Entrants to this competition must be:

• Established businesses, academic institutions, start-ups, SMEs, or individual entrepreneurs
• UK based or have the intention to set up a UK base

Your project must:

• Have a grant funding request up to £60,000 (eligible costs only, see below).
• Start 1st March 2027 and complete by 31st May 2027
• Carry out its project work in the UK.
• Intend to exploit the results from or in the UK.

The solution should provide a series of prototype Edge AI compute boards spanning a range of computational capability, power consumption, physical size and cost.

The purpose of the board series is to allow Tekh to deploy and evaluate different AI algorithms across progressively constrained hardware, enabling us to determine the minimum level of compute required to deliver acceptable inference performance for a given task.

The solution should therefore:

• Provide multiple board variants rather than a single fixed design
• Offer a deliberate progression in compute capability across the board series
• Enable deployment of different AI models and inference workloads across the range
• Explore different compute architectures where appropriate, recognising that dedicated GPU or AI acceleration may not be required for all workloads
• Avoid assuming that greater computational capability or dedicated AI acceleration necessarily represents a better solution
• Allow meaningful comparison of inference performance, power consumption, physical footprint and cost between board variants
• Use a consistent software and deployment approach wherever practical to support comparison between hardware configurations
• Support repeatable experimentation and benchmarking
• Be suitable for integration into constrained edge platforms
• Be capable of further iteration as testing identifies useful performance thresholds or gaps between board variants

The board series should enable Tekh to determine not only how little computational capability is required, but also what type of compute is appropriate for different classes of AI workload.

The solution should:

• Use only readily available commodity components, with consideration given to component cost, availability, supply chain resilience and potential for manufacture at scale
• Provide a range of computational capabilities across the proposed board series, enabling progressively more constrained AI workloads to be investigated
• Include variation in compute architecture across the board family where appropriate. This may include CPU only configurations as well as boards incorporating GPUs, NPUs, FPGAs or other specialist acceleration
• Include sufficient variation to allow Tekh to understand when dedicated acceleration becomes necessary, and when a lower capability architecture can still provide acceptable inference performance
• Support the deployment and execution of AI inference workloads supplied by Tekh
• Provide sufficient memory and storage for the intended workload of each board variant, while avoiding unnecessary capacity that increases cost, size, weight or power consumption
• Include power monitoring, or an equivalent integrated measurement capability, that allows power consumption to be recorded during representative AI workloads
• Where practical, support measurement of both overall board power consumption and major compute subsystems to allow meaningful comparison between different architectures and workloads
• Enable power measurements to be captured alongside inference performance, allowing performance per watt and energy consumption per inference to be evaluated
• Provide appropriate interfaces for integration with sensors, communications systems and other edge platform components
• Support repeatable benchmarking across the board series
• Where practical, maintain sufficient commonality between board variants to simplify software deployment, reduce development cost and allow meaningful comparison of results
• Make pragmatic use of common board designs, interchangeable components, compute modules or other approaches where these allow a useful range of experimental configurations to be produced within the available development budget
• Avoid unnecessary components, interfaces or capabilities that increase size, weight, power consumption or cost without contributing to the intended workload
• Be designed with future manufacture, reproducibility and scaling in mind rather than solely as a laboratory prototype

The objective is not to identify a single preferred processor architecture. The board series should allow Tekh to experimentally determine the relationship between AI workload, compute architecture, inference performance, size, weight, power consumption and cost.

The solution should:

• Be suitable for operation as an Edge AI compute platform without reliance on cloud based processing or continuous external connectivity
• Be capable of operating from standard, readily available power supplies using standard power connectors
• Be designed with future battery powered operation in mind, including consideration of supply voltage, power efficiency and power management
• Avoid requiring unusual or specialist external power infrastructure for normal operation and testing
• Support operation under sustained AI inference workloads for sufficient periods to allow meaningful performance and power characterisation
• Allow power consumption, temperature and other relevant operating characteristics to be monitored during testing
• Be sufficiently compact and lightweight to support integration and experimentation on representative small edge platforms
• Be robust enough for laboratory testing and representative field trials

Full environmental or military qualification is not required as part of this challenge. However, the design should take account of the potential for future operation in constrained environments where size, weight, available power, thermal management and connectivity may be limited.

The ultimate aim is to enable deployment on battery powered edge platforms. While battery integration itself is not required within this challenge, the proposed hardware should not introduce design choices that would unnecessarily constrain future battery powered use.

The following are outside the scope of this challenge:

• Development of a single high performance AI compute board without a range of lower capability alternatives for comparison
• Development of AI models or algorithms. Tekh will provide representative AI workloads for deployment and evaluation
• Solutions that require cloud based processing or continuous external connectivity to perform AI inference
• Development of batteries, battery management systems or bespoke power supplies. Future battery operation should be considered within the board design, but battery development is not required
• Full military, environmental or regulatory qualification of the prototype hardware
• Development of complete autonomous platforms, vehicles, sensor systems or communications systems
• Optimisation for a single application or AI workload at the expense of the wider experimental purpose of the board series
• Solutions dependent on highly specialised or difficult to source components where this would prevent realistic future manufacture or undermine the low cost objective
• Production scale manufacture during the challenge

The challenge is focused on developing a practical series of experimental Edge AI compute boards that allows the relationship between AI performance, compute architecture, size, weight, power and cost to be investigated.

The successful applicant will have the opportunity to work directly with Tekh to develop and manufacture a new family of low SWaP C Edge AI compute boards.

The resulting boards will be evaluated by Tekh using representative AI workloads to establish their performance across different compute architectures and capability levels. Successful designs may provide the basis for continued collaboration, further development and future procurement as Tekh develops and deploys Edge AI solutions.

The initial opportunity is driven by Defence applications where size, weight, power and cost can significantly constrain the deployment of AI at the edge. However, the underlying requirement is not Defence specific. Affordable, power efficient Edge AI hardware has potential application across sectors including agriculture, manufacturing, infrastructure, robotics and remote sensing.

The longer term opportunity is to establish a scalable hardware platform and supply relationship supporting multiple Edge AI applications, rather than development of a single board for a single use case.

• Challenge launch: 5 October 2026
• Deadline for applications: Friday 6 November 2026
• Review of applicants: Monday 9 – Friday 13 November 2026
• Selection and notification of finalists: Friday 13 – Tuesday 17 November 2026
• Pitch Day: Wednesday 25 November – Wednesday 3 December 2026
• Confirm selected solution provider: By Friday 4 December 2026
• Submission to Innovation Funding Service (IFS): December 2026
• Project length: 3 months. (Starting by 1 March 2027 and complete no later than 31 May 2027)