GrantRadar

Innovate UK (UKRI)

Efficient, Structured and Controllable AI Systems

Innovate UK competition for UK SMEs to experimentally validate novel, reusable AI or machine learning mechanisms (new learning, causal, modular or geometric/physics-structured architectures) that could underpin future generations of efficient, structured and controllable AI systems. Projects are short (1 to 3 months), start at TRL 1 to 3 and must have total eligible costs of £50,000 to £150,000; the grant covers up to 70% of costs for micro and small businesses and 60% for medium-sized ones. It is aimed at research-intensive AI start-ups, deep tech SMEs and university spinouts, sits under the Sovereign AI Unit's mandate, and is phase 1 of a possible two-phase competition. Innovate UK says similar competitions suggest about a 4% chance of success; it closes at 11am on 18 November 2026.

Apply on the funder’s site
You could get
£50,000 – £150,000
Deadline
31 days left
18 November
Effort
Heavy
budgets, references, often several stages

Who can apply

  • A UK registered micro, small or medium-sized enterprise (SME) applying alone; the competition is intended primarily for research-intensive AI start-ups, deep tech SMEs, university spinouts and pre-commercial companies developing proprietary AI technology
  • An organisation can lead only one application, subcontractors are not allowed, and the work must be carried out in the UK with results exploited in the UK and most of the funding spent in the UK
  • Projects must cost £50,000 to £150,000, last 1 to 3 months, start by 1 April 2027 (on the first of a month) and end by 30 June 2027
  • Funding is a subsidy under the Research, Development and Innovation Streamlined Subsidy Scheme; firms in financial difficulty, or that have failed to exploit a previous Innovate UK project, have an overdue accountant's report or breached previous award terms, will not be funded

What it funds

Feasibility studies and industrial research projects that experimentally validate a novel AI or ML mechanism, where the principal technical advance comes from one or more of:

  • A new learning or credit assignment mechanism
  • An adaptive or dynamically reconfigurable architecture
  • A learned causal or intervention grounded model
  • A graph-native or relational computational architecture
  • A compositional representation, skill or operator mechanism
  • A mechanism, model or architecture exerting measurable control over an agent
  • Projects must fall within at least one theme and one specific priority in it: Theme 1 resource and data adaptive learning (active and probabilistic learning; continual and selective learning; data efficient learning); Theme 2 causal and intervention grounded models (causal representation learning; interventional or counterfactual learning; causal discovery under partial observability); Theme 3 learned modular computation and routing; Theme 4 geometric, equivariant and physics structured learning
  • Deliverables: a prototype algorithm architecture or subsystem, an experimental and validation environment, results against named baselines or a hypothesis, ablation and negative control evidence, and a plan for further development

Projects must start at TRL 1, 2 or 3 and advance 0.5 to 1 TRL. A technical whitepaper is required at the end. Where an existing frontier model is used, the advance must remain technically material above the base model. A possible Phase 2 has no funding currently allocated.

What it will not fund

  • Projects without clear background IP ownership or rights
  • Projects without a specific defensibility route, or that do not result in defensible foreground IP
  • Projects that do not align with a competition theme and specific priority
  • Projects without a named baseline with metrics and numeric targets, or without a falsifiable, hypothesis-driven experimentation plan
  • Primarily literature reviews or requirement gathering without substantive experimental R&D
  • Routine integration, deployment, orchestration or productisation of existing AI tools or third-party models without novel AI or ML development
  • Projects focused primarily on non-AI or non-ML R&D, including cross-cutting projects
  • Prompt engineering, generic agent orchestration, routine Graph Neural Network (GNN) application and passive-only agentic layers
  • Projects without measurable, specific objectives or clear technical novelty and feasibility challenge
  • Subcontractors (not allowed in this competition)
  • Previously submitted applications
  • Subsidies dependent on export performance or on use of domestic inputs

How to apply

  1. Read Innovate UK's guidance on applying for a competition on the Innovation Funding Service, and watch the briefing recording (16 October 2026) and slides
  2. Start an application on the Innovation Funding Service (via 'Start new application' on Find a Grant or the competition page)
  3. Complete project details: title, start date and duration, research category, project summary (400 words), public description (400 words) and a scope answer (800 words) covering technical novelty and hypothesis, the technical asset, how it could underpin future products, and TRL justification
  4. Answer the application questions: unscored questions on theme, location, animal testing, permits, international collaboration, export licence and Trusted Research, then scored questions on IP and defensibility, need and technical development, approach, team, market, wider impacts, project management, risks, added value and costs
  5. Complete the finances and Project Impact sections
  6. Submit by 11am UK time on 18 November 2026; three independent assessors score the application

Applications out of scope are not sent for assessment. Contact support@iuk.ukri.org or 0300 321 4357 at least 15 working days before the deadline if you need reasonable adjustments.

Have ready

  • Optional appendix to the need and technical development question (PDF, up to two A4 pages)
  • Optional appendix to the approach and innovation question (PDF, up to two A4 pages)
  • Appendix of one-page CVs for each known member of the project team (the question is capped at 5 points without them)
  • Project plan or Gantt chart (PDF, up to two A4 pages)
  • Risk register (PDF, up to two A4 pages)

Appendices must be PDFs of no more than 10MB, legible at 100% zoom. Answers must not contain URLs.

More like this

Other grants open to organisations like yours.