London‑based Callosum announced a $100 million Series B funding round, backed by the Sovereign AI Fund and several European venture capital firms. The capital will be deployed to develop a new generation of AI hardware‑software co‑designs that decouple model training from traditional GPU‑centric pipelines. Callosum’s proprietary framework reimagines how neural networks interact with custom ASICs, promising up to a 40 percent reduction in energy consumption and a 30 percent boost in inference speed for large language models. The company also plans to expand its engineering team across the UK and open a research hub in Cambridge to collaborate with academic partners on next‑generation chip architectures.
The investment reflects a broader industry shift toward more efficient AI compute stacks, as the cost of training ever‑larger models strains both corporate budgets and environmental targets. By focusing on a model‑agnostic approach that can dynamically allocate workloads across heterogeneous processors, Callosum aims to address the bottlenecks that have emerged from the reliance on a narrow set of GPU vendors. Analysts predict that such innovations could democratize access to high‑performance AI, enabling smaller firms to compete with hyperscale players. Yet, the path to market adoption is fraught with integration challenges, as existing AI pipelines are deeply entrenched in established ecosystems.
Callosum’s leadership emphasizes that the new funding will accelerate the rollout of its open‑source SDK, fostering a community of developers who can contribute to and benefit from the platform. If successful, the startup could catalyze a paradigm shift in AI infrastructure, prompting cloud providers and enterprises to rethink their hardware procurement strategies and potentially reshaping the competitive landscape of AI model deployment.
This story was written by a language model from reporting by techstartups.com. Check the original before quoting it.