Artificial intelligence scaled on a bet: that bigger models, more identical chips and more data would keep delivering. It worked remarkably well. But what began as a practical choice has hardened into the only way we know how to build. We now design everything - our models, our inference systems, our entire compute infrastructure - around the premise that scaling means more of the same models on the same chips.

That bet is breaking down.

It is breaking because the vast majority of intelligence deployed in the real world will not come from a single model. It will come from systems. Different models accelerated by different chips, each with distinct trade-offs, in direct feedback with the world. The problems we actually need AI to solve - in science, in industry, in security - are heterogeneous. They demand many forms of intelligence working together, different but complementary, across a vast and growing range of problems that no single model can serve. The next huge leaps in capability, cost and speed will not come from building single models bigger on homogeneous systems. They will come from heterogeneity.

Heterogeneous Intelligence is the defining shift of the next era of AI.

The supply side is catching up to this reality. New chips are emerging from every direction: photonic processors, in-memory inference architectures, neuromorphic chips, optical interconnects, analog accelerators, quantum processors, thermodynamic compute. Each with radically different strengths, while the software barriers that once locked us into single vendors are quickly dissolving. But no infrastructure exists to bring them together - no one is building vertically integrated Intelligent Systems that are greater than the sum of their parts.

This is why Callosum exists.

We are defining a new class of company at the intersection of frontier AI and compute. We call ourselves the Intelligent Systems Company because we believe intelligence comes from the system, not the models. We co-evolve chips and intelligence together, because co-evolution unlocks a configuration space that didn't exist before - where the properties of diverse chips create new algorithmic possibilities and new architectures reveal new ways to exploit the silicon. This is where the greatest surface area for discovery lives. The improvements in performance and cost are orders of magnitude beyond what optimising within the current paradigm can deliver.

We are building for the teams bringing AI to the world's hardest problems and for the chip makers whose silicon deserves better than the stack it's trapped in and the market that keeps it there. Everything we discover ships directly into production, powering our partners in AI and silicon alike.

We envision a future where diversity of compute gives rise to entirely new forms of intelligence. Where AI systems solve problems that today remain beyond reach. The discoveries we do not yet know to look for. We believe this future is not only possible - it is brighter, more open and more ambitious than the path we are currently on.

Our world-class founding team has spent their careers at the frontier of new paradigms of intelligence, neuroscience and hardware-software co-design: designing chips, writing kernels, building clusters, operating clouds, creating new model architectures and systems from scratch - across Cambridge, Oxford, MIT, Imperial College London, Microsoft, ETH Zurich and Intel. No other team on earth sits deeper across both intelligence and compute. We came together around a shared conviction: that the current paradigm is not enough and that the era of heterogeneous intelligence must be built. We are building it.

We have raised $10.25 million, pre-seed led by Plural, alongside 22 world-class investors and supported by the Advanced Research + Invention Agency.

Today we are publishing the first results from our technology. Read our discoveries, our scaling principle and our roadmap. This work demands extraordinary people. Come join us.