This week The Register reported that Revolut, the British fintech, has cut its AI coding costs by roughly 90% in under a year. It did not do it by negotiating a better price or rationing access. It did it by stopping renting intelligence by the token and starting to run its own models on its own infrastructure. Commercial AI services are now the exception, not the default. It is a big-company story, but the principle underneath it is exactly the one Lansafe AI Studio is built on — and it applies to a twenty-person firm just as much as to a bank.
What Revolut actually did
According to the report, Revolut's engineers were burning through metered AI services at up to $10,000 a month each, with little to show for it. So the company imposed strict budgets, then went further: it adopted open-weights models and served them itself, at roughly a tenth of the per-token cost of the commercial tools. Around 2,000 engineers now use the system, with coding time down 60% and lead time down 70%.
Three details in the story matter more than the headline number:
The biggest wins were not glamorous.The largest gains came from letting AI deal with the ordinary, repetitive work — debugging, tickets, documentation — the things that eat a working week.
The bill became flat.No more bill shock at the end of the month. When the capability is yours, the cost is predictable, and nobody has to save their allowance for a special occasion.
The frontier models are still there — for the exceptions.Revolut did not throw the commercial services out. It routes the unusual cases to them and keeps everything else in house. Expensive AI became a scalpel, not a utility bill.
You can read the full story in The Register's report.
This is the approach we build on
We have said from the start that AI is powerful, but it should not be the answer to everything. The known, repeated parts of a business process belong in tested rules — software that does the same thing correctly every time, costs almost nothing to run, and does not need a meter running. AI earns its place on the uncertain, exceptional cases, with a person reviewing what it does. Revolut has just demonstrated the same thinking at enormous scale: own the everyday capability, rent the exotic.
The same goes for where your software lives. Everything we build is portable. It can run with us, or it can come home and run on your own hardware, inside your own walls, under your own policies. For regulated businesses that is not a nice-to-have — compliance designed in from the ground up beats compliance bolted on later, every time. Technology moves faster than legislation; a system that was designed to be compliant does not need heart surgery when the rules catch up.
You do not need 56 graphics cards
Revolut spent serious money on serious hardware, because it has thousands of engineers. That is not the bar for everyone else. The lesson for a smaller business is simpler: start with one problem that costs you real hours every week. Prove a solution against it. Keep what works. Because the solution is yours — not a subscription to someone else's roadmap — every improvement you pay for stays with you.
That is what we mean when we say build value, don't embed dependency. The world is catching up to the idea that renting intelligence forever is a cost centre, not a strategy. We would rather help you own the capability from day one.
If there is a process in your business that eats hours and behaves predictably, tell us about it — or if you would rather talk it through first, get in touch.