AI Champions Frontier AI Phase 1: Fundamental AI is the most capability-led theme in the competition. It is aimed at UK SMEs developing foundational AI advances in abstract reasoning and generalisation, novel AI and ML architectures and methodologies, online learning and world model learning systems, or generalisable explainable AI. As with the rest of the competition, Phase 1 is for UK-registered SMEs applying alone, with £150,000 to £250,000 project cost and 3 to 6 months duration.
This theme lines up closely with current UK policy. The AI Opportunities Action Plan: One Year On says government has already delivered 38 of 50 actions, while the UK is also using the Sovereign AI Unit to support homegrown frontier capability. The same update says the Unit has already started investing in data assets to support UK companies. Separately, the AI Security Institute says frontier AI capabilities are improving rapidly across the domains it has tested.
This is the theme for companies whose core value is the AI advance itself.
That could mean:
The official priorities are:
This is the theme for companies whose core value is the AI advance itself.
That could mean:
The official priorities are:
A good bid in this theme is usually narrow, technical, and measurable.
For instance, a strong application might focus on:
The competition does not require you to prove full market scale in Phase 1. It does require you to prove that the core approach works in practice and can translate into later scalable products or platforms. Innovate UK asks for proof of concept of the whole stack, the core architecture, or a critical subsystem, plus technical novelty, validation against clear metrics, and a scaling rationale.
The first mistake is calling something “foundational” when it is really just fine-tuning or packaging an existing model.
The second is chasing broad benchmark language without naming the actual uncertainty. Assessors need to know what you are trying to prove in Phase 1, not just what long-term vision you have.
The third is weak defensibility. In foundational AI, you need to tell a believable story about how the advance becomes hard to replicate. That may come from architecture, training method, data asset design, evaluation know-how, systems integration, or some combination of them.
The fourth is confusing research ambition with project fit. This competition is still a commercial innovation competition. The application must describe serviceable market and customers, background IP and route to defensibility, and the ability of the technology and business model to scale.
For this theme, the best applications usually have five ingredients:
That last point matters because only successful Phase 1 applicants may be invited into a later Phase 2 competition, which Innovate UK says is expected to support demonstrator projects of up to 12 months and up to £1 million, subject to separate launch and approval.
FAQs
