Planned UK Data Center Delayed Until 2030s Over Power Grid Delays. Microsoft Pulls Out (telegraph.co.uk) 34
July: Nvidia-backed Nscale is building a £2bn data centre at Loughton in Essex, with Microsoft signed up as its anchor tenant. But it can't get the electricity to switch on, reported The Next Web:
Nscale's problem is not really Nscale's. It is Britain's. Grid-connection times have become the single biggest brake on new data centres. Some UK projects face waits stretching close to a decade for enough power. One estimate found more than a quarter of data-centre capacity slipped in 2025 alone.
"The data centre was due to receive its power supply at the end of the year," reports the BBC. "But as first reported by the Telegraph on Sunday, Nscale has been told this will not happen until the 2030s." (Though Nscale and Microsoft are still working together on other projects in Norway, Portugal and America.)
But now anchor-tenant Microsoft has pulled out of the Essex deal, reports the Telegraph, adding that Microsoft's move "leaves the site in apparent limbo. The company was its only announced customer, having agreed to rent 23,000 Nvidia AI chips at the facility..." Microsoft's decision threatens to undermine Nscale's attempts to push ahead with a Wall Street flotation worth up to $35bn (£26bn), despite a brutal sell-off in AI computing stocks. Rival data centre developers have seen their share prices plunge over the last week on concerns that they will struggle to repay their debts as interest rates and inflation continue to rise.... Tech analyst Gartner said last week that $3.5 trillion could be written off by 2030 if the AI boom fails to deliver on its promises.
And "our research indicates that enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures during this period," Gartner wrote: "In 2026, AI providers are expected to spend nearly $1 trillion on AI infrastructure, and are forecast to invest an additional $6 trillion between 2027 and 2030," said Will Sommer, Sr Director Analyst at Gartner.... Failure to achieve sustainable return on investment capital (ROIC) on rapidly depreciating assets would almost certainly necessitate write-downs. Based on Gartner projections, a continued AI infrastructure investment trajectory could leave more than $4 trillion in asset value exposed to write-down risk through 2032. In this scenario, Gartner does not expect the 12% ROIC to be achieved until the end of 2032 at earliest...
Gartner estimates AI-optimized IaaS spending will grow from $66 billion in 2027 to $174 billion in 2030. "The growth in spending on AI-optimized IaaS will be too low for most AI compute providers to recover their AI investments before the current generation of AI semiconductors is replaced," said Sommer. By 2030, Gartner predicts that cost pressure and a lack of a differentiating business model will force over 80% of neocloud [AI-specialized infrastructure] providers to cease independent operations.
"The data centre was due to receive its power supply at the end of the year," reports the BBC. "But as first reported by the Telegraph on Sunday, Nscale has been told this will not happen until the 2030s." (Though Nscale and Microsoft are still working together on other projects in Norway, Portugal and America.)
But now anchor-tenant Microsoft has pulled out of the Essex deal, reports the Telegraph, adding that Microsoft's move "leaves the site in apparent limbo. The company was its only announced customer, having agreed to rent 23,000 Nvidia AI chips at the facility..." Microsoft's decision threatens to undermine Nscale's attempts to push ahead with a Wall Street flotation worth up to $35bn (£26bn), despite a brutal sell-off in AI computing stocks. Rival data centre developers have seen their share prices plunge over the last week on concerns that they will struggle to repay their debts as interest rates and inflation continue to rise.... Tech analyst Gartner said last week that $3.5 trillion could be written off by 2030 if the AI boom fails to deliver on its promises.
And "our research indicates that enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures during this period," Gartner wrote: "In 2026, AI providers are expected to spend nearly $1 trillion on AI infrastructure, and are forecast to invest an additional $6 trillion between 2027 and 2030," said Will Sommer, Sr Director Analyst at Gartner.... Failure to achieve sustainable return on investment capital (ROIC) on rapidly depreciating assets would almost certainly necessitate write-downs. Based on Gartner projections, a continued AI infrastructure investment trajectory could leave more than $4 trillion in asset value exposed to write-down risk through 2032. In this scenario, Gartner does not expect the 12% ROIC to be achieved until the end of 2032 at earliest...
Gartner estimates AI-optimized IaaS spending will grow from $66 billion in 2027 to $174 billion in 2030. "The growth in spending on AI-optimized IaaS will be too low for most AI compute providers to recover their AI investments before the current generation of AI semiconductors is replaced," said Sommer. By 2030, Gartner predicts that cost pressure and a lack of a differentiating business model will force over 80% of neocloud [AI-specialized infrastructure] providers to cease independent operations.
No problem (Score:1, Insightful)
China (Score:3, Interesting)
Re: (Score:2)
It is a general problem in the UK though. Not just data centres are affected, everyone is. People who want solar find that because a few other people in the area already have it, they are limited to about 3.6kW export, and worse still the DNO (the people who run the local grid) won't let them have more than a 3.6kW inverter so they can dump the rest into batteries. They don't trust higher power inverters to properly limit export power.
Re: China (Score:2)
Re: (Score:2)
Re: (Score:2)
It is much harder to get planning permission for a data centre in Scotland though.
Re: (Score:2)
Except a) I read that China has capped the number of hyperscale datacenters. (Check out the dictionary def of the word "enough"), and b) with Texas' electricity grid?
Good (Score:2)
Re: Good (Score:3, Informative)
Linear algebra is uncannily useful. As such, it has more ways of being monetized than there are grains of sand on the beach.
Re: (Score:2)
Re: Good (Score:1)
Speculative overinvestment is the only mechanism by which good ideas can actually get discovered and bad ideas can be identified. Everything else is armchair philosophy limited by the imagination of the philosopher.
Re: (Score:2)
Re: Good (Score:1)
Yeah. I have this great idea for predicting the future. It just needs some startup funding to get off the ground...
Re:Good (Score:5, Interesting)
Sooner or later people are going to realize that AI is just another name for Linear Algebra and all this will come crashing down
There are at least four problems with this. First, linear algebra is from a math standpoint insanely powerful. For example, all of representation theory https://en.wikipedia.org/wiki/Representation_theory [wikipedia.org] is using linear algebra to understand groups. Much of the study of modular forms https://en.wikipedia.org/wiki/Modular_form [wikipedia.org] involves understanding the behavior of modular forms over SL_2(Z) which has heavy linear algebra elements to it. People who see linear algebra as somehow a weak thing are often people who took a single semester of it and didn't do much else with it.
Second, the idea that these AI systems are just linear algebra isn't accurate anyhow. While there's a lot of linear involved, a major part of how transforms like LLMs work is the intervening non-linear layers, which use ReLu https://en.wikipedia.org/wiki/Rectified_linear_unit [wikipedia.org], Softplus, or a similar function. So, summarizing this as just linear algebra isn't accurate. It is linear algebra with some spice.
Third, even if this were somehow just linear algebra, that shouldn't therefore be a reason to think that there's any limits to what is going on. At the end of the day, the empirical success is what matter, and systems which can do things like prove the quasi-Riemann Hypothesis or solve other major math problems are pretty impressive regardless of what is under the hood. So, if one does think this is just linear algebra, that's not a reason to dismiss the systems but a reason to increase your estimate of what can be done with mere linear algebra.
Fourth, slowing down connections like this may make crashes *less likely.* In particular, the main way we could have a serious bubble here is over-investment. Delays including power grid delays force bottlenecks on the investment level, and make that sort of problem less likely.
We may be in a bubble here. Certainly some elements look very bubble-like. In the late 1990s, during the dot-com boom, one had people making perfectly normal businesses but just tacking on "on the internet" to it and it would get millions of dollars of funding automatically. See for example pets.com . And we're seeing some of that now; we've had a whole bunch of businesses see their stock price shootup just because they mention AI. So we may be in a bubble, and it may crash now, or might crash never or decades from now. (Arguably the US had an 80 year automobile bubble that didn't crash until the late 2000s which then forced the auto bailout.) But the other thing to keep in mind is that a crash does not mean the tech goes away. In 1873, the US had a devastating collapse of the railroad bubble, and yet railroads continued to grow, with the amount of track not peaking until around 1910 https://transportgeography.org/contents/chapter5/rail-transportation-pipelines/rail-track-mileage-united-states/ [transportgeography.org]. Similarly, although the internet bubble burst, by 2015, one had more internet use than even the most internet hyping person in 1998 would have likely dreamed of. A bubble bursting does not mean the tech itself goes away.
Re: (Score:2)
Oh I understood the power of linear algebra as a kid when my uncle showed me various ways of using them and matrix maths to solve various problems and my little brain exploded at the possibilities.
And I am *very* certain we are in a bubble. Wall street is has been doing pretty frigging insane things for the last 4-5 years, with absolutely irrational valuations of various key stocks, and the thing with markets, is the *always* correct eventually. And sometimes those corrections involve entire markets catchin
Re: (Score:2)
We may be in a bubble here. Certainly some elements look very bubble-like. In the late 1990s, during the dot-com boom, one had people making perfectly normal businesses but just tacking on "on the internet" to it and it would get millions of dollars of funding automatically. See for example pets.com . And we're seeing some of that now; we've had a whole bunch of businesses see their stock price shootup just because they mention AI. So we may be in a bubble, and it may crash now, or might crash never or decades from now. (Arguably the US had an 80 year automobile bubble that didn't crash until the late 2000s which then forced the auto bailout.) But the other thing to keep in mind is that a crash does not mean the tech goes away. In 1873, the US had a devastating collapse of the railroad bubble, and yet railroads continued to grow, with the amount of track not peaking until around 1910 https://transportgeography.org... [transportgeography.org] [transportgeography.org]. Similarly, although the internet bubble burst, by 2015, one had more internet use than even the most internet hyping person in 1998 would have likely dreamed of. A bubble bursting does not mean the tech itself goes away.
I think the major difference here is that all those things had and still have undeniable utility. Of course, even the most useful things have limits to their usefulness - there's places where building railroad doesn't make much business sense, there's businesses which are never going to be viable even when going online, or businesses which aren't going to benefit that much by going online etc and that's why the related bubbles became bubbles . Not sure there can be such a thing as an 80 year bubble. But any
Re: (Score:2)
Re: (Score:2)
The business case for LLMs is a bit broader than that. They don't need to replace that many workers if they make enough workers more efficient or more reliable (such as simply having LLMs look over drafts).
As things stand 'reliable' is not a word you associate with AI. This being the case, you either need to get people to check the AI output, negating most if not all of the benefit from using it in the first place or accept that what AI produces may simply be not that good - far from ideal in either case. Given the insane amount of money invested, adoption by businesses has to be overwhelming. My impression is that this is very far from being the case. Also some of the organisations that are currently using A
Re: (Score:2)
The business case for LLMs is a bit broader than that. They don't need to replace that many workers if they make enough workers more efficient or more reliable (such as simply having LLMs look over drafts).
As things stand 'reliable' is not a word you associate with AI.
That's why it is using it to look over drafts. You take the AI output and then go through its suggested edits or concerns. Essentially it functions as a very fast, half-way decent proof-reader. For example, my spouse looks over a lot of drafts of things I write, but if I give her a 15 page paper, there's good work on the first 8 or 9 pages, probably a joke on page 10 about my notation, one or two comments on the next page, and then a few comments on the bibliography. Claude is not as good a proof-reader as
I propose ... (Score:2)
We need: Resource Complete.
Something we can use to keep the investors engaged with the idea that somewhere, somehow, this will be delivered and the project will be right back on track. See: Real Soon Now [grokipedia.com]
Good, nothing of value to the public is lost. (Score:2)
The UK is already overcrowded and property punitively expensive.
Re: (Score:3)
The UK is already overcrowded and property punitively expensive.
Really? Got evidence?
I am baffled that they were going to build this centre in Essex, which is more crowded and expensive than other parts of the UK. We need to convince foreign businesses that other parts of the UK, apart from London and the SE of England exist. There are lots of places in the north of England, Scotland and Wales where connection to the National Grid is easier, and where other costs are substantially lower.
Not only UK (Score:2)
The problem is UK energy policy (Score:3)
The problem is UK energy policy. It is to build large amounts of intermittent power generation in places where there is no demand.
The business case for doing this rests on LCOE calculations which omit the costs of remedying intermittency and also the cost of getting the power to where there is demand, in the present case, Essex, but in general in the Midlands and South.
The result is that the UK has to build a hugely expensive new transmission network from the wind farms off the north coast of Scotland to the Midlands and South. It still has no specific problem for how to supply continuous reliable power from wind - the only solution is of course gas fired generation, but that is politically incorrect.
The country has the highest electricity prices in the OECD. Which as inevitable - however you fake the numbers with LCOE, the costs are real and have to be recovered somehow.
The program also involves large amounts of solar. However, the UK is the second worst country in the world for solar power generation (the worst is Ireland). This doesn't help either, and just increases the size of the intermittency problem.
What is coming towards the UK is probably a nationwide blackout, and certainly rolling regional ones. If they are sucessful in the other part of their program they will increase demand substantially by converting heating to heat pumps and transport to EVs. And by persuading companies to install lots of data centers for AI etc. Sooner or later these plans will run into the one or two week period which occurs most years, when there is a blocking high to the southwest and dead calms. Wind generation falls well below 10% of faceplate and stays there for a couple of weeks. At 4pm, this being winter, solar vanishes - there has been negligible amounts anyway even earlier in the day, this being winter at these latitudes. There are no plans to keep supply going when this happens. They need plans to build out masses of new gas - because the old installations they depend on at the moment are reaching end of life. Are there any such plans? Of course not. And if you were building out that much gas, why do you need all that wind and transmission?
You don't!
This cannot work. You would have to be blinkered or silly (or paid off with tax breaks, and even then its not clear these could be big enough) to contemplate investing heavily in a data center operation in the UK till they get this turned around, as in U-turn, abolish the net zero silliness and build reliable low cost generating capacity. And stop building transmission links to nowhere.
As for Global Warming and Climate Change, its irrelevant. Whatever is happening with the climate, there is nothing that UK energy policy can do to have any effect on it. All the current programs can do is make a nationwide blackout more likely. And the world will then, if it happens, watch a world first: a country trying to restart its grid using wind power. Good luck with that one.
Re: (Score:2)
Your description of the problem is mostly good. I'm not sure I agree with your conclusions though.
The UK has built out a lot of renewables, which regardless of other issues has to be a good thing. You're right though, the grid isn't big enough to get that power to where it is needed very effectively, so we have to pay the generators to turn wind farms off (for example). That's clearly a crazy situation which can't go on for long. We also have high electricity costs because we have a janky market where we pa
Re: (Score:2)
if you were building out that much gas, why do you need all that wind and transmission?
You need the wind so you can use less gas. You need the transmission so you can use the wind. Burning fossil fuels is bad in many ways so doing as little of it as possible makes sense. I tried to use small words for you there.
Re: (Score:2)
"You need the wind so you can use less gas. You need the transmission so you can use the wind. Burning fossil fuels is bad in many ways so doing as little of it as possible makes sense. I tried to use small words for you there."
Not what they are trying to do. What they are trying to do is run the country on wind and solar. Without having any provision for the one to two week calms which take place most winters, when wind production falls below, and often well below, 10% of faceplate.
Burning fossil fuel ma
Re: (Score:2)
The problem is UK energy policy. It is to build large amounts of intermittent power generation in places where there is no demand.
Not really, or rather only in as much as this is because we have other problems. The problem is a massive lack of investment in infrastructure (i.e. the grid), and absolutely zero long term planning. Sure there might be plans, but they'll change every 5 minutes anyway, just to ensure that no companies invest properly either.
The result is that the UK has to build a hugely expensiv
Re: (Score:2)
Anyone building an AI data centre in Essex needs their head examining. Build it somewhere in Scotland we're with DLC you can do 365x24 free air cooling. Oh and your close to all that renewable energy so no need for building massive amounts of grid infrastructure to transport energy down south.
The problem (Score:2)
No, it's their's that there isn't just infinite energy to go around being pulled out of nowhere.
What Does This Mean? (Score:2)
Gartner predicts that cost pressure and a lack of a differentiating business model will force over 80% of neocloud [AI-specialized infrastructure] providers to cease independent operations.
Does this mean the State takes over?