks to build in 2021 now takes two engineers nine days when the application is generated from a JSON descriptor and a business specification. The descriptor is the source of truth. The code regenerates on demand.
We’ve measured this across 23 internal migrations. Median time from requirement to production-grade screen dropped from 41 days to 6. The cost per function point fell under $180.
The three categories where build now wins
Three categories flip to build under the new math. First, workflows that require deep integration with a proprietary data model — the kind that generate 40% customization inside a SaaS tenant. Second, regulated processes where the audit trail has to live inside the company’s own control plane. Third, any capability where the SaaS vendor charges per seat for a workflow touched by more than 3,000 employees.
In each case, the five-year TCO of generated software comes in 45% to 70% below the SaaS equivalent. The delta is no longer labor. It’s licensing.
What doesn’t change
Email, CRM pipelines, general ledger, and payroll still belong in packaged software. The SaaS vendors have 20 years of domain depth and compliance certifications that no generation pipeline can replicate in a year. Build-vs-buy isn’t a reversal. It’s a redrawing of the line.
The line moved because generation quality crossed a threshold. When a system can produce a 200-screen application with correct validation, role-based access, and audit logging in a weekend, the portfolio of candidates for custom build expands by an order of magnitude.
How the decision should be framed in 2026
The question is no longer “build or buy.” It’s “where does each dollar of software spend earn the most leverage over the next five years?” For workflows unique to the business, the answer is increasingly to generate. For commodity processes, the answer is still to buy. The mistake is assuming the 2015 ratio still applies.
The bottom line
SaaS isn’t going away. But the portion of enterprise software that justifies its license fee is shrinking. CIOs who re-run the build-vs-buy calculation with 2026 generation costs are finding 20% to 35% of their SaaS footprint sitting on the wrong side of the line. That’s the opportunity.