AI is being sold as the engine of the next economy, but the way we build it will decide whether it strengthens communities, protects sensitive data and supports climate goals – or if it simply shifts more power, energy demand and risk into fewer hands.
In Swansea, Cory Hughes of Lantela and Liam Betsworth of Pendra got together to discuss that question from different perspectives. Cory is focused on helping organisations turn sustainability commitments into embedded operational choices. Liam is building sovereign AI infrastructure for teams that need control over sensitive data.
In this conversation the two Swansea-based founders talked about closing the gap between ambition and delivery, right-sizing AI, data sovereignty, and why Wales is a good place to build.
Here are the key takeouts – but do read on for a transcript of the full conversation.
- Designed, not declared. Cory says organisations have plenty of sustainability intent. What’s missing is a way of turning commitments into the day-to-day decisions teams make.
- AI both helps and hurts. A top-tier AI server can cost £500,000 and draw 14 kilowatts. But AI can also speed up the search for more sustainable alternatives.
- Right-size the model. Liam argues that most tasks don’t need the biggest model. Open-source models that run on a laptop now match the flagship models of 12 to 18 months ago.
- Sovereignty is about more than where data sits. Under the US CLOUD Act, US providers can be required to hand over data even when it’s stored in the UK.
- Show, don’t tell. Both founders win people round with evidence: stress-testing a strategy, or benchmarking a smaller model against the state of the art.
- Test before you plan. Their advice to founders is to test the problem and the solution quickly rather than write a full plan first.
Meet the founders
Who are you, and what does your company do?
Cory: I’m Cory Hughes, Managing Director at Lantela. We want a world where the future is designed, not declared: where the sustainability commitments organisations make are the ones they actually deliver. If you have a net zero target or a social value promise, how does it get embedded into the daily decisions services make, rather than sitting in a strategy document on a shelf?
Liam: I’m Liam Betsworth, CEO at Pendra. Pendra is a sovereign AI infrastructure company for regulated industries. Everyone’s using AI right now, but people who handle sensitive data, like financial records or personal information, worry about where that data goes. Is it being used for training? Is it being retained? Pendra offers a credible UK alternative for processing that data.
How did you come to build it, and when did you start to believe in it?
Liam: I got the idea from a problem I saw at my previous company, Amplify. We worked with large regulated customers like banks and pharmaceutical companies, and they kept asking: are you training on our data? Where’s it going? Is it kept in the UK? When we handled the data ourselves, that was easy to answer. When we used providers like OpenAI or Anthropic, people got less comfortable, because those services can feel like a black box.
I really started to believe in it after my first customer conversations. Companies told me their own sales were being blocked because they didn’t have a solution like this. I kept seeing the same pattern. This isn’t a nice-to-have; it’s a genuine need.
Cory: The story goes back to 2016. I was on stage at a conference, talking to digital transformation leaders about the next ten years, and one of my predictions was that climate change would blindside us. Climate rarely came up in rooms about technology, but I knew technology could both help and hinder our wider ambitions.
For the last four years I’ve been back in Wales. Lantela isn’t a startup from scratch. It’s powered by Perago, the digital transformation consultancy it sits within. Across NHS trusts, local government and the wider public sector, we’ve seen no lack of intent. Commitments are made, targets set and boards sign off. What’s consistently missing is how that ambition becomes the day-to-day decisions teams make.
In design, a recurring pattern tells you something can be redesigned. This is a design failure, not a failure of motivation. My role as a digital innovation adviser to the UK Design Council, through its Design for Planet mission, gave me the language to name it and build a practice around closing the gap from ambition to delivery.
Does AI help or hurt sustainability?
Liam: The cop-out answer is both. How it hurts is obvious. A top-tier AI server can cost £500,000, contain rare earth minerals and draw 14 kilowatts. That’s like fifteen microwave ovens running constantly. Then there’s water use. But AI can also help us find more sustainable alternatives and get tasks done with less energy. Paradoxically, the fact that AI hurts sustainability is also pushing sustainability up the agenda.
Cory: In 2022 we delivered a project for the Welsh Government called Tech Net Zero. It asked whether digital technologies help or hinder the race to net zero, and the research found it’s both. The Sustainable UX Network has a canvas for the responsible use of AI, and its test cases found that every positive impact came with a linked negative consequence.
That makes it a design choice, not a given. The net effect depends on whether an AI use case is designed against a real, verified need, or picked by default because it’s the tool to hand. As UNEP puts it, we need to make sure AI’s net effect on the planet is positive before we deploy it at scale.
How can we use AI more sustainably?
Liam: Right-size it. If you ask a top-of-the-range model to summarise an email you could read in 30 seconds, you’re using a £500,000 machine drawing 14 kilowatts. Most tasks don’t need that. Open-source models are now much smaller and can run on a MacBook, and they match the flagship models of 12 to 18 months ago, like GPT-4. A lot of AI is massively overpowered for what we’re asking of it.
Pendra lets you run any model, from the largest state-of-the-art systems down to models on your own laptop. You choose the right size for the task, and ultimately use less energy.
Cory: It’s exciting to see right-sized AI being built in Wales. We need case studies like this so we can show the organisations we work with that there’s an alternative to going straight for the big models.
Sovereignty is a sustainability question too, because it’s also about where the compute lives. One stat I read suggests AI infrastructure could soon use six times more water than Denmark. Sovereignty and sustainability belong in the same room.
There’s also a healthy tension between adoption and scepticism. Design Council research found that 91% of designers use AI, but 80% still prefer human judgement. The question I’m fond of asking at the start of any transformation is: is this even something we want to build? Is it a genuine AI use case, or just habit?
What does doing AI “the right way” mean?
Liam: We’ve given too much control to the handful of companies that own the models, the compute and the data. There are very few credible options that let you choose whether your data is trained on, where it goes, or whether it runs on renewable energy. Doing it the right way means giving people those choices.
Where people get lost, and what wins them round
What’s the objection you hear again and again?
Cory: To be blunt, the language of climate activism can lose people, especially in highly regulated organisations juggling long-term strategy against short policy cycles. Leading with “design” or “sustainability” often doesn’t match the operational reality service owners live in. Others tell us they already have an approved strategy and roadmap, so why intervene now? People aren’t resistant; they’re frustrated. They can see what needs to change but lack a practical way to get there.
Liam: I hear two. First: Microsoft, Amazon and others already offer UK regions and sovereignty packages, so how are you different? Second: people are used to the largest models and assume nothing else will do.
What finally flips them?
Cory: Showing, not telling. Lantela isn’t there to replace reporting or strategy. It’s there to stress-test the strategy and quickly show the impacts and unintended consequences of decisions. In our diagnostic workshops we might take a 2040 target and ask how it shapes the weekly conversation about capital allocation. Then we test it in small increments, without asking for a big upfront commitment.
Liam: For the model objection, we benchmark. I can run document processing or extraction and show a smaller model is right 99.5% of the time at 10% of the cost of one that’s right 99.8% of the time. Is ten times the money worth it?
For the hyperscaler objection, I explain that sovereignty is about more than where the data lives. US providers are subject to US law, including the CLOUD Act. It lets the US government legally request data from any US provider, even if that data sits in a UK data centre. This isn’t theoretical. It happened recently when the Dutch government blocked the sale of a company and the US subpoenaed data through Microsoft, including meeting notes and personal information. Add US export controls, and the question becomes: do you want your technology governed by another country’s laws? That argument is winning people round, including government teams that are already Microsoft customers.
Why projects stall
Most organisations aren’t short on ambition. So why do sustainability projects still stall?
Cory: Lately, many organisations have pushed back their ESG targets and called it pragmatic. My question is: who’s designing the pragmatism?
Early wins, like retrofits or switching from print to digital, are things an in-house team can control. It gets harder with something like the supply chain, where responsibility passes from the sustainability team to procurement, finance, operations and product. Those teams don’t naturally speak the language of sustainability. If no one has designed how the vision translates into their work, things stall, and the “pragmatic” declaration turns out to be anything but. That’s why we focus on translating between functions, so everyone around the table understands the commitment and can make it real.
And why do AI projects stall?
Liam: There’s been a gold rush. Lots of organisations rushed to adopt AI, expecting 10x or 100x gains, and many of those projects have stalled and failed. You can’t succeed just by buying AI and hoping it solves everything. You need a clear idea of how it will solve your specific problems, rather than buying a Copilot or Claude licence because everyone else has.
That’s why companies like Microsoft and Anthropic are investing heavily in forward-deployed engineers: experts embedded in an organisation to build something effective alongside its own subject matter experts. It’s about building relationships so people actually succeed with AI, rather than buying the hype and leaving the subscription unused.
Building in Swansea, and advice for what comes next
What’s brilliant about building a business in Swansea, and what’s not so great?
Cory: Wales didn’t wait for the AI boom. It had already legislated for long-term thinking, so building for the long term is a legal default, not a value-add. It’s also a small ecosystem where people know each other and can test ideas quickly. The downside is brain drain. I left Wales young and didn’t come back to work in my discipline until my mid-to-late thirties. We need to make the Welsh and Swansea economies vibrant, resilient and future-proofed, and push back on the question, “Is this as credible as a London firm?”
Liam: For me it’s the quality of life. Friends in London and Manchester struggle with housing costs, while you can have a lovely home in Swansea for the same money. With remote working, there’s no reason you can’t compete with businesses anywhere in the UK. The pain is distance and credibility: people still ask why you haven’t moved to London, where it’d be easier to raise money. The answer is some big winners in Wales that people can point to. I’ve made that my life’s mission. I’m staying in Wales to build interesting tech companies and show it’s possible.
Someone in Swansea has a business idea but hasn’t acted on it. What do they need to hear?
Liam: Just do it. Tech is moving so fast it feels like the next industrial revolution, and there’s never been more you can do with less. If you’ve got an interesting idea, jump on it.
Cory: The idea stage feels vulnerable, and my instinct is to write a full, robust plan. But what an idea needs early on is quick, testable evidence. You don’t need the full plan to make progress. Find the part you can test without spending a huge amount of money, emotion or energy, and start there.
Liam: You can test two things: that the problem is valuable, and that your solution is valuable. Test the problem by talking to people who have it and asking whether they’d pay for a fix. Then, with AI, you can mock up a solution quickly and get feedback. A valuable problem plus a valuable solution means you’re on to a winner.
Final question: what’s the one thing you’d tell a leadership team to do differently over the next year?
Cory: I’ll be a sustainability nerd here. There’s an idea from carbon markets being applied to AI governance: you don’t need perfect control to create meaningful incentives. You need one measurable unit, clear rules, and a system that makes doing the right thing cheaper than doing the wrong thing. So pick one decision already in flight, match it to one of your ESG commitments, and redesign it until you’ve proved it flows from strategy to operational delivery. Do that before the strategy document goes to your investors or operational teams.
Liam: Especially in regulated industries: don’t accept a status quo where you have to give up control of your data or of what’s most valuable in your business. Sovereign AI is taking off, and there are good opportunities emerging. Consider them seriously, rather than assuming you have to use the biggest providers and the biggest models.