Potentially AI raises £4.9m and lists ‘collective AI’ platform on AIM

Potentially AI’s listing is notable because it’s one of the first pure‑play “AI orchestration” companies to go public rather than build its own foundation models. Its pitch is that the frontier‑lab arms race is wasteful and fragile: access to models like GPT‑5.x or Claude Mythos can change overnight, and training your own models is capital‑intensive. Instead, Potentially wants to sit on top of 1,000+ open and closed models, routing users to the right system while managing tokens, data and sovereignty.

If it works, that’s a powerful lever in the AGI era. As models proliferate, the scarce resource becomes not parameters but trust, routing and control. Aggregators that can enforce data‑retention limits, regional constraints, and compliance rules across multiple backends could become indispensable to enterprises and governments that fear being locked into one US lab or another. It also gives open‑weight models a commercial distribution channel alongside closed APIs.

From a competition standpoint, this is an explicit bet that the “meta‑layer” of AI—routers, agents, marketplaces—will capture significant value even as foundation models commoditise. How Potentially executes will be an early test of whether that thesis supports a listed company or remains the province of developer tooling startups and cloud providers.