Callosum Raises $100M Seed for AI Infrastructure Optimization

Callosum, a UK-based innovator in AI infrastructure, has successfully closed a $100 million seed funding round. The investment, notably backed by DCVC, aims to fuel the development of a novel orchestration layer designed to manage and optimize the execution of complex, heterogeneous AI workloads. This funding arrives as the demand for advanced AI capabilities, particularly in enterprise and scientific computing, continues to surge, driving up operational expenses.

The core challenge Callosum addresses is the escalating cost of AI inference. As AI models become more sophisticated and workflows increasingly involve multiple, interconnected agents, the consumption of computational resources—measured in tokens, GPU hours, or model runs—is expanding exponentially. Current hyperscale infrastructure, while powerful, is not seeing a commensurate decrease in unit costs, creating a significant economic hurdle for organizations deploying cutting-edge AI. This is particularly true as AI moves beyond single-turn interactions to persistent, autonomous operational loops.

Callosum's foundational insight is that AI is inherently multimodal, not monolithic. The company is building intelligent algorithms capable of decomposing complex tasks and routing them to the most cost-effective model and hardware combination available, while adhering to specified performance and scheduling constraints. This approach allows for the utilization of a diverse range of AI models, from large, general-purpose systems like ChatGPT 5 potentially running on high-end NVIDIA chips, to smaller, specialized models like Llama 70B or Gemma 31B, which can be highly efficient on dedicated silicon from partners such as Cerebras.

Early demonstrations highlight the potential for significant savings and performance gains. In one comparative analysis, Callosum demonstrated that Llama 70B running on Cerebras's specialized, low-latency silicon achieved comparable accuracy to ChatGPT 5, but at five times the speed and one-fifth the cost. This efficiency is a key driver for the company's mission. Callosum also announced a strategic partnership with Cerebras, enabling access to Cerebras's compute capacity via Callosum's APIs.

“The industry has been betting that one model or one kind of chip will rule them all,” stated Danyal Akarca, Co-Founder and CEO of Callosum. “But nature shows us the opposite. Intelligence is collective and will emerge from many specialized systems working together. We believe AI needs a new axis of scaling, where models and hardware co-evolve as a single system. That’s how we build AI that is not only more capable, but dramatically faster, more affordable and far more energy-efficient.” Akarca, along with co-founder Jascha Achterberg, brings a background in neuroscience PhDs from the University of Cambridge, providing a unique perspective on complex systems.

The company positions itself as a neutral orchestrator, akin to VMware's role in decoupling hardware and operating systems in the 1990s. By avoiding allegiance to specific model providers or chip manufacturers, Callosum aims to offer unbiased optimization across the rapidly diversifying AI hardware and software ecosystem. This independence is crucial, as AI companies and hyperscalers often have vested interests in promoting their own proprietary solutions. The market for AI infrastructure and optimization is rapidly expanding, with projections indicating substantial growth driven by the increasing adoption of AI across all industries.