Meryem Arik

Meryem Arik is co-founder and chief executive of Doubleword, the London AI inference company formerly called TitanML. She founded it in 2021 with Dr Jamie Dborin, now chief scientific officer, and Dr Fergus Finn, now chief technology officer, building on the pair’s postdoctoral research at University College London. Arik studied theoretical physics and philosophy at Oxford and has been named in Forbes 30 Under 30 as a technology leader.

The cost of running models

Doubleword started as a platform for enterprises to run open-source, proprietary or custom models inside their own environments. When it raised a $12m (£9m) Series A led by Dawn Capital in May 2025, with Hugging Face chief executive Clément Delangue and Dataiku chief executive Florian Douetteau among its angel investors, Arik described the aim plainly: “We eliminate the heavy lifting of inference at scale so they can go from idea to production faster, without racking up technical debt.”

Its current product focuses on asynchronous and batch inference, the high-volume jobs behind AI agents, evaluations and data pipelines that don’t need an instant answer. Doubleword says this runs up to 90% cheaper than closed-source APIs, and cites its own work on 70 times faster model cold starts and fourfold compression of the memory models use while generating text. Customers and partners it names include Dataiku, nPlan, Element Materials and the University of Colorado Boulder.

Backed by the Sovereign AI Unit

On 16 April 2026 the government’s £500m Sovereign AI Unit named Doubleword as one of seven UK AI companies in its first round of support, giving it access to the AI Research Resource supercomputer network. The government described it as providing optimized inference and model governance tooling that lets regulated users run models securely in their own environments.

Arik is also a regular voice on the economics of AI. She spoke on running asynchronous agents in production at QCon London in March 2026 and discussed inference costs and UK investment on Bloomberg in June.

Why it matters to UK buyers

As AI moves from pilots into daily use, the bill shifts from building models to running them, and much of that work does not need to happen in real time. For UK IT leaders, cheaper batch inference on open models offers a way to control costs, and self-hosting keeps sensitive data inside the organization’s own environment. For partners building AI services for clients, Doubleword is a UK-based alternative to buying all inference from US model providers.

Arik has spent five years on one problem, the cost of running AI, and has the UK government’s compute behind her approach. If running models becomes the larger share of AI spending that focus is likely to look well chosen.