Doubleword exists to solve a problem that grows with every enterprise AI deployment: running models reliably, cheaply, and securely on your own infrastructure. Originally founded as TitanML in 2021 by Meryem Arik, Dr. Jamie Dborin, and Dr. Fergus Finn, the company emerged from the founders' physics and quantum machine-learning research at Oxford and UCL. They discovered that model-compression techniques from that work could dramatically improve inference performance, and built a product around making self-hosted inference simple.
The platform unifies the messy parts of inference deployment. Instead of stitching together serving frameworks, hardware-specific optimizations, and orchestration, teams can deploy and manage models through a single self-hosted layer that runs across different hardware. The company reports up to 90% reductions in compute costs and up to 20x latency improvements within hours of deployment, which matters most to enterprises trying to control the spiraling cost of running models at scale.
Doubleword's positioning is explicitly enterprise and security-first. Many regulated organizations cannot send data to third-party API endpoints, so the ability to run models inside their own environment, on their own hardware, is a hard requirement rather than a preference. Partnerships with Snowflake and Dataiku place Doubleword inside data and analytics workflows where models need to be served close to enterprise data.
In May 2025, Doubleword raised a $12 million Series A led by Dawn Capital, with K5 Tokyo Black among investors and angels including Hugging Face CEO Clement Delangue and Dataiku CEO Florian Douetteau. The funding supports the company's push to make self-hosted inference effortless for enterprises and to grow its presence in the US after relocating its center of gravity from its UK roots. The founding team's deep technical background remains a core part of the company's identity.