London-based AI startup Callosum has raised $100 million in seed funding to build software that helps AI tasks run on the chips best suited to handle them. Atomico led the round, with Plural, DCVC, and the UK Sovereign AI Fund also taking part.
Callosum was founded in 2025 by Danyal Akarca, its CEO, and Jascha Achterberg, its CTO. The company’s central idea is simple: AI should not depend on a single type of chip. Callosum calls this approach “heterogeneous intelligence.”
That idea arrives as the AI industry shifts from building models to running them at scale. The process of running trained models to produce outputs is called AI inference, and global spending on inference will surpass spending on model training in 2026.
Why AI chip choice is becoming a bigger issue
AI systems need hardware to train models and produce results from them. For years, the standard way to scale AI was to run bigger models on more of the same chip, usually Nvidia GPUs. That approach created a familiar path for companies building and operating AI systems, but it does not mean every task works best on the same hardware.
Specialised chips already target specific types of AI work. Cerebras, Rebellions, and Axelera each represent this direction, with hardware designed for particular workloads. The challenge is deciding which chip should handle each task and moving work between different types of hardware.
No adopted system exists to route different tasks to the hardware that handles them best. Callosum is developing software around that gap, with the goal of optimising AI task execution across different chips.
The distinction matters because AI workloads do not all operate in the same way. A system that can match tasks with suitable hardware would address the problem Callosum has identified: AI infrastructure built around one chip type may not fit every kind of work.
A new government investor joins the round
The UK Sovereign AI Fund participated in Callosum’s seed round and made the startup its first equity investment. The fund is a £500 million government venture initiative launched in April 2026.
Its investment gives Callosum a direct connection to a UK government-backed venture initiative while the company develops its software in London. The fund’s first equity investment also places Callosum at the start of that investment programme.
The timing reflects the scale of the market Callosum is targeting. Global spending on AI inference will reach $23.3 billion in 2026, out of a total AI infrastructure market worth $42 billion. Spending on inference will surpass spending on model training during the year.
Callosum’s place in the changing AI infrastructure market
As companies move from building models to running them at scale, the hardware supporting those systems becomes part of the operating challenge. Running larger models on more of the same chip has been the standard approach, but specialised processors offer another path for specific AI work.
Callosum’s “heterogeneous intelligence” approach focuses on coordinating those different options rather than treating one chip type as the answer for every task. Its software is intended to optimise execution across different chips, linking the choice of hardware to the work an AI system needs to perform.
The company has not been described as making a new chip. Its focus is software that manages AI task execution across existing categories of hardware, including general-purpose chips and specialised processors designed for specific AI work.
That leaves Callosum addressing a coordination problem in a market where inference spending is growing to $23.3 billion in 2026. The company’s seed funding will support its effort to develop software for that task, while Atomico leads a group of investors that includes Plural, DCVC, and the UK Sovereign AI Fund.
Callosum was founded only in 2025, making the $100 million seed round the first major funding event identified for the company. Its founders, Danyal Akarca and Jascha Achterberg, are building around the view that AI workloads should be matched with the chips that handle them best rather than routed through a single hardware type.
For now, the company’s core proposition is clear: AI infrastructure needs a way to work across different processors, and Callosum is developing software to provide that routing. The company’s funding comes as AI inference becomes a larger part of the $42 billion AI infrastructure market in 2026.
