AI will create real inequality between engineering firms. But the dividing line is not the one most people draw. The lazy version of the story says the split runs between firms that use AI and firms that ignore it. Use it, you win; ignore it, you fall behind. Tidy, and not really true.
For an engineering business the real line sits somewhere else entirely. It runs between the firms that rent their AI from a hyperscaler and the firms that own it, on a box, in their own building.
Renting looks like the safe side
And it is seductive. It is cheap to start, nobody has to learn anything difficult, and you can tell yourself you have “adopted AI” by the end of the afternoon. Then reality arrives in instalments:
- The bill climbs month on month, indexed to your usage.
- Your data takes a trip to a data centre on the other side of the planet every time someone asks a question.
- You are helping train a model you do not own, on terms you do not control.
None of that is hypothetical. It is the standard arrangement. And for a firm handling confidential drawings, test data or customer records, the second point alone can stall a project for weeks while legal works out what is even allowed to leave the building.
Owning it is the side people understand least
Mostly because it is so badly explained. The reality is duller, and better, than the fear. A box that costs five or six thousand pounds, then close to nothing to run. Self-hosted models that now match the cloud frontier on the jobs that actually matter to most departments: R&D, sales, support, finance, production, procurement. The data never leaves the building, so there are no weeks lost to screening documents before you can start.
And a detail that still surprises people: on one of our deployments the energy use came out around a hundred times lower than the cloud equivalent, from a single GPU sitting in a normal office. Payback in a couple of months, and then it simply keeps earning.
Why this becomes a competitive gap
Because the two sides compound in opposite directions. The renting firm’s costs rise with success: the more useful the AI becomes, the more it charges. The owning firm’s costs fall to near zero after the initial outlay, while the capability it has built stays in-house and keeps improving. Give that a few years and you no longer have two firms using AI. You have one firm that owns an asset and one that rents a liability.
This is the whole thesis behind Ascentis AI: self-hosted systems you own outright, deployed on your infrastructure, built around the commercial functions that actually move revenue. If you want to see what “owning it” looks like in a finished product, Cortex runs entirely on your hardware, with no external AI calls.
So if you have looked at self-hosting and walked away, I am genuinely curious: what stopped you, the cost, the skill gap, or simply not believing it could be that simple?
Weighing up deployment models? Read our full comparison: self-hosted AI vs cloud AI for engineering and manufacturing businesses.