When it unveiled the AI Opportunities Action Plan, the UK government promised £14bn in private investment and more than 13,000 new jobs, all with the goal of positioning the UK as a global AI superpower. But after nine months, it’s increasingly clear that the initiative suffers from a huge blind spot: infrastructure.
Most upskilling programmes focus on the two extremes on the AI scale. On the one end, educators are offering sophisticated training to data scientists to build cutting-edge LLMs. On the other, the average person is provided with very basic training on topics such as prompt engineering; they’re essentially being taught how to use proprietary, black-box generative AI tools. TechGrad’s £96.8m investment in AI scholarships, for instance, exemplify this approach. While there is value in these schemes, there is a huge upskilling challenge that remains unaddressed.
If the challenge is ignored, organisations will inevitably end up in what I call “pilotitis” – they’ll craft endless proof-of-concepts that never spread beyond the laptops they were built on. Every department of every company seems to have an AI pilot on the go, but few have AI that delivers services at scale.
The real UK AI skills gap
The government is not blind to this reality. In March, it completed the AI upskilling fund, which will help to fund AI training for SMEs. But if infrastructure skills are not included in such efforts, the UK risks producing many thousands of prompt engineers who lack the ability to turn ideas into working systems.
The consequences are real. The NHS, for instance, needs better technology deployment to ease waiting lists, while the courts system is crying out for trustworthy automation to clear its case backlog. These aren’t problems that will be solved by pilot schemes. They involve extremely complex deployment challenges that our current approach is not addressing.
AI engineering isn’t about just building models or writing prompts. It also involves creating the underlying infrastructure that turns experimental AI into powerful systems. Take the UK Compute Roadmap, published in July 2025, which commits up to £2bn to a sovereign public-compute ecosystem. With a significant expansion of the AI research resource by 2030, this is the kind of strategic investment the UK has long needed.
But building this infrastructure requires AI platform engineers: professionals who can deploy, manage, secure and scale out AI across environments where it’s most needed. Failing to train people to manage and orchestrate this infrastructure would be like building new airport terminals without air traffic control systems.
Building national AI infrastructure
Imagine if the UK had specialised models that were trained on national datasets and deployed efficiently across government and industry. Such systems would enable rapid experimentation, create the foundation for agentic AI and provide digital sovereignty while remaining secure and globally interoperable.
This is achievable if we stop treating AI as separate from core digital capability. Tomorrow’s civil servants won’t just prompt AI, they’ll orchestrate it across workflows. Our developers, administrators and senior leaders will all work side by side with AI. But none of this will be possible without strong foundational skills. The UK can continue down the current path of producing AI users at the two extremes: world-class researchers whose work never escapes the lab, and administrators whose AI interactions remain shallow. Or it can fit the missing piece by developing a core of AI-platform engineers who can turn pilot projects into transformational capabilities.
The global open-source community has already provided many of the technical foundations. What we need is the wisdom to build on those foundations and train the professionals who can turn today’s pilots into tomorrow’s public services. The government has the opportunity to embrace projects such as vLLM, an open-source library for AI, to ensure that the UK’s infrastructure investment is used as efficiently as possible. But still, public servants must be able to implement it effectively and securely.
To turn its AI ambitions into reality, the UK urgently need professionals who understand AI-ready platforms to build resilient, scalable systems that support diverse workloads. And those platforms must be secured across complex, multi-tenant environments, and run on open standards to ensure flexibility and choice across vendors.
These skills are the dividing line between organisations and nations that have AI ambitions and those that can achieve genuine AI capability.
Jonny Williams is chief digital adviser, UK public sector, at Red Hat.
When it unveiled the AI Opportunities Action Plan, the UK government promised £14bn in private investment and more than 13,000 new jobs, all with the goal of positioning the UK as a global AI superpower. But after nine months, it’s increasingly clear that the initiative suffers from a huge blind spot: infrastructure.
Most upskilling programmes focus on the two extremes on the AI scale. On the one end, educators are offering sophisticated training to data scientists to build cutting-edge LLMs. On the other, the average person is provided with very basic training on topics such as prompt engineering; they're essentially being taught how to use proprietary, black-box generative AI tools. TechGrad’s £96.8m investment in AI scholarships, for instance, exemplify this approach. While there is value in these schemes, there is a huge upskilling challenge that remains unaddressed.
If the challenge is ignored, organisations will inevitably end up in what I call “pilotitis” – they'll craft endless proof-of-concepts that never spread beyond the laptops they were built on. Every department of every company seems to have an AI pilot on the go, but few have AI that delivers services at scale.
