Donald and his Silicon Valley entourage are in town - and kerching - the US dollars are following them freely into the UK cash register. Promises of lucrative partnerships, a new AI Growth Zone in the North East, and a step-change in US-UK cooperation on AI, quantum computing, and big tech are all piling up. ChatGPT has been going into overdrive over the last 12 hours creating images of GPUs with union jack flags on, and all in all, it’s high time for the UK’s high tech!
The news about the North East’s AI Growth Zone (AIGZ) is welcomed and a sign we’re moving in the right direction. However it’s always important to look beyond the PR-spin and see this for what it is. The OpenAI – Nvidia – Nscale deployment in the North East will be circa 10MW initially, rising to maybe 40MW to power “in-time” (and I do love the ambiguity in that phrasing from the government’s press release) the predicted 31,000 GPUs.
Compare that to the AIGZ ambition of 500MW a site, and the 100,000 GPUs Nscale, a brilliant UK headquartered company - are deploying in Norway by the end of 2026 and its definitely the inference, rather than the deep training end of the AI compute pool. None the less, we’ve made a start to tackle one of the UK’s most frustrating conundrums, which as Jenson pointed out, is that 'the UK remains the world’s largest AI ecosystem without its own infrastructure'.
But 'infrastructure' is a broad word - GPUs, servers, data centres, connectivity—all essential, all part of the same AI mission. Yet one crucial element was missing from last night’s cascade of announcements: power.
Tech prosperity deals are only as good as the energy supply behind them. Zahra Bahrololoumi CBE, Salesforce’s dynamic CEO, spoke of “tapping into the UK’s legacy of global AI breakthroughs and expertise.” That’s fortunate - because tapping into the UK’s power grid is a far tougher proposition. And even if you can connect before you’ve retired, the price of it is going to sting somewhat compared to the US.
AI devours energy, and the UK’s power infrastructure is stuck in an antiquated state after decades of underinvestment. Energy here is more expensive than in any other G7 nation, and across 27 European countries. A year’s supply of 1GW costs £1.8 billion in the UK, compared with approximately £430 million in the US and parts of Europe. Even for companies with deep pockets, there is still the challenge of getting the power in the first place, and in the UK that isn’t straightforward. Grid connection times have ballooned from 18 months in 2019–20 to as long as 15 years in some regions. Offer dates are already creeping into the 2040s. Zoisa North Bond, CEO of Octopus Energy Generation, rightly calls it “a national scandal.”
These problems need to be addressed, and addressed quickly. Otherwise, as soon as the big investment announcements are passed out of the media limelight and onto delivery teams to make the headlines happen, they’re going to hit good, old fashioned British institutional failings, red-tape and systemic infrastructure obstacles. Despite the PR bluster, the very real risk is the UK’s AI superpower dream isn’t going to happen.
If we want AI to boom in the UK and for our nation to truly be an AI Maker and not an AI Taker, we need to trust in the timeless advice of another UK institution - Sir Michael Caine, and “Blow the Bloody Doors Off”!
That means ripping up outdated public policy on energy, reforming privatised frameworks, and moving beyond endless energy talking shops. It means big, radical thinking. It’s not sexy, and it’s a long way from a PR scoop, but data centres could immediately be made eligible within the government’s Energy Intensive Industries (EII) criteria, reducing 23% of the bill overnight. Thankfully, techUK is already lobbying on this front on behalf of the sector. Industrial electricity costs must be decoupled from gas prices, especially when 41% of the UK’s power over the past year came from renewables and only 27% came from gas. DSIT and DESNZ need to urgently fix grid bottlenecks enabling more green electrons to flow south from Scottish wind farms, reducing costly curtailment fees. Tech giants and data centres should be incentivised to build near renewable sources with zonal pricing - as they are in the Nordics, which will reduce curtailment issues further. This will also turbo-boost economic growth across former industrial areas like Apatura's impressive Ravenscraig site in Scotland, and the new AI Growth Zone in the North East, ensuring this becomes more than just a branding exercise and showing that all of Britain can benefit from AI, not just those with a London postcode. And, the 70GW queue of data centre projects in the UK must be cut back to those who actually have a shot at delivering.
Alongside power, we also need urgent reform to our copyright laws so the UK has, at the very least, parity with the EU on text and data mining for commercial purposes. Training workloads – the fastest growing and dominant form of compute in the world right now – needs the ability to train large language models on data and in the UK we’re making this an almost impossible proposition.
Yesterday at DCD Connect, I had the pleasure of meeting Jennifer Chinenye Olafenwa, the impressive founder behind UK company, Brilliant AI – a specialist AI engineering company focused on designing, building and deploying custom AI systems for real-world impact. Jennifer’s idea and innovation is based in Britain, but because copyright laws here are prohibitive, she’s Making AI in France (on Mistral’s infrastructure) and Taking AI back to the UK. Without reform to copyright she won’t be the only AI entrepreneur following that path.
Yes, reforms like these are not glamorous and they will take hard work, and even harder choices. The government PR machine isn’t going to get many column inches out of these, and they won’t produce Trump–Keir handshake photo ops. But through this hard endeavour, correcting our energy provision and pricing in the UK will create part of the fundamental environment within which a thriving UK AI sector can evolve. As Sir Michael also once said, “the harder thing to do, and the right thing to do, are usually the same thing.”
Fundamentally AI Making is training, AI Taking is inference, and if you’d like to explore why the UK must blow the bloody doors off and address the issues that threaten the success of real AI Making in our country—read Kao Data’s excellent whitepaper: https://kaodata.com/taker-not-maker
Adam Nethersole
Adam Nethersole is VP of Marketing at Kao Data and his experience covers a decade of working within data centres and digital infrastructure. Interested in green tech and sustainable computing.