How I learned to stop worrying and love the datacenter • The Register
Comment The UK has bitterly expensive power, an energy minister who sees electricity as bad, a lethargic planning system, and a grid with a backlog for connections running to 2039.
So why would Google spend five giga-quid on its latest datacenter, while Nvidia and others like Oracle raise the Brit barn investment fund to a claimed £31 billion (c $42 billion)?
The glorious visit of US President Donald Trump to the British Isles last week came with promises of thousands of precious GPUs, with Nscale, Microsoft, Google, and more teaming up to build a giant datacenter within walking distance of my house. But is the Tech Prosperity deal really a portal to a bright future or will it suck us in?
Even before Nvidia’s Jensen Huang touched down in the UK, real estate expert CBRE was reporting that DC construction wasn’t aligned with demand; some investors were taking a “build it and they will come” approach. CBRE pointed out in a January forecast that such an approach is sensitive to risks from the general economy – it urged onlookers to consider the risk of tenant flight in a recession and the need to balance business ambitions with rising energy costs and planning limitations.
As for the coming growth for hyperscalers, it will come partly at the expense of on-prem. Except for our large financial sector, enterprise customers running their own datacenters are in decline. There are just so many risks and hassles. Do you know whether your cabinets in five years are going need 10 kW or 100? Nvidia threatens to roll out 600 kW racks within 18 months and a megawatt after that.
But what will be the mix of inefficiently coded AI, moderately efficient big data, and bog standard database and compute in these new data barns? No, I don’t know either and it’s a big bet. Air cooling is cheaper but retrofitting for liquid isn’t just expensive; construction work in your delicate datacenter brings its own risks.
What if this is a bubble?
There are risks on a national scale as we build power and water infrastructure to feed our new robotic overlords. If you’re building a DC, you have to commit to five to 10 years of consumption – and it takes two years to build an electrical substation after you have the permits. You’re smarter than me if you can forecast where AI will be then. Yet generation and grid are amortized on the assumption of having a customer for 20-30 years. That’s a big gap if AI is a bubble or if inference or some other technology makes it less hideously inefficient. You may not love the energy biz, but if it has a financial hole, so do we. And just what effect will all this have on the Net Zero pledge?
Google has signed up Shell to provide renewable power, but Nvidia wants Ed Miliband, UK Secretary of State for Energy and Climate Change, to go for gas to help Britain meet the 80 gigawatt shortfall we’re already looking at. This was not at all what he wanted to hear.
When we add in the three years to build the DCs themselves, there is a big risk term in guessing what volume and type of demand might be there as well as the loss to the economy of what is now officially designated Critical National Infrastructure.
So the government is “calling in” planning applications for this specific type of infrastructure, over-riding the tedious planning process, and having a measurable effect on the whole economy. Has there ever been a government industrial strategy that has ever worked though? If so, I am not aware of it.
Come friendly bits and fall on Slough, it isn’t fit for DCs now
The standard measure for the scale of a datacenter is megawatts, not Tera Operations Per Second (TOPS) and the awful price of UK power is only the first problem. Telehouse’s Mark Pestridge, exec veep and GM for Europe, is clear that one of the first questions colocation customers ask is whether they can access renewable power at the right scale. This gets audited, because financial regulators are at last clamping down a little on the fantasy numbers we associate with any ESG claim.
Slough & west London are hitting limits on power and that’s why, despite connectivity pushing large datacenters to huddle together, they are instead sprawling out. Cambridge and Oxford are hitting the limits of their existing water supply, so placing DCs is becoming ever more tricky even without dealing with local government. To be fair, it’s not just Luddism; the superscalers do put a bit of effort into helping the local economy, in fact Google’s PR kindly sent me a two-page letter on how they loved the locals and will care for us and the construction blips up the local economy.
In the longer term, components, energy, and tech are all sourced from far away. The DCs themselves may have more security staff in place than actual engineers. They might represent a loss of potential jobs that the site might have supported.
Bronze medal for Britain
The government’s ambition is for the UK to hold third place in the AI pecking order, which – given the UK has twice as many of the world’s top 100 universities than the whole EU put together and is not encumbered by the EU AI Act – might be more feasible than most Brits themselves realize.
But even before Trump’s second term, technology relations with the US have always been transactional and often one-sided, with the States ignoring and bending commitments. There is a talk of exchange programs of leading researchers, which, given the behavior of the US’s Homeland Security and ICE of late, is going to be a challenge for top techies who are not straight or white. So the UK may well become a convenient compromise for those who want to work at the top tier but without the personal risks of going to the US, or as we’ve seen in the last couple of days, arbitrary $100,000 “fees” for H-1B visas.
Staffing problems
Remember outsourcing to India and what a good idea that seemed like at the time? Fewer jobs in tech support and mediocre pay, so guess what? The intake of fresh blood became a trickle years ago, made worse by many HR departments that would rather hire rabid wolves than those over 45 so the datacenter industry is facing a demographic crisis as the skills to commission and support are retiring. No HR director, they can’t just “go on a course next week.”
Skilling up what colo provider Telehouse calls “bright young things” for its apprenticeships, a multi year process and the industry as a whole is only slowly moving to work out how to plug the holes. That should mean better pay and maybe that will filter down to schools, encouraging kids to consider options more realistic than becoming an influencer. But in the meantime, a shortfall of skills – as so often with the UK economy – will hold us back from exploiting the advance.
Of course we could and should welcome the immigration of engineers – but someone should tell the Home Office that.
Tech sovereignty
Sorry, you can’t have that.
The business of Telehouse and other colos is partly driven by companies and their customers who need to feel comfortable that their data is physically stored in the UK/EU, away from the notorious US Patriot Act which the superscalers have made clear they will apply in any country the Trump administration asks them to outside the US.
The vast majority of investment is for the UK to process data in systems owned and developed by American firms. The scale of US R&D investment utterly dwarfs that of the UK, so whatever benefit the UK gets from the technology is from using it to improve the country’s business and government. That could be huge, if we take the Microsoft stack as an example. Something north of 90 percent of the value generated is generated by users rather than disappearing in tax-optimal ways into Redmond.
Despite being desperate for tax revenue, it seems this AI DC investment deal has killed the idea of the UK increasing Digital Service Tax on large US firms for fear of strangling the agreement at birth.
Government attitudes to UK cloud providers have always been somewhere between disdain and indifference, so expect the vast majority of its cloud spend to go to AWS, Google, Oracle, et al as we saw with the collapse of UK Cloud.
However this expansion does mean that UK biz and government may develop the skills to exploit the new techs, which has historical precedent in the 1980s where a lot of UK growth was driven by having a more computer literate workforce than pretty much anyone else.
What have datacenters ever done for us?
Is this is a good move? Our government has decided to throw in its lot with the American leviathan. The stakes are high; we could try going it alone but we don’t have the cash. Joining the EU projects will be attractive to many but the bloc’s strategy of “regulate then innovate, then regulate some more” is optimized by fear not greed.
A (mostly) modern economy like ours relies on a whole pile of digital services, streaming, shopping, numerical modelling, sensors, and databases. If you want to keep on working from home, we’ll need that as well.
To Reg readers, this is not news, but the industry has done a miserable job of telling the general public about it. The brutal reality is: if you want the UK economy to grow, just like the first industrial revolution, you’re going to need dark satanic mills. ®
