For a decade or more, “cloud first” was the responsible answer, and “SaaS first” its natural extension. I’m now convinced that, as a default, it should be dead. Somewhere along the way it stopped being a strategy and became a habit.
If your pockets are deep, the habit is affordable. Outsource everything, absorb the subscription inflation, move on. But for those of us who must demonstrate value for money to customers and regulators, handing our compute decisions to someone else’s business model is becoming indefensible as a default.
Two things changed. First, our developers. Be honest about what most business applications are: create, read, update, delete, wrapped in workflow. With AI, our engineers now build and support applications at a pace where their capacity is no longer the constraint it was. Compare that to subscriptions that rise every year for features we never use, plus the standing overhead of administering licences every time someone changes roles or leaves. The build-versus-buy question we considered closed is wide open.
Second, the models themselves. As we embed GenAI into business solutions, running everything on frontier models gets expensive fast. For specific, well-bounded needs, open-source and small language models do the job at a fraction of the cost and the economics may stack up to host those models on your own compute.
And if you run critical infrastructure, you never went all-in anyway. Your OT environment, maybe your records management, parts of your data estate: on-prem for regulatory or operational reasons. Hybrid isn’t a future state. It’s your present. This isn’t contrarian anymore. IDC found 86 per cent of CIOs planned to repatriate at least some workloads, while almost none plan a full exit.
The answer isn’t to swap one dogma for another. On-prem first would be equally lazy. We’re building a decision framework for this, and it isn’t finished, but its first principle is settled: no defaults. Every workload earns where it runs, on the merits of the best long-term decision for the business, not on the momentum of a decade-old mantra.
Cloud first was the right answer to a question we’ve stopped asking. It’s time to start asking again.