Rethinking AI infrastructure: Callosum’s $10.25 million bet against the one-size-fits-all approach

Take one architecture, add more of the same chip, and scale: that was the dominant strategy for building more capable deep learning systems until now. But the problems that AI is increasingly being asked to solve (from drug discovery to industrial automation, ecc.) are not homogeneous. They are composed of many distinct sub-tasks, each with different computational demands and latency requirements.

It is this structural mismatch that London-based startup Callosum, founded by Danyal Akarca and Jascha Achterberg, has been built to resolve. The two cofounders crossed paths during their PhDs at Cambridge, publishing research in several Nature journals and collaborating with Google DeepMind. The company name is deliberate, referring to the corpus callosum, the neural bridge linking the brain's two hemispheres, allowing specialised regions to function as one coherent system. Intelligence in biological systems does not arise from replicating a single cell type at scale, but from the coordinated interplay of many specialised ones: Callosum's conviction is that AI infrastructure should follow this same principle.

Their software orchestrates AI workloads across chips from multiple manufacturers, distributing each sub-task to the hardware best suited to handle it, whether that means a different processor architecture, a different cost tier, or a different latency profile. The result is a system that is simultaneously more capable and cost-efficient than its homogeneous counterpart by using the right component for each job.

The company recently raised $10.25 million in pre-seed funding led by Plural, alongside other angel investors. The capital will go towards expanding the team, entering the US market, and laying the groundwork for hardware infrastructure, with longer-term ambitions that stretch to rethinking how data centres themselves are designed.

As debates about digital sovereignty intensify across Europe, infrastructure bets that reduce concentration risk in the AI supply chain carry strategic weight beyond their commercial value.

Sources: Callosum | Crunchbase

Founders: Danyal Akarca, Jascha Achterberg