Microsoft pens $15B love letter to the UK with 23,000 Nvidia GPUs attached
- Reference: 1758106262
- News link: https://www.theregister.co.uk/2025/09/17/microsoft_uk_ai_cash/
- Source link:
Tony Blair Institute: UK needs bit barns to lead in AI deployment, not training [1]READ MORE
The money is part of a $30 billion [2]spending spree between 2025 and 2028, although Microoft has been coy as to exactly where the cash will be spent. Google, on the other hand, has been very clear about where it is building its [3]multibillion-dollar bit barn .
The announcement came on the eve of US President Donald Trump's state visit, and Microsoft was keen to highlight its more than four decades of involvement in the US-UK relationship. "Today," the company said, "this alliance enters another new chapter, as we work together to expand access to trusted American technology and strengthen the infrastructure that will drive economic growth and technological advancement in the AI era."
The issue of American tech giants in the era of Trump has become vexing for countries on this side of the Atlantic, where [4]efforts are underway at some biz customers to extract themselves from the clutches of the hyperscalers.
Such controversies, including Microsoft [5]admitting during a hearing in France that it couldn't guarantee data sovereignty, have not deterred the megacorp. "This marks the largest financial commitment we've ever made in the UK," Microsoft said.
[6]
"This investment will enable us to build the country's largest supercomputer – with more than 23,000 Nvidia GPUs – in partnership with Nscale."
Datacenters guzzled more than a fifth of Ireland's electricity in 2023 [7]READ MORE
In 2023, Microsoft [8]announced a three-year, £2.5 billion investment in the UK, also aimed at AI infrastructure.
Microsoft is not alone in splashing the cash – or at least announcing that it will – in the UK to mark Trump's visit. Nvidia also [9]announced that investment would be headed to Britain. It reiterated Microsoft's supercomputing ambitions, although it put the GPU count as more than 24,000. What's a thousand or so Nvidia GPUs between friends?
[10]
[11]
CoreWeave is to [12]invest another £1.5 billion as part of an initiative to deploy Nvidia GPUs in Scotland.
Jensen Huang, founder and CEO of Nvidia, said: "Today marks a historic chapter in US-United Kingdom technology collaboration. We are at the big bang of the AI era – and the United Kingdom stands in a Goldilocks position, where world-class talent, research, and industry converge."
[13]UK Cabinet Office hands stalled Microsoft migration to another department
[14]Google unmasks itself as mystery hyperscaler behind yet another UK datacenter
[15]Google lands £400M MoD contract for secure UK cloud services
[16]UK government trial of M365 Copilot finds no clear productivity boost
Mark Boost, CEO of UK cloud vendor Civo, sounded a cautious note, commenting: "Clearly, international investment is vital for our AI industries, and a package of this size will do a huge amount of good for quantum, AI and nuclear projects. However, we should also be building up our homegrown tech ecosystem, and setting harder limits on the extent to which we're willing to allow overseas control of our digital infrastructure. If not, we risk sacrificing our future digital sovereignty on the altar of short-term gains."
Microsoft can't guarantee data sovereignty – OVHcloud says 'We told you so' [17]READ MORE
While it is not clear exactly where the next investment will be directed (other than the aforementioned supercomputer), the UK's infrastructure has struggled to keep up with the demands of tech giant datacenters, particularly in the age of AI. Earlier this year, [18]planning rules were loosened to permit more development, but [19]power generation problems could still slow growth.
In May, the UK government [20]stepped in to overturn a local council that refused planning permission. Datacenters are, after all, [21]critical national infrastructure now.
[22]
Music to the ears of tech giants, judging by the investments, though perhaps less so for locals forced to deal with potential power shortages, or electricity price rises as is [23]feared in the US , not to mention the [24]environmental challenges that could come from all that AI and cloud infrastructure development. ®
Get our [25]Tech Resources
[1] https://www.theregister.com/2025/08/04/tony_blair_institute_says_uk/
[2] https://blogs.microsoft.com/on-the-issues/2025/09/16/microsoft-30-billion-uk-ai-future/
[3] https://www.theregister.com/2025/09/16/google_hertfordshire_datacenter/
[4] https://www.theregister.com/2025/04/17/us_hyperscaler_alternatives/
[5] https://www.theregister.com/2025/07/25/microsoft_admits_it_cannot_guarantee/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aMrbGwRDxdtwTDuY8y0BmwAAAhI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[7] https://www.theregister.com/2024/07/25/ireland_datacenter_power_consumption/
[8] https://www.gov.uk/government/news/boost-for-uk-ai-as-microsoft-unveils-25-billion-investment
[9] https://nvidianews.nvidia.com/news/nvidia-and-united-kingdom-build-nations-ai-infrastructure-and-ecosystem-to-fuel-innovation-economic-growth-and-jobs
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aMrbGwRDxdtwTDuY8y0BmwAAAhI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aMrbGwRDxdtwTDuY8y0BmwAAAhI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[12] https://coreweave.com/news/coreweave-announces-significant-commitment-to-power-uk-ai-innovation-and-growth
[13] https://www.theregister.com/2025/09/17/cabinet_office_microsoft_migration/
[14] https://www.theregister.com/2025/09/16/google_hertfordshire_datacenter/
[15] https://www.theregister.com/2025/09/12/google_cloud_mod_contract/
[16] https://www.theregister.com/2025/09/04/m365_copilot_uk_government/
[17] https://www.theregister.com/2025/08/27/ovhcloud_interview/
[18] https://www.theregister.com/2025/01/13/uk_datacenter_planning_rules/
[19] https://www.theregister.com/2025/02/10/london_has_400_gw_of/
[20] https://www.theregister.com/2025/05/16/uk_overrules_local_council_approve_datacenter/
[21] https://www.theregister.com/2024/09/17/objections_to_datacenter_builds_cni/
[22] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aMrbGwRDxdtwTDuY8y0BmwAAAhI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[23] https://www.theregister.com/2024/11/22/ai_hike_energy_bills/
[24] https://www.theregister.com/2025/01/04/how_datacenters_use_water/
[25] https://whitepapers.theregister.com/
Re: Fixing AI with local resources.
You've just described a Facebook group.
Re: Fixing AI with local resources.
There are other ways of getting information out there, but AI is being offered as a platform for what Musk wanted his X thing to be. That's why they are adding stuff like agentic shoppers on to it. Nobody has actually managed to take any individual service and expand it that far. They want it to be your personal assistant rather than an app on your device, your go-to interface for everything you do.
The tech is not there yet for that, most users won't want it, and the star topology, with everything going back to a single entity, where it can be searched and harvested, is really not a good way of doing it anyway.
AI may offer a half-decent user interface for the general public, but they have cut corners and ballsed it up. The quick and dirty route doesn't work well and greed undermines functionality. Facebook has too many adverts. Even radio nowadays has too many adverts.
What I'm suggesting is a distributed aftermarket fix for the areas we find problematic. As with Tampermonkey and the saving of streams, they may not want it and may block it. But it is a way for AI to be useful, reliable, localised - data doesn't have to go back to the mothership, and generate cash. The whole billion dollar/datacentres/magic stuff is just a scam. But we can dig something useful out of it if they allow it.
All your sensitive personal data held on UK government computers
are belong to US.
Bollocks...again!
A £20billion investment.
What a fucking lie but another lying scumbag.
Where will most of the "investment" be spent.
Oh yes, US chip companies and server makers.
Apart from the building of the DC, which will then be over in 6 months, and about 40 - 50 jobs. That's it.
So for 40 or 50 jobs, there will be a huge drain on electricity and water supplies impacting the entire region.
But governments (all sides) are to fucking stupid / corrupt to notice.
"Microsoft pens $15B"
Too expensive.
Mightier than the sword - so pretty cheap as defence procurement goes.
Fixing AI with local resources.
This is a desperate government clutching at an expensive straw offered by snake oil salesmen.
AI is simply not that good. But it may be possible to fix it with an aftermarket hack, to make it useful and reliable in some areas. And part of it, at least, can be onshored. Longer explanation below.
Distributed systems could have fixed most of the problems folk have with social media. A little bit of distributed tech might also fix some of the problems associated with AI. This is a version of a paper that explains a way that it might be done.
GAFA move fast and break things, so they produced a general AI product, using all available content to train software and deliver results. Ideally this would have been two processes. The first could use as much content as is available to teach the software how to process queries - effectively a natural language interface for Search. The second would utilise good quality data to provide high quality responses.
The 'quick and dirty' method has left AI software with polluted data sources, fiction, opinion, obsolete material etc. They could have included a mechanism for choosing the data sources that are considered when creating output, but generally have not.
Can we use the capabilities of general (but flawed) AI services to deliver reliable information?
I think we can, using a distributed topology, relying on GAFA (or alternatives) for a functional 'AI'-like interface, and distributed (local) data for reliable answers to queries.
This is a little like using a Chrome extension like Tampermonkey, that intervenes between webpage content and the user, to filter or manipulate the original intentions of the webpage provider, for example, you used to be able to save a YouTube stream with it. Middleware can be used to fix flaws and inadequacies.
So, create a single general purpose application that can be used in localised environments, such as a shop, office, tourist site or school. A stash of high quality, reliable and relevant source data is then built locally and made usable by this application. It can easily be kept up to date and supplemented from other sources: A brand owner can supply information about their products to a shop in a suitable format to be included. A tourist board can identify things in views and supply route information. A corporation can supply in-house information. None of this data needs to be passed to the AI services. It could be kept local.
This data is then prioritised over, or entirely replaces the data that AI systems on users' devices utilise, when they are in a specific area. So you are using the AI's general interface and processing capabilities, but without its generalised, 'polluted' data sources. You are switching the AI service back to a two stage process, implementing the second in a distributed, localised manner as an aftermarket fix.
This would typically operate via WiFi to mobile devices, but could operate via a corporate intranet. For public use, a simple 'click yes to use' button would be enough, but services could be password restricted or require age verification, as needed.
Local site owners would build up their own cache of information to be used by AI services. It could be downloaded as required, or the process could operate on the system that held the data and simply interact with users directly.
Third parties could pay to have their information added to localised repositories. So a chain of eateries could deploy information about their menu to anyone using AI at a site near one of their branches. The chain would pay the local repository operator, rather than GAFA. This would create a more widespread AI economy and keep data local.
Whatever the variables involved, a single application should be able to manage any iteration, locally storing and serving data, and work with any compliant AI service, publicly or with gating. A matching app would be used on devices, or the AI services would tweak their apps to work with this alternative model.
Whilst it may require some compliance from the AI vendors (if you could not simply order the AI to restrict its data sources), this could well be the killer app of AI - one that can make it useful and reliable in some circumstances. Whilst rogue repositories could be set up, any software can be used for good or ill. That's more of a people problem than a tech problem.
This should offer some utility and commercial value for AI. It would never match the expense being laid out for data centres, court actions and the like, or the promises, but nothing AI does or will do will generate that sort of ROI.
As for cloud storage, I would everyone to avoid it. You are simply placing yourself at the mercy of a third party that wants as much of your cash as it can get. A huge red flag for any business.