By early 2025, AI was already reshaping fraud detection, customer service and compliance across UK financial services. But for most firms the FCA regulates — particularly mid-sized incumbents and smaller fintechs — the GPU infrastructure, datasets and technical capacity needed to experiment safely were not readily available. The FCA's existing digital sandbox, launched a decade earlier, was not built for compute-heavy AI workloads.
Two things at once: a controlled environment where firms could prototype AI models against realistic data without exposing customers, and the processing power to run it. Without both, the regulator risked seeing a two-tier market — frontier firms experimenting freely while everyone else waited — and losing the ability to learn from those experiments in real time.
In June 2025, the FCA announced a partnership with Nvidia and its existing sandbox infrastructure provider NayaOne to build the "Supercharged Sandbox": GPU-enabled compute, the Nvidia AI Enterprise software suite, more than 200 anonymised datasets and over 1,000 APIs, all wrapped in regulator-supervised testing. The idea: the FCA would learn faster from live experiments than from consultation papers, and giving firms access to compute would open innovation to more of the market.
The first cohort of firms began testing in October 2025, focused on early-stage proof-of-concept models. Applications for a second cohort opened in early 2026, with the FCA signaling particular interest in agentic AI use cases — payment agents, compliance agents, and customer-service agents.
The arrangement is not without trade-offs. Industry commentators have noted that the partnership functions, in part, as a showcase for Nvidia's stack — increasing exposure to one vendor's tooling within a regulated environment. The FCA has been explicit that participation does not constitute approval, endorsement or authorization of any product.
Sandboxes usually exist to give firms room to test ideas under lighter rules. The Supercharged Sandbox reframes them as infrastructure: the regulator brings the compute, the data and the supervisory capacity, and lowers the cost of experimentation for firms that couldn't otherwise afford to try. The harder question — one other jurisdictions will need to answer — is how to provide that infrastructure without entrenching dependence on a single technology vendor.