Britain's schools are adopting AI, but who controls the infrastructure behind it?

Four in five UK teachers now use artificial intelligence for work.

That is an extraordinary level of adoption for a technology that, just a few years ago, most people outside the technology industry had barely encountered. But there's another figure from the same research that I find considerably more interesting. Only 35% of teachers using AI say they're actually working fewer hours because of it. Those findings come from Bett's 2026 AI in Education research, based on a YouGov survey of 1,033 UK teachers. Weekly AI use has doubled in a year, yet the relationship between adoption and meaningful workload reduction remains much less convincing.

That should give us something to think about.

We've reached a point where AI is becoming a normal part of teaching, but we haven't necessarily established how to make it work effectively across an entire school. I think the reason goes beyond the capabilities of the technology itself. We're introducing increasingly sophisticated AI into schools that already operate across fragmented digital systems, inconsistent processes, and limited time for training.

At the same time, we rarely discuss the infrastructure underneath those AI tools. Who owns it? Where does the data go? Who controls the models? And how much control does a school really have over the technology it's beginning to depend on?

I think those questions are going to become increasingly important.

The problem we're actually trying to solve

I founded LESSO because of something I spent years watching happen at home. My wife, Amanda, is a classroom teacher and I've watched her spend evenings planning lessons, preparing resources, replying to messages, and trying to stay ahead of the following week.

I'm not a teacher, and I wouldn't pretend to understand every pressure that comes with standing in front of a classroom. But, I understand what happens when a working day doesn't really end when someone leaves their workplace. Amanda's experience is far from unusual.

According to the Department for Education, 38,600 full-time equivalent teachers left England's state-funded school sector in 2024/25. Education Support estimates that the wider annual cost associated with teacher attrition exceeds £1.5 billion. Those numbers represent a serious workforce challenge. They also raise questions about how effectively we organise the work surrounding teaching.

AI has enormous potential here.

It can help prepare lesson sequences, create classroom materials, draft communications and support assessment. Used appropriately, these capabilities could remove significant amounts of repetitive work. But saving ten minutes generating a worksheet is only useful if the teacher doesn't then spend those ten minutes correcting, reformatting or moving it between several systems.

Teachers already have enough tools to manage.

If AI simply adds another collection of disconnected applications, we risk improving individual tasks while leaving the overall workload largely unchanged. This is an argument I explore in our whitepaper, Teacher Retention, Workload and Governed AI. The opportunity is to change how work moves through a school, giving teachers more time for the professional judgement, relationships and classroom interactions that matter most. But making that possible requires something the education AI conversation often overlooks. Infrastructure.

What actually happens when a teacher uses AI?

Imagine a teacher uploading a knowledge organiser or curriculum plan into an AI platform. They ask it to produce a sequence of lessons, supporting slides, differentiated activities and assessment questions. On the surface, that looks like a straightforward interaction. Underneath it, several different systems may be involved.

The document could be stored in one environment. Relevant information might be retrieved through another service. A foundation model processes the request. Generated materials are saved, edited and potentially exported into other applications. Depending on how the product is designed, prompts, responses, logs and temporary data may be handled by additional providers. Every one of those stages introduces questions about security, privacy, ownership, and operational control:

• Where is the original material stored?
• Where does AI inference actually happen?
• Which companies operate the underlying infrastructure?
• What information is retained, and for how long?
• Can a school audit what has happened to its information?
• What happens if the underlying model provider changes its terms, increases prices or withdraws access?

These are practical questions about the systems schools are purchasing and relying upon. They become particularly important when AI moves beyond generic curriculum content and begins supporting real school workflows.

Assessment information, SEND-related records, safeguarding processes and communications with parents can involve sensitive personal information requiring appropriate protections. A school should understand how information flows through its technology before allowing those systems to handle increasingly sensitive work. That responsibility cannot simply be passed to individual teachers.

UK hosting and UK sovereignty are different things

Britain has become increasingly interested in sovereign AI. I agree with that direction, particularly for sectors handling sensitive information. But I think we need to be much more precise about what sovereignty actually means.

A service can store information in a British data centre while relying on a foreign-owned company to operate the infrastructure. A British software company might host its application domestically while sending AI requests to models operated in another jurisdiction. And a platform can use British infrastructure while still depending on foreign suppliers for critical components.

None of these arrangements is automatically inappropriate.

Some of the best AI technology comes from international companies. British businesses should be free to use it when doing so makes sense and the necessary protections are in place. The issue is whether organizations understand those dependencies and retain meaningful control over them. Data residency tells us where information is physically stored or processed. Legal jurisdiction determines which laws and disclosure obligations may apply. Infrastructure ownership tells us who ultimately controls the underlying environment. Model ownership and operational control determine which organizations can change access, pricing, capabilities and availability.

These distinctions matter because AI is becoming something organizations depend on operationally, rather than simply a service they occasionally use.

For a school, supplier resilience and continuity are particularly important. Teachers cannot reasonably rebuild their entire working environment whenever a technology provider changes direction. UK data protection law does not universally require every school workload to remain in Britain. Lawful international transfers are possible with the appropriate protections.

But when schools are choosing infrastructure for long-term AI adoption, greater transparency and control over data processing, jurisdiction and supplier dependency should carry real weight in procurement decisions. Sovereignty needs to be something an organization can demonstrate through its architecture and governance.

Schools need governance before deeper AI dependency

There's another part of the conversation that I think deserves more attention.

In the same Bett research, 48% of surveyed teachers said their school had no policy governing staff AI use, while 41% reported receiving no AI training or support from their school. That's concerning given how quickly adoption is progressing.

Schools need to establish which systems staff can use, what information can be processed, where human approval is required and how the impact of AI will be measured. The Department for Education's generative AI product safety standards already place clear expectations on suppliers around security, privacy, safeguarding, transparency and accountability. The DfE also recognises that teacher-facing AI currently offers more immediate benefits with fewer risks than direct pupil-facing use.

I think that sequencing is sensible.

We should give teachers the confidence, training and tools to benefit from AI before rushing towards widespread autonomous AI experiences for children. And school leaders need visibility over what is happening across their institutions. This means measuring much more than logins or the number of documents generated.

Are teachers spending fewer evenings preparing lessons? Are they moving between fewer tools? Do they trust the materials being produced? Can they identify where professional review is required? Is the overall working experience improving?

If schools cannot answer those questions, adoption statistics alone tell us very little about whether AI is delivering meaningful value.

Why we chose Civo for LESSO

These considerations shaped the decisions we made when building LESSO.

We're developing Teacher Workforce Infrastructure... A connected AI platform designed to help teachers prepare, review, and approve the work surrounding teaching, while giving school leaders greater visibility and control. The architecture needs to support multiple services, connected workflows, persistent school context and different levels of access and governance. It also needs to accommodate the sensitivity of the information that schools may eventually entrust to it.

When choosing our infrastructure provider, we wanted a foundation that supported those requirements from the beginning. That led us to Civo.

LESSO's production application environment runs on Civo's UK Kubernetes infrastructure, allowing us to operate and scale a multi-service platform without relying on a large internal infrastructure team. It also gives us a British infrastructure partner whose wider ambitions around sovereign cloud and AI compute align with our long-term thinking. This is one component of the wider sovereignty challenge. A UK-hosted application does not automatically make every connected AI model or external service sovereign.

We still need to evaluate the complete processing path, supplier dependencies, and safeguards as the platform evolves. Our immediate priority is proving that connected, governed AI workflows genuinely improve teachers' working lives. You can read more about the infrastructure decisions behind LESSO in Civo's case study.

Britain has an opportunity to build something more substantial

I think education presents an interesting opportunity for Britain's wider AI industry. Schools need technology that is safe, reliable, affordable and capable of delivering measurable benefits. They also need a level of accountability that many consumer AI applications were never originally designed to provide.

If British technology companies can build AI infrastructure and services that meet those requirements, the expertise developed could be valuable well beyond education. Healthcare, financial services, local government and other sensitive sectors face similar questions about data protection, procurement, operational resilience and control.

There is also a wider economic consideration.

Britain can become a substantial user of AI while remaining dependent on infrastructure, models and platforms controlled elsewhere. We should be building domestic capability alongside adoption. That requires investment in British infrastructure, accessible compute, technical skills and companies capable of developing useful products on top of it. It also requires customers who are prepared to evaluate British alternatives on their technical and commercial merits.

Sovereignty shouldn't mean isolating ourselves from international technology. It should mean having credible choices and the ability to maintain essential services without being permanently dependent on decisions taken outside our control.

The question Britain needs to answer

AI will almost certainly become a more significant part of how British schools operate. I can see why teachers are adopting it. The opportunity to remove repetitive preparation and administrative work is substantial. But the long-term success of AI in education will depend on more than the quality of its outputs.

It will depend on whether schools can govern its use, whether teachers genuinely benefit, whether sensitive information is handled responsibly and whether the infrastructure behind those services remains reliable and accountable. We should be asking these questions while schools still have the opportunity to make deliberate decisions about their AI infrastructure. Because once AI becomes embedded in the working week of thousands of teachers, changing the technology underneath it becomes considerably harder.

Britain has an opportunity to develop AI infrastructure that its schools can trust, that its companies can build upon and that its institutions can meaningfully control.

But that opportunity depends on thinking about who owns and operates the infrastructure before we've built an entire education system around it.