Callosum raises $10.25 million to challenge the monopoly in AI computing

While the entire world of artificial intelligence continues to revolve around a handful of major platforms, and the queue for NVIDIA GPUs stretches for months, London-based startup Callosum quietly closed a funding round of $10.25 million — and offered the industry food for thought on whether it's time to rethink the very architecture of AI computing.

The round was led by Plural — an early-stage European venture fund specializing in infrastructure technologies. Among business angels was Charlie Songhurst, a well-known serial investor and former Microsoft strategist. By AI industry standards, where funding rounds often reach hundreds of millions, the sum looks modest. But behind these dollars lies an idea capable of reshaping the market for AI computing infrastructure.

Callosum is developing software for multi-chip AI infrastructure. To simplify, the company creates an abstraction layer that allows computational tasks to be distributed among processors from different manufacturers — whether GPUs from NVIDIA, accelerators from AMD, specialized chips from Intel, Google, or any other vendor. The startup's name, incidentally, is no accident: the corpus callosum is a structure in the brain that connects the two hemispheres and ensures their coordination. An apt metaphor: Callosum aspires to become the connective tissue between disparate computing resources.

To understand why this matters, one must look at the current state of the market. Today, AI computing is practically synonymous with NVIDIA. Jensen Huang's company controls approximately 80-90% of the GPU market for neural network training. Its CUDA ecosystem has become the de facto standard: developers write code for CUDA, cloud providers purchase NVIDIA cards, and alternative chip manufacturers find it extremely difficult to penetrate this vertical integration. The result is a shortage of computing power, dependence on a single supplier, and rapidly rising prices. For companies training large language models, this means billion-dollar infrastructure budgets and strategic vulnerability.

This is where Callosum sees its niche. If a software layer allows efficient use of chips from different manufacturers, companies gain freedom of choice. They can combine equipment, optimize costs, and reduce dependence on a single vendor. For the market, this potentially means increased competition among chip manufacturers — AMD, Intel, and dozens of startups developing specialized AI accelerators will get a real chance to compete not just on hardware specs, but also on accessibility to end users.

The European context of this deal is also significant. Venture capital in AI infrastructure has traditionally been concentrated in Silicon Valley. Plural, betting on Callosum, is effectively declaring: Europe is capable of producing not just consumers of AI technologies, but creators of infrastructure layers. This echoes a broader trend — the European Union actively promotes the idea of digital sovereignty, and companies that reduce dependence on American and Asian chipmakers fit perfectly into this strategy.

Of course, Callosum's path will not be easy. The CUDA ecosystem took over fifteen years to form, and its network effects are colossal. Attempts to create alternative frameworks — from OpenCL to AMD's ROCm — have not yet seriously shaken NVIDIA's position. However, the situation is changing. Growing demand for AI computing makes single-vendor dependence increasingly risky, and major cloud providers — from Google with its TPU to Amazon with Trainium chips — are already investing in their own hardware. The market needs software glue to unite this fragmenting ecosystem.

Ten million dollars is a starting position, not the finish line. But the very bet on multi-chip architecture reflects a fundamental shift in industry thinking. The era when one chip manufacturer could dictate terms to the entire AI market may be reaching a turning point. And if Callosum can prove that heterogeneous computing works no worse than monolithic systems, everyone will feel the consequences — from cloud giants to startups training their first model.