Innovate UK is offering £1.66 million to UK SMEs developing AI architectures, learning mechanisms, and controllable systems to develop and experimentally validate new AI technologies.
The competition, titled Efficient, Structured, and Controllable AI Systems, is being run by part of UK Research and Innovation (UKRI). It targets research-intensive startups, university spinouts, and pre-commercial companies developing proprietary AI technology that could bolster future products and services.
The programme excludes projects centred primarily on integrating existing AI products, routine deployment of third-party models, generic AI-agent orchestration or prompt engineering without substantive technical innovation.
Applicants must also demonstrate that their technology has a credible route to commercial defensibility, such as intellectual property, proprietary data advantages or specialised technical expertise.
Applications must establish a clear technical hypothesis, measurable validation criteria and evidence that the proposed research could produce a defensible technical asset.
Projects must be conducted in the UK, with most of the funding spent domestically, and each eligible organisation can lead only one application.
The competition opens on October 12, 2026, and closes on November 18, 2026.
What the funding will support
Individual projects must have eligible costs between £50,000 and £150,000 and run for between one and three months. Projects must start by April 1, 2027, and finish by June 30, 2027.
The funding is awarded as a grant, but it will not cover the entire cost of their projects. Micro and small businesses can receive up to 70% of eligible costs, while medium-sized businesses can receive up to 60%.
Applicants must demonstrate that their technology can advance beyond its current stage of development through measurable experimental results. Projects are expected to begin at Technology Readiness Level (TRL) 1, 2, or 3 and make progress of approximately half to one TRL level during delivery.
The required outputs include a prototype algorithm, architecture, or subsystem; an experimental validation environment; results measured against defined baselines or hypotheses; and a technical white paper documenting the findings and remaining limitations.
Four areas of AI research eligible for funding
The competition identifies four technical themes that applicants must align with. Proposals must select a specific priority within at least one theme.
1. Resource and data-adaptive learning
This category covers AI systems designed to learn more efficiently by reducing dependence on large datasets, repeated model training, or unnecessary computation.
Eligible research could involve systems that identify the most useful information to acquire, adapt to new knowledge without repeatedly retraining an entire model, or learn effectively when observations are limited or expensive.
2. Causal and intervention-grounded models
Innovate UK is also seeking approaches that help AI systems understand causal relationships and predict what may happen when conditions change.
Projects could explore causal representation learning, counterfactual prediction, or causal discovery under partial observability. The objective is to improve how systems generalise, make predictions, and support decisions beyond the circumstances represented in their original training data.
3. Learned modular computation and routing
This theme focuses on AI systems that can discover, select, and reuse computational components across different tasks.
Research may involve modular architectures, dynamic routing of computation, or reusable learned functions. Such approaches could enable AI systems to adapt how they process information rather than relying on a fixed computational pathway.
4. Geometric, equivariant, and physics-structured learning
The fourth theme supports learning systems that incorporate geometric, relational, symmetry-based, or physical structures.
Potential applications include AI models that reason more efficiently about structured data, maintain consistency under transformations, or incorporate physical constraints into their learning processes.
Eligible businesses can review the full requirements and apply through the official Efficient, Structured, and Controllable AI Systems funding competition portal.
