A decade ago, every serious company was moving to the cloud. This year, AI is quietly bringing a lot of that software back on-prem.
The trade-off that built SaaS
The move from CAPEX to OPEX let companies smooth costs and avoid heavy upfront investment. Less cash upfront, no configuration, no consultant time. The catch, rarely said out loud: bought hardware disappears from your bottom line after 36 months, but software subscriptions never do. That was usually someone else’s problem, and less visible to investors, because licence costs started lower than depreciation. Building and maintaining your own software used to be slow, expensive and risky. That is precisely why the turnkey SaaS gold rush succeeded.
AI changes the maths
AI development tools have collapsed the cost of building and the effort of maintaining. The original trade-off no longer holds, and the market has noticed. I read a striking piece on TechCrunch about the “SaaS Apocalypse”: $285 billion in software market cap vanished in 48 hours. Salesforce down 33%. Atlassian down 35%. Software price-to-earnings ratios below the S&P 500 for the first time in history. Investors have understood that AI reshuffles the make-or-buy cards for every line of enterprise software. I can say it with some confidence, because I am building one of those alternatives right now.
Fit, flexibility, and freedom from lock-in
Forrester calls 2026 the year of tech nationalism in AI. The sharpest leaders I speak to are not asking which AI tool is best. They are asking how fast they can build a tailored solution that gives them fit, flexibility and freedom from rigid vendor lock-in. They do not fear AI. They learned something expensive the last time they outsourced a critical layer of their business.
None of this means SaaS is dead — it still lowers barriers and offers real flexibility. It means there is now a serious alternative that was impossible a few years ago: owning the capability outright, on hardware you control, at a cost that does not climb with every query. It pairs with the reality that your cloud AI bill is currently subsidised, and with the case for building on what only your company knows.
Is your software still running on someone else’s terms?
Weighing up deployment models? Read our full comparison: self-hosted AI vs cloud AI for engineering and manufacturing businesses.