Shiny new AI data centers and duller questions – what could possibly go wrong?

In the wake of Trump’s arrival in the UK, major US AI companies have announced tens of billions of dollars in new AI investments for developing more and better AI capabilities in the UK. According to this narrative, the UK will gain the largest UK-based AI supercomputer ever built, numerous additional AI data centers, increased job opportunities, and substantial financial benefits - unless it doesn’t. 

US Companies, including Google, Microsoft, NVIDIA, OpenAI, CoreWeave, Scale AI, BlackRock and AI Pathfinder, have all announced support for this 'Special Relationship'.  NVIDIA alone plans to deploy 120,000 high-end GPUs across Britain, which is the largest deployment in Europe to date. UK-based Nscale plans to deploy half of these to support OpenAI's ambitions to build Britain’s largest AI supercomputer. 

Advocates claim that this will support economic growth, scientific research, and energy security in both the US and UK, thanks to innovations in healthcare, quantum computing, and streamlined nuclear projects. What could possibly go wrong? A lot actually -  like higher water and power bills, more piles of radioactive pollution, loss of data sovereignty, and more efficient wealth extraction schemes. A great deal is at stake in determining how this investment can be leveraged to foster growth in environmental, human, and social capital. 

The elephant in the room

One big question that seems to be glossed over in the rush to build more data centers is who or what will be subsidizing the costs of the power that goes into or the heat that comes out. Also, what can be done to reduce the slow permit approval processes and relatively high construction costs compared to other markets? 

The power infrastructure conversation generally seems to get glossed over in the excitement on the road to AI data center nirvana. A large data center can easily consume more power than a small city. This could mean more carbon-emitting fossil fuel plants or perhaps a lot more wind and solar that no one wants to see blighting their view. 

This also means building a lot more power lines that no one wants to see, either. The US AI industry has a handy wealth extraction scheme where it leaves these grid connection costs up to consumer rate payers. The same approach is also being tested for subsidizing the water infrastructure costs required for cooling these data centers. This kind of thinking will surely be tested by existing water management challenges in the UK.

Small modular nuclear reactors are being proposed as one way to mitigate the power and grid requirements of new data centers. They could be built on-site and don’t emit air pollution or greenhouse gases. They might also be safer than traditional nuclear plants. Except no one has figured out how to safely and durably dispose of the nuclear waste from the current facilities, much less the new ones. Some estimates suggest that these could generate anywhere from two to thirty times as much nuclear waste as traditional nuclear plants per watt of power. Newer nuclear designs will also require new processing techniques due to the use of novel materials and designs. 

What’s more, the construction costs of SMRs are coming in at significantly more than expected. The cost of NuScale’s first US SMR surged from $5.3 to $9.3 billion before it terminated the project. Estimates place the average levelized cost of energy at $200 per megawatt hour compared to $45 to $74 for gas, $26 to $50 for wind, and $59 to $91 for rooftop solar. 

All of these should leave regulators and ratepayers wondering if the UK’s data center capacity can scale quickly and sustainably enough to meet the demand. Konstantin Hartmann, Managing Director, EMEA, at NTT Global Data Centres, argues in a statement:

As the industry scales to meet this demand, sustainability must remain front and center. Major builds will and should face scrutiny from media, regulators, and communities.

The data dilemma

Another concern is how this partnership will balance the UK’s more stringent data protections with the more permissive move fast and fix things later attitude favored by US firms. 

Mark Boost, CEO of UK-based cloud and AI provider Civo, says in a statement that while tech investment is always welcome, this deal ultimately takes another step towards surrendering the UK’s digital sovereignty:

Any investment into our tech sector is worth celebrating, and will help accelerate the AI ambitions of countless British businesses. That said, even though the ‘Tech Prosperity Deal’ is an impressive deal with a catchy name, I have to question which side is prospering more. If the UK isn’t careful, sooner or later, the majority of our critical AI infrastructure will be owned by Big Tech. The US Cloud Act means none of the ‘Big Three’ providers can offer true digital sovereignty, leaving British businesses and public bodies completely at the mercy of American data laws.

The silver lining?

That said, this could also represent an opportunity to combine the UK’s traditional strengths in scientific research, transparency, collaboration, and social benefit with America’s strengths in financial engineering and digital innovation. This middle ground will require investing as much in people, trustworthy data infrastructure, new processes, and physical AI as in the shiny new data centers that tend to hog all of the photo ops. 

One ray of hope is Salesforce’s strategy for investing in the R&D and talent required to cultivate local AI skills and capabilities.

Another ray of hope is that the significant investment in UK operators like Nscale might also help UK engineers collaborate on more efficient data centers and construction processes that could guide better practices worldwide. 

Additionally, there is an opportunity to leverage the UK’s extensive scientific expertise in the next phase of physical AI tools and processes, which can improve efficiency, design more cost-effective power plants and batteries, and potentially streamline regulatory and permitting processes by utilizing more advanced digital twins. These kinds of things might benefit from Nvidia’s UK R&D investments in applying AI to material innovation, electrostatic chemistry, data center digital twins, pollution models, and better training tools. 

Perhaps some of this will translate into more sustainable nuclear fusion power plants that produce less radioactive waste per watt. It may also reduce the need for freshwater or support new approaches to quickly and cost-effectively rewiring the grid with higher capacity upgrades. 

Allan Kaye, director of Manchester-based data center specialist Vespertec, argues in a statement that the US investment could clearly give the UK a leg up. Still, it's important to recognize the need to proceed carefully:

It’s vital that we work with our US partners on an equal footing to build up British AI, rather than just becoming more real estate for American data centers. There is a path forward that ensures the rising tide of AI lifts ships on both sides of the Atlantic, so we need to work with our US partners to make sure that’s the path we’re pursuing in the future.

My take
The prospect of billions of dollars in new investments and shiny new data centers makes for good press releases. Building a large number of data centers may create numerous short-term construction jobs. However, the bigger opportunity will be figuring out how to grow the talent pipeline and developing the physical AI tools and processes that translate into more sustainable and trustworthy AI, which in turn generates even bigger gains in the long run.