UK AI infrastructure startup Callosum is reportedly seeking up to $100 million in fresh funding as it works to build software that can make different AI chips work together. The company is taking a different approach to AI infrastructure, betting that the future will not depend on one type of processor or one AI model.
Callosum previously raised $10.25 million in seed funding, led by Plural, in February 2026. In May, reports said the London-based startup was in talks to raise as much as $100 million (£75 million) in a new round.
Important: the $100 million figure was reported as a target/fundraising effort, not a confirmed completed $100 million round in the sources available. Callosum’s own website still identifies its announced funding as $10.25 million.
What Is Callosum Building?
Callosum is developing software designed to orchestrate AI workloads across different types of chips and computing systems.
Today, much of the AI industry is built around large numbers of similar processors, particularly Nvidia GPUs.
Callosum believes that model will eventually become inefficient.
Instead, it wants AI systems to use different processors depending on the job.
These could include hardware from:
- Nvidia
- AMD
- Amazon
- Cerebras
- SambaNova
- Other emerging AI-chip companies
The company’s software acts as an orchestration layer that can decide how AI workloads should be distributed across this different hardware.
Why Different AI Chips Matter
AI computing is becoming more diverse.
Nvidia remains the dominant force in AI accelerators, but many companies are developing specialized processors designed for particular workloads.
Some chips focus on:
- Training
- AI inference
- Memory efficiency
- Low power consumption
- High-speed networking
- Specific AI models
- Edge computing
The problem is that these chips do not automatically work together efficiently.
A company could have Nvidia GPUs in one cloud, AMD accelerators in another and custom chips somewhere else.
Managing all of that infrastructure can become extremely complicated.
Callosum wants to solve that problem.
Callosum’s “Heterogeneous Intelligence” Approach
The company calls its approach “Heterogeneous Intelligence.”
The basic idea is simple:
Different problems may need different models and different chips.
Instead of forcing one AI model to handle everything on the same hardware, Callosum wants to build systems where multiple models and types of compute work together.
Callosum argues that this can create systems that are faster, cheaper and more capable than a system built around identical hardware.
The company describes itself as an “Intelligent Systems Company” rather than simply an AI software company or chip company.
The Startup’s Founders Come From Cambridge
Callosum was founded by Danyal Akarca and Jascha Achterberg, who met while doing PhD research at Cambridge.
Their background combines neuroscience, AI and computing.
The company’s founding idea is partly inspired by the human brain.
The brain does not use billions of identical computing units to solve every problem.
Instead, different types of neurons and circuits perform different functions and work together.
Callosum believes AI infrastructure could benefit from a similar idea.
Callosum Wants to Challenge Nvidia’s AI Dominance
The company is not simply trying to build another Nvidia competitor.
Its strategy is different.
Instead of saying:
“We have a better AI chip than Nvidia.”
Callosum’s approach is closer to:
“AI should be able to use Nvidia, AMD, Google and many other chips together.”
That could become increasingly valuable as companies look for alternatives to Nvidia and try to reduce the cost of AI computing.
Callosum’s platform is designed to work across multiple cloud providers and different types of hardware.
Why Investors Are Interested
The AI infrastructure market is becoming enormous.
Companies are spending billions of dollars on:
- GPUs
- AI accelerators
- Data centers
- Networking
- Storage
- Cooling
- Electricity
But simply buying more hardware isn’t always the best solution.
Companies also need to make the hardware they already have work more efficiently.
That creates an opportunity for software that can intelligently manage computing resources.
Callosum says its early work has demonstrated significant improvements in certain heterogeneous workloads, including claims of higher accuracy, faster performance and lower costs compared with homogeneous systems.
These claims will need to be proven across larger real-world deployments, but they explain why the company has attracted investor attention.
The UK Government Is Also Backing Callosum
Callosum has received support from the UK government.
In April 2026, the UK’s Sovereign AI Fund announced Callosum as its first equity investment.
The government sees the startup as part of a broader effort to build strategic AI infrastructure in the UK.
Callosum has also received research support from ARIA, the UK’s Advanced Research and Invention Agency.
The company announced a $2.9 million ARIA grant to research a co-located heterogeneous compute cluster designed for multi-agent AI workloads.
This government support gives Callosum an important advantage as it develops its technology.
The AI Industry Is Moving Toward More Specialized Hardware
The timing of Callosum’s fundraising effort is important.
The AI industry is moving away from the idea that one processor architecture will dominate every workload.
New types of AI hardware are appearing, including:
- Custom inference chips
- Optical processors
- Neuromorphic hardware
- In-memory computing
- Specialized accelerators
- New networking architectures
That creates a problem:
How do you make all these different systems work together?
Callosum believes this could become a major infrastructure layer.
AI Inference Could Make This More Important
AI inference is becoming one of the biggest challenges for the industry.
Training a frontier AI model requires huge computing resources, but once the model is released, millions of users can interact with it every day.
AI agents could increase demand even further because they may perform dozens or hundreds of model calls to complete one task.
That means companies will need to optimize:
Cost + speed + power + accuracy
Using different chips for different parts of a workload could become one way to achieve that.
Callosum’s Bigger Vision
Callosum’s long-term vision goes beyond simply managing cloud GPUs.
The company says it wants to co-evolve chips and AI intelligence, allowing software and hardware to be designed around each other.
That could eventually create a new type of AI infrastructure where:
Models + chips + algorithms + data centers
are optimized together.
This is a much bigger ambition than building another software layer for existing cloud infrastructure.
What Could the $100M Be Used For?
If Callosum successfully closes a funding round of up to $100 million, the capital could significantly accelerate the company’s plans.
The company has previously said it wants to expand its team, grow into the U.S. and develop its own complementary hardware infrastructure.
The additional capital could help fund:
- Engineering
- AI research
- Hardware development
- Cloud infrastructure
- U.S. expansion
- Enterprise partnerships
- Large-scale deployments
It could also allow Callosum to move from research demonstrations toward production systems.
There Are Still Major Challenges
Callosum’s idea is ambitious, but heterogeneous computing is difficult.
Different chips have different:
- Programming models
- Memory systems
- Performance characteristics
- Software stacks
- Networking requirements
Moving workloads between them can introduce overhead.
Callosum therefore needs to prove that the benefits of using different hardware are greater than the complexity of connecting it.
It also needs to convince major AI companies and cloud providers to integrate its technology into production environments.
Why This Could Become a Big AI Infrastructure Story
The AI industry has spent years focused on scaling up.
More data.
More parameters.
More GPUs.
More data centers.
Callosum is betting on another direction:
Scale through diversity.
Instead of simply adding more identical chips, future AI systems could combine many different types of computing hardware.
If that happens, the software that coordinates those systems could become extremely valuable.
The Bigger Picture
Callosum’s reported effort to raise up to $100 million shows how investors are looking beyond AI models and individual chips toward the infrastructure connecting everything together.
The company’s previously announced $10.25 million seed round, UK Sovereign AI backing and ARIA support have given it a strong starting position.
But the $100 million figure should currently be described as a reported fundraising target, not as a completed financing round.
The larger idea behind Callosum is nevertheless significant.
As AI becomes more expensive and computing becomes more diverse, companies may no longer want to depend on a single chip architecture.
They may want AI systems that can choose the right model, the right processor and the right cloud environment for every task.
If Callosum can make that practical, it could become an important piece of the next generation of AI infrastructure.
The future of AI may not be built on one chip. It could be built on many chips working together—and Callosum wants to build the software that connects them.
By
