Callosum Raises $10.25M to Challenge Nvidia
AI hardware orchestration startup Callosum has raised $10.25 million in a new funding round. Founded by Cambridge-trained neuroscientists, the company aims to break Nvidia's dominance in AI data centers by developing a platform that orchestrates computations across a mix of GPUs, CPUs, and other specialized chips. The platform is designed to optimize for price and speed while reducing vendor lock-in, targeting both hyperscale and enterprise clients.
The funding round was led by European VC Plural and included participation from the UK's Advanced Research and Invention Agency (ARIA), a government-backed body focused on high-risk, high-reward R&D. Angel investors such as Charlie Songhurst, Stan Boland, and John Lazar also joined the round. The investment aims to expand the London-based team and scale its software stack. Callosum's founders, Danyal Akarca and Jascha Achterberg, met during their PhDs at the University of Cambridge, where their research merged neuroscience and computing. Their academic work, published in prominent journals like *Nature*, and industry experience with Intel and Google DeepMind, informs the company's core belief: true AI will emerge from specialized components working in concert, much like a brain, rather than from a single monolithic model on uniform hardware. The company's software platform is designed to orchestrate workloads across a diverse array of chip architectures. This includes hardware from Nvidia, AMD, AWS (Trainium/Inferentia), Cerebras, and SambaNova. Callosum claims this approach can yield 7x speed and a 4x cost reduction on complex tasks compared to single-hardware solutions. Nvidia's market share in the AI data center sector is estimated to be between 80-86%, creating a significant bottleneck and high costs for enterprises. This dominance is increasingly challenged by hyperscalers like Google, Amazon, and Microsoft developing their own custom silicon, and a wave of startups creating alternative hardware. Callosum's hardware-agnostic software approach directly targets the vendor lock-in created by Nvidia's CUDA ecosystem. This investment aligns with a broader venture capital trend of funding AI hardware and infrastructure startups to mitigate reliance on a single supplier. Nearly half of all global venture funding went into AI in 2025, with a significant portion targeting infrastructure and hardware acceleration. The move signals a growing recognition that hardware diversity and orchestration are critical for building scalable and cost-effective AI. The commercialization of deeptech from university settings is a key pathway for innovation, though Europe has historically lagged the US in this area. Initiatives like the UK's ARIA are designed to bridge this gap by funding ambitious projects with scientific roots. Callosum exemplifies the model of turning doctoral research into a fundable startup, a trend gaining traction to translate academic intellectual capital into market-ready products. While the Turkish deeptech scene is growing, with 653 startups identified, activity is concentrated in software applications. A recent partnership between InnovationQuarter and ITU ARI Teknokent aims to help Turkish deeptech scale-ups with validated technologies enter the European market, signaling a move toward international expansion and hardware-focused ventures.
