Revolut cut AI coding costs 90% by self-hosting on 56 Nvidia B200s
Revolut pulled 2,000 engineers off metered AI APIs and moved them onto self-hosted open models running on 56 Nvidia B200 GPUs. According to a StartupHub.ai account of a talk by David Abitbol on the AI Engineer YouTube channel, the company reports that move cut its AI coding costs by 90%.
The source article notes it did not watch the talk directly and did not contact Revolut, so the 90% figure is Revolut’s own claim as relayed through that summary.
The Pattern Worth Copying
Revolut did not replace frontier models entirely. The approach was a split: keep paid frontier models for the narrow slice of tasks where they genuinely outperform, and move routine, repeatable work to open models the team runs or fine-tunes on its own infrastructure. Decagon, per Jesse Zhang, followed the same logic.
This is the part operators should pay attention to. The cost argument for self-hosting has always been there on paper. What Revolut is claiming is that it is real at scale, with a specific GPU count and a specific cost reduction attached to it.
The Operator Takeaway
If you’re running any kind of high-volume AI workflow, the two-tier approach is worth modeling out. Frontier API calls for judgment-heavy tasks, self-hosted open models for the repeatable, structured ones. The crossover point where self-hosting pays back the GPU investment depends on your volume, but 2,000 engineers generating continuous coding assist requests is clearly past it.
For solo operators and small teams not running 56 GPUs, the same logic applies at smaller scale through providers that offer open model hosting at lower per-token rates than the frontier APIs. The underlying principle holds even if the infrastructure is rented rather than owned.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
