Callosum
Callosum is a London-based AI infrastructure startup that orchestrates heterogeneous AI models across diverse chip architectures (NVIDIA, AMD, AWS Trainium/Inferentia2, Cerebras, SambaNova, TPUs, and next-gen silicon), targeting enterprise AI teams and chip makers seeking cost and latency improvements.
- Company typePrivate
- Founded2026
- HeadquartersLondon, United Kingdom
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What Callosum does
Callosum Technologies Ltd is a London-headquartered AI infrastructure company that builds an orchestration framework for running heterogeneous AI models across heterogeneous silicon — including AWS Trainium and Inferentia2, Cerebras, SambaNova, NVIDIA accelerators, AMD, Google TPUs, and a growing set of next-generation compute substrates (Normal Computing physics-based ASICs, Mixx silicon photonics, Cortical Labs biological computing, Great Sky superconducting optoelectronic networks). The company's vertically integrated stack is organised so that workflows are aware of hardware, models are aware of the task graph, and kernels are aware of output constraints, with each layer co-optimised against the others. Core techniques include Heterogeneous Recursion (decomposing recursive language model workflows across diverse model–chip pairings), a Heterogeneous Vision-Language-Action system (benchmark SOTA on VisualWebArena at 66%), Topology-Aware cache management using Furthest Future Use eviction (up to 2.4x faster than LRU in vLLM/SGLang), and an on-die NKI kernel on AWS Inferentia2 that performs constrained JSON grammar decoding 1,767x faster than CPU masking at batch 64.
The company targets two primary customer segments: enterprise AI infrastructure teams running heterogeneous workloads at scale who need to escape single-vendor accelerator lock-in and improve the cost-latency-quality Pareto frontier, and next-generation chip makers whose novel silicon needs a software integration path into production AI stacks. A secondary channel is the UK Government via the £500M Sovereign AI Fund, which made Callosum its inaugural equity investment in April 2026. Pricing is not publicly disclosed and appears quote-based; commercial relationships to date include production deployments (Coworker AI), bespoke hardware co-engineering (AWS Inferentia2 custom kernels, four neo-silicon partners), and grant-funded research (ARIA $2.9M, ARIA Scaling Inference Lab). The company emerged from stealth in February 2026 alongside a $10.25M pre-seed led by Plural, and as of the latest available data has a London team of approximately 14 named members with 11 open engineering roles being actively hired.
Callosum firmographics
Firmographics- Name
- Callosum
- Legal name
- Callosum Technologies Ltd
- Website
- https://callosum.com
- Company type
- Private
- Founded year
- 2026
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Callosum is a London-based AI infrastructure startup that orchestrates heterogeneous AI models across diverse chip architectures (NVIDIA, AMD, AWS Trainium/Inferentia2, Cerebras, SambaNova, TPUs, and next-gen silicon), targeting enterprise AI teams and chip makers seeking cost and latency improvements.
- Ownership category
- akta.pro rank
Callosum industry classification
Industry- Product category
- AI Compute Orchestration Infrastructure
- NAICS
- Computer Systems Design and Related Services (54151), Computer Systems Design and Related Services (5415), Computer Systems Design Services (541512), Custom Computer Programming Services (541511)
- SIC
- Services-Computer Integrated Systems Design (7373), Services-Computer Programming Services (7371), Services-Computer Programming, Data Processing, Etc. (7370)
- akta.pro primary industry
- AI Integration & Orchestration Platforms (Connectors, Workflow, iPaaS for AI) (HDAEANAI)
Keywords
Where Callosum is headquartered
LocationHeadquarters
- HQ city
- London
- HQ country
- United Kingdom
- HQ region
- Europe
Offices1 record
Markets served
Callosum business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Personnel, Technology or R&D, Operations, Marketing or Sales, Infrastructure
Revenue model
- Enterprise orchestration deployments: Pre-revenue / commercial model not explicitly disclosed. Callosum partners with named enterprises (e.g., Coworker AI) to deploy its heterogeneous intelligence orchestration infrastructure in production AI workflows; results from one such partnership are published on the company's technology blog, suggesting enterprise deployment and managed infrastructure relationships rather than a self-serve product.
- Bespoke hardware integrations: Capital from the seed round is allocated to "bespoke hardware integrations" with chip makers, indicating commercial co-development with silicon partners (e.g., AWS Inferentia2 custom NKI kernels, partnerships with Normal Computing, Mixx, Cortical Labs, Great Sky).
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Other | — | Quote-based / Not publicly disclosed |
Go-to-market motion3 records
Distribution channels4 records
Marketing channels8 records
Callosum product offering
Product offeringCore offering
Callosum builds a vertically integrated Intelligent Systems orchestration platform that coordinates heterogeneous AI models across heterogeneous silicon (NVIDIA, AMD, Google TPUs, AWS Trainium/Inferentia2, Cerebras, SambaNova, and next-generation physics-based, photonic, biological, and superconducting compute) to deliver orders-of-magnitude improvements in cost, latency, and quality for enterprise AI inference workloads. The company operates at the systems level rather than at individual model level, co-evolving workflows, models, kernels, and silicon into production-grade infrastructure for enterprise AI teams, chip makers, and sovereign-AI programmes.
Differentiator
Problem solved
Functional benefit
Products and services
- Heterogeneous Intelligence Platform Vertically integrated Intelligent Systems orchestration platform that operates at the systems level (not individual models) to coordinate heterogeneous AI models across heterogeneous silicon (NVIDIA GPUs, AMD, Google TPUs, AWS Trainium/Inferentia2, Cerebras, SambaNova, and next-generation physics-based, photonic, biological, and superconducting compute) for enterprise AI inference workloads. Delivered to enterprise AI teams via cloud-agnostic deployment across any provider exposing an access endpoint.
- Enterprise Orchestration Deployment (Coworker AI production integration)
Quantifiable outcome
- Up to 12x lower cost and 5.5x faster than recursive GPT-5 (Cerebras Llama-70B on OOLONG)
- +6 more outcomes
Companies that use Callosum
Customer profileNamed customers1 record
Segments4 records
Ideal customer profiles4 records
Callosum technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration19 records
AI capability11 records
Feature8 records
Callosum partnerships and signals
Strategic signalPartnerships
Eight partnerships are on record, tiered minor, flagship and core.
- Oriole NetworksminorPhotonics-based AI infrastructure partner in advanced talks to join the ARIA Scaling Inference Lab alongside Callosum, CommonAI CIC, and Sequrity.AI.
- AWS (Amazon Web Services)flagshipCo-development partner on AWS Inferentia2 silicon: Callosum built a custom NKI kernel that performs constrained grammar decoding directly on NeuronCore SBUF, enabling O(1) tool-calling grammar enforcement (1,767x faster than CPU masking at batch 64). The work is being deployed to enterprise customers.
- Normal ComputingcoreStrategic partnership with a physics-based ASIC company targeting diffusion-based generative AI workloads (image and video generation). Callosum is co-evolving the silicon into its orchestration stack for orders-of-magnitude energy efficiency and lower latency.
- MixxcoreStrategic partnership with a silicon photonic interconnect company integrating optics directly with ASICs. Targets switchless clusters with dramatically lower power and latency.
- Cortical LabscoreStrategic partnership with a biological computing company that fuses lab-grown human neurons with silicon to create adaptive networks learning from minimal data at a fraction of the energy cost.
- Great SkycoreStrategic partnership with a superconducting optoelectronic networking company combining semiconductors, superconductors, and photonics — approaching physical limits of neural computation.
- Coworker AIcoreEnterprise integration / customer deployment: Coworker AI runs Callosum's heterogeneous recursion on its autonomous-agent platform that handles millions of complex enterprise workflows. A public benchmark on GitHub activity-log summarisation shows Callosum infrastructure beats single-call Claude Opus 4.5 baselines across 30k, 88k, and 200k context lengths.
- CommonAIcoreCo-development partner on the ARIA-funded co-located heterogeneous compute cluster for multi-agent intelligence. CommonAI CIC also leads delivery of ARIA's Scaling Inference Lab in which Callosum is the inference orchestration partner.
Scale indicators12 records
Recent moves6 records
Expansion highlights6 records
Callosum competitors and assessment
Company assessmentDirect peers
- Anyscale: Anyscale (built on Ray) provides distributed compute orchestration for AI workloads, including inference and training across heterogeneous clusters. It is comparable to Callosum in providing a developer-facing platform that abstracts heterogeneous infrastructure for AI teams.
- Replicate: Replicate runs a cloud platform that hosts and serves open-source AI models with a developer API. It is comparable to Callosum as a third-party inference and deployment layer that abstracts hardware differences for AI application builders.
- Together AI: Together AI operates a cloud platform for open-source and custom AI model inference and training, with optimization across multiple accelerator types. It competes with Callosum in the AI inference platform category, particularly for cost-optimized inference on heterogeneous hardware.
- Modular AI: Modular AI builds the MAX platform, a unified inference and serving stack designed to optimize AI model execution across heterogeneous hardware. It is the closest direct competitor to Callosum in vertically integrated inference orchestration and heterogeneous compute routing.
- Fireworks AI: Fireworks AI provides a low-latency AI inference platform with model-serving infrastructure optimized across GPU hardware. It is a direct comparable to Callosum in the AI inference orchestration and serving category targeting enterprise AI teams.
Broad incumbents
- Groq: Groq designs its own LPU silicon and offers a hosted inference cloud serving open-source models at low latency. It is comparable to Callosum as a chip-plus-inference stack, but operates a single proprietary architecture rather than orchestrating across heterogeneous silicon.
- SambaNova Systems: SambaNova builds RDU accelerators and offers enterprise AI inference and fine-tuning stacks. Like Cerebras, it is both a Callosum partner and a comparable inference provider, but with a vertically integrated single-chip product strategy.
- Cerebras Systems: Cerebras builds wafer-scale AI accelerators and offers cloud inference services. It is a Callosum integration partner but also a comparable inference provider, owning its own silicon and serving customers directly rather than orchestrating across chips.
- Hugging Face: Hugging Face is a broad AI/ML platform spanning model hub, datasets, training, and inference endpoints (including Inference Endpoints and dedicated hardware integrations). It overlaps with Callosum as a layer that mediates between AI developers and underlying compute, but as part of a much wider portfolio spanning the entire model lifecycle.
Others
- Lambda Labs: Lambda Labs provides GPU cloud infrastructure and clusters for AI training and inference, primarily on NVIDIA hardware. It is adjacent to Callosum as a compute provider that AI infrastructure teams use, though it does not offer heterogeneous orchestration itself.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
Callosum social profiles
Digital presenceCallosum compliance and trust
Trust signalCompliance1 record
Callosum financial estimates
Financial estimateRevenue estimate
Valuation estimate
Callosum leadership team
Management profileNumber of profiles
Profiles5 records
Callosum funding detail
Funding detailFunding overview
Funding rounds3 records
Investors5 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Callosum M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Callosum
What does Callosum do?
Callosum builds a vertically integrated Intelligent Systems orchestration platform that coordinates heterogeneous AI models across heterogeneous silicon (NVIDIA, AMD, Google TPUs, AWS Trainium/Inferentia2, Cerebras, SambaNova, and next-generation physics-based, photonic, biological, and superconducting compute) to deliver orders-of-magnitude improvements in cost, latency, and quality for enterprise AI inference workloads. The company operates at the systems level rather than at individual model level, co-evolving workflows, models, kernels, and silicon into production-grade infrastructure for enterprise AI teams, chip makers, and sovereign-AI programmes.
Is Callosum a public or private company?
Callosum is a private company. It is classified as venture growth investor backed and is currently operating.
When was Callosum founded?
Callosum was founded in 2026. It employs 11 to 50 people.
Where is Callosum based?
Callosum is headquartered in London, United Kingdom, in the Europe region.
How does Callosum make money?
Two revenue lines are on record. Enterprise orchestration deployments are the primary driver. The others are bespoke hardware integrations.
Who are Callosum's main competitors?
Direct peers on record are Anyscale, Replicate, Together AI, Modular AI and Fireworks AI. Broad incumbents are Groq, SambaNova Systems, Cerebras Systems and Hugging Face. Lambda Labs is listed as an others.
Does Callosum have an API?
Yes. Callosum is hiring an "Inference Engineering - API Platform Architect" and a "Platform Engineer" for an inference API platform, indicating it exposes an API surface for orchestration of AI inference across heterogeneous hardware. The platform exposes endpoints for any provider that can be integrated into the orchestration layer and deployed alongside hyperscaler clouds, enabling enterprise partners to route AI workloads across diverse chip architectures.
What industry is Callosum in?
Callosum's product category is AI Compute Orchestration Infrastructure. Its primary akta.pro industry code is HDAEANAI, AI Integration & Orchestration Platforms (Connectors, Workflow, iPaaS for AI). Its NAICS code is 54151 and its SIC code is 7373.