This is all starting to smell like financial engineering games. Traditionally nobody in their right mind would give a startup billions to build data centers. For a long list of reasons, that’s kinda nuts.
However what it does allow all these companies investing to do is fund significant capital expenditure but hide it on their balance sheet. They all know if they funded capex directly it would create a deprecation storm that would tank their future earnings. Instead they give the money to another entity to do the building and magically it’s (the equity) now just an asset on their balance sheet with no deprecation. It’s “worth” a lot as a line item there, but only because the hype driving this financial engineering keeps the shares valuable.
Meanwhile the startup isn’t public and thus the fact that it has this massive deprecation on the books is mostly out of sight and out of mind, with some random sky high valuation that’s not based in any normal sense of business reality.
That all works great… until the bubble busts of course.
The business plan makes sense to me. They are a company that is focussed specifically on building AI data centers, which is a huge part of the economy at the moment. The big cloud players know about generic data centers, but there are likely big efficiency wins to be gained by specializing on AI. There is also the geopolitical angle: European countries (and others!) will likely trust a UK-based company more than one of the American BigCos. NVidia is a great partner and investor for them: NScale will buy billions worth of NVDA chips, and also send information and learnings about the unique needs of the market to the chipmaker.
That being said, financial engineering tricks like depreciation and tax sheltering are of course hugely important in the global economy. It's likely that NVDA has a lot of cash sitting in Europe that it doesn't want to repatriate because it would have to pay taxes on it.
> there are likely big efficiency wins to be gained by specializing on AI
I've seen this suggested before as well, but I don't think I've heard a lot of actual concrete things. What big efficiency wins are to be had specializing on "AI" datacenters as opposed to what the past mega hyperscalers have done? What techniques seem to be out there that cloud providers and others have slept on? What makes them so different, in terms of operating a datacenter?
I know nothing about the inner workings of a large scale AI datacenter but I'd imagine the power and cooling requirements are more specialised than your average datacenter that mostly has to handle transmitting large amounts of data over the internet (not that that isn't computationally expensive, I just imagine LLMs (especially at the current scale of their deployment) are much more demanding)
> I'd imagine the power and cooling requirements are more specialised than your average datacenter
But are they actually doing things differently than the high compute parts of the hyperscaled datacenters? Are there radical new ways of distributing heat in the datacenter that only makes sense at that level of energy usage per square foot? Is AI energy use that much higher per square foot of other high-compute parts of datacenters, or is it just that its now something like 90% of the floor plan versus maybe only 50-60%?
> handle transmitting large amounts of data over the internet
I certainly can't speak for all datacenters, and I've never been in a hyperscaler datacenter. But of all the datacenters I've spent time in, the space for the outside network connectivity was rather small compared to the rest of the space for storage and compute. Think a few small office suites dedicated to outside networks coming in and connecting to the clients in the datacenter compared to a medium to large sized warehouse full of compute and storage.
There's "high compute", and then there's proper HPC. AI these days is way more on the HPC end of the scale. The GPUs are doing computations using 2-bit and 4-bit numbers and not 64-bit, but everything else is going to be comparable.
Deprecation is the right way to spell it when it comes to GPUs, because the hardware becomes totally obsolete and uncompetitive after only a few years and you have to replace it wholesale. It's a slow-moving trainwreck.
How does Nvidia's backing of this startup shape competitive dynamics in the AI infrastructure space? If Nscale relies on Nvidia for both capital and GPU supply, does that create an uneven playing field for alternative GPU vendors looking to get their hardware into large multi-tenant AI data centers? Is this $14.6B valuation mostly predicated on continued preferential access to Nvidia's GPU supply, rather than any unique technical or operational advantage that Nscale itself owns?
My observation has been that the areas where data centers makes the most sense (colder climate, cheap energy, cheap real-estate, trustworthy countries) to build, are also areas where resistance against data centers has started to gain momentum. The spike in energy prices alone will make building these centers an uphill battle.
Good, if these multi billion dollar companies can't afford to build a data center that... At the bare minimum doesn't affect their cost of electricity over the next 10 years, nor the infrastructure, then it's totally reasonable for them to resist it. It's not like building data centers is for the common good like building new nuclear power plants or other key infrastructure. Instead it's literally just profit motivated, and not even by consumer spending.
I don't get the economics behind building AI DCs in the UK instead of a middle income country near the equator where there's plentiful solar or fossil fuels.
Labor and land is expensive, energy is scarce and expensive, and colocation is not that valuable because latency is dominated by compute instead of transmission.
Jeez what a way to dismiss the second largest continent and the second most populous landmass in the world.
I'd argue Africa is the best place to start such a thing. Cheap labour, plenty of renewable energy. The biggest issue it probably has is how little bandwidth it has, and might require additional sea cables.
But "africa is africa" is not a way reason to dismiss the continent.
> india is too politically risky. people will just steal your shit
India is a huge place and some parts of it are vastly better managed than others.
> latin america no infrastructure (highways, power grid, etc)
The governance problem is quite real in Latin America actually, but I think it may have potential. If some infrastructure gets funded in the process it would have beneficial side effects all around.
I'd imagine cooling? I could be wrong but a lot of the cost is there and putting what's effectively a megawatt scale electric heater certainly isn't going to make that any cheaper
That's a public good that occurs if the government puts their thumb on the scale with subsidies or regulations, because otherwise the market can't get there itself (example: TSMC building plants in US only because of government incentive). But the Nscale case looks like private investors deciding this is the most profitable place to build it on the margin, which is anomalous.
Huh, you think the equator is better for datacenters? I've never heard that.
I mean, there's a reason Luleå is a popular datacenter spot: access to hydro & wind- but also "free cooling"; datacenters create a lot of heat and having a cold environment makes it easier to maintain ambient temperatures that are sustainable with lower energy consumption.
Additionally, people tend to prefer their servers are in geopolitically stable countries that are less likely to be bombed or undergo civil war (and the US likes to ensure civil war in countries that have a lot of oil) - and you want your datacenters to avoid any environmental hazards like earthquakes.
I can't imagine preferring the equator for datacenters.
Chances of this startup pulling a Theranos are? I mean data center construction is something that couple of non startup companies do and do it well. What is the problem that they solve? The article is quite light on what they actually do.
Arbitrage most likely. They take money from big players, and probably outsource the different parts of construction, then they hold the capex on their balance sheets instead of the big players
I think it’s lagging indicator. There were signs of saturation in the end of 2025. From 2025 to March 2026 average developer ai spent gone from $20/mo to at least $200. In my company it’s $1000-1500/mo to anthropic only.
In many cases this 20x+ increase over 3 mo.
This is going in other industries too. Claude cowork is just an example and beginning of the trend.
Show me any product that / company that charge you 20x and customers are happy , well its anthropic
I dont know what world you/I are living in. I do ask claude to enumerate/explain concepts i am not familiar with. I never approached the free tier limit (is there one?). At work, we have a webpage which ia basically a chat to different models, sometimes i use it
Would I be paying 20€ to ask those questions? I dunno, i dont feel any particular need. Would I be paying 200€?! Are you insane hell no
The question should really be what is the reservation price of existing buyers.
At some point the price will rise. But the value has to have risen for existing buyers to be ok with that - but can they perceive the value? Hmmm difficult to tell. Benchmarks are not an objective way to measure that.
In the long run google is most likely to acquire a serious cost advantage given their level of vertical integration.
We've been experimenting with claude code handling jira tickets and opening PR -- we're starting with Opus. It costs about $1 per PR that gets merged-- how much does it cost to have a software engineer do that PR? That's your price sensitivity. It will only get cheaper as models get more efficient and people get better at using them, though.
Managers of firms care about impact of financials. They don’t care about the metrics you are measuring / gaming. Ultimately all ‘progress’ has to show in the cash flows.
Are you taking more cost reduction projects and more revenue-generating projects? Are you actually delivering? Are customers perceiving you to be as trusted as before? Etc. are the only things that matter. ‘Show me the money’.
To me this is akin to the discussion re. Scrum, agile etc. Who cares? Show me the money.
My work pays for the $200 sub for me. At work in a typical 5h slot I tend to use 35~40% of the usage limit, so the $100 sub wouldn’t let me keep my current workflow and AI usage but I am more often than not using 2x the $100 sub. We’ll see if they keep paying for this or not, because we are not yet sure if using AI like this is sustainable and worth the tech debt of moving faster
But that is your employer paying a license (probably in bulk, I have no idea how much B2B sales cost for AI). My employer pays licenses for a lot of things that I probably wouldn't pay personally.
The person I was replying to said the average developer pays 200$. I call bullshit on that one.
This is all starting to smell like financial engineering games. Traditionally nobody in their right mind would give a startup billions to build data centers. For a long list of reasons, that’s kinda nuts.
However what it does allow all these companies investing to do is fund significant capital expenditure but hide it on their balance sheet. They all know if they funded capex directly it would create a deprecation storm that would tank their future earnings. Instead they give the money to another entity to do the building and magically it’s (the equity) now just an asset on their balance sheet with no deprecation. It’s “worth” a lot as a line item there, but only because the hype driving this financial engineering keeps the shares valuable.
Meanwhile the startup isn’t public and thus the fact that it has this massive deprecation on the books is mostly out of sight and out of mind, with some random sky high valuation that’s not based in any normal sense of business reality.
That all works great… until the bubble busts of course.
The business plan makes sense to me. They are a company that is focussed specifically on building AI data centers, which is a huge part of the economy at the moment. The big cloud players know about generic data centers, but there are likely big efficiency wins to be gained by specializing on AI. There is also the geopolitical angle: European countries (and others!) will likely trust a UK-based company more than one of the American BigCos. NVidia is a great partner and investor for them: NScale will buy billions worth of NVDA chips, and also send information and learnings about the unique needs of the market to the chipmaker.
That being said, financial engineering tricks like depreciation and tax sheltering are of course hugely important in the global economy. It's likely that NVDA has a lot of cash sitting in Europe that it doesn't want to repatriate because it would have to pay taxes on it.
It makes sense till it doesn’t. History tells us that suppliers funding demand like this tends to end badly.
> there are likely big efficiency wins to be gained by specializing on AI
I've seen this suggested before as well, but I don't think I've heard a lot of actual concrete things. What big efficiency wins are to be had specializing on "AI" datacenters as opposed to what the past mega hyperscalers have done? What techniques seem to be out there that cloud providers and others have slept on? What makes them so different, in terms of operating a datacenter?
I'm genuinely asking.
I know nothing about the inner workings of a large scale AI datacenter but I'd imagine the power and cooling requirements are more specialised than your average datacenter that mostly has to handle transmitting large amounts of data over the internet (not that that isn't computationally expensive, I just imagine LLMs (especially at the current scale of their deployment) are much more demanding)
> I'd imagine the power and cooling requirements are more specialised than your average datacenter
But are they actually doing things differently than the high compute parts of the hyperscaled datacenters? Are there radical new ways of distributing heat in the datacenter that only makes sense at that level of energy usage per square foot? Is AI energy use that much higher per square foot of other high-compute parts of datacenters, or is it just that its now something like 90% of the floor plan versus maybe only 50-60%?
> handle transmitting large amounts of data over the internet
I certainly can't speak for all datacenters, and I've never been in a hyperscaler datacenter. But of all the datacenters I've spent time in, the space for the outside network connectivity was rather small compared to the rest of the space for storage and compute. Think a few small office suites dedicated to outside networks coming in and connecting to the clients in the datacenter compared to a medium to large sized warehouse full of compute and storage.
There's "high compute", and then there's proper HPC. AI these days is way more on the HPC end of the scale. The GPUs are doing computations using 2-bit and 4-bit numbers and not 64-bit, but everything else is going to be comparable.
A few billion here and there are needed to keep the game going. Small cost of doing business, considering the alternative
Deprecation is the right way to spell it when it comes to GPUs, because the hardware becomes totally obsolete and uncompetitive after only a few years and you have to replace it wholesale. It's a slow-moving trainwreck.
>> This is all starting to smell like financial engineering games.
"Starting to"? :)
How does Nvidia's backing of this startup shape competitive dynamics in the AI infrastructure space? If Nscale relies on Nvidia for both capital and GPU supply, does that create an uneven playing field for alternative GPU vendors looking to get their hardware into large multi-tenant AI data centers? Is this $14.6B valuation mostly predicated on continued preferential access to Nvidia's GPU supply, rather than any unique technical or operational advantage that Nscale itself owns?
Seems like you already know the answer, mate.
My observation has been that the areas where data centers makes the most sense (colder climate, cheap energy, cheap real-estate, trustworthy countries) to build, are also areas where resistance against data centers has started to gain momentum. The spike in energy prices alone will make building these centers an uphill battle.
Good, if these multi billion dollar companies can't afford to build a data center that... At the bare minimum doesn't affect their cost of electricity over the next 10 years, nor the infrastructure, then it's totally reasonable for them to resist it. It's not like building data centers is for the common good like building new nuclear power plants or other key infrastructure. Instead it's literally just profit motivated, and not even by consumer spending.
I don't get the economics behind building AI DCs in the UK instead of a middle income country near the equator where there's plentiful solar or fossil fuels.
Labor and land is expensive, energy is scarce and expensive, and colocation is not that valuable because latency is dominated by compute instead of transmission.
But there must be a good reason I am missing.
good question, but there is a rational answer. remember if youre going to tie up a few billion in a project you want strong laws around ownership
singapore ran out of land. malaysia ran out of water and electricity
indonesia is unstable politically and just started confiscating foreign owned farms
india is too politically risky. people will just steal your shit
middle east - israel/iran chabad trump war etc. all within ballistic missile range, unstable
africa is africa
latin america no infrastructure (highways, power grid, etc)
labor is other bottleneck. it actually borderline impossible to bring in skilled engineers and technical specialist to remote energy rich areas
> africa is africa
Jeez what a way to dismiss the second largest continent and the second most populous landmass in the world.
I'd argue Africa is the best place to start such a thing. Cheap labour, plenty of renewable energy. The biggest issue it probably has is how little bandwidth it has, and might require additional sea cables.
But "africa is africa" is not a way reason to dismiss the continent.
> india is too politically risky. people will just steal your shit
India is a huge place and some parts of it are vastly better managed than others.
> latin america no infrastructure (highways, power grid, etc)
The governance problem is quite real in Latin America actually, but I think it may have potential. If some infrastructure gets funded in the process it would have beneficial side effects all around.
I'd imagine cooling? I could be wrong but a lot of the cost is there and putting what's effectively a megawatt scale electric heater certainly isn't going to make that any cheaper
Sovereignty, the days of peaceful geopoltics are behind us.
That's a public good that occurs if the government puts their thumb on the scale with subsidies or regulations, because otherwise the market can't get there itself (example: TSMC building plants in US only because of government incentive). But the Nscale case looks like private investors deciding this is the most profitable place to build it on the margin, which is anomalous.
[dead]
Huh, you think the equator is better for datacenters? I've never heard that.
I mean, there's a reason Luleå is a popular datacenter spot: access to hydro & wind- but also "free cooling"; datacenters create a lot of heat and having a cold environment makes it easier to maintain ambient temperatures that are sustainable with lower energy consumption.
Additionally, people tend to prefer their servers are in geopolitically stable countries that are less likely to be bombed or undergo civil war (and the US likes to ensure civil war in countries that have a lot of oil) - and you want your datacenters to avoid any environmental hazards like earthquakes.
I can't imagine preferring the equator for datacenters.
singapore/malaysia/indonesia one of the best places in the world for datacentres. just have to build absurd cooling systems and use coal power plants
the bottleneck is they keep running out of water for cooling
Singapore?
They were among the worst (and most expensive) datacenters I've worked with, but they were used because we were latency sensitive (game servers).
Haven't heard of anyone deploying to Indo/Malay though, maybe I missed something here.
Latency?
Does that even really matter with AI? If you already are waiting >1sec for a response/output, what is 0.2 seconds more?
I get what you're saying but do we really think the AI emanating from these data centers is always going to be limited to multisecond response times?
More cynically, someone in the UK might want to do something that makes money with these data centers once the AI thing goes bust.
Country risk..
It's not actually building many datacenters there right? It's just incorporated there.
[dead]
Unless nScale is going to help the Ellisons prop up their borrowing power to consolidate a media empire they're a little late to feeding frenzy.
Chances of this startup pulling a Theranos are? I mean data center construction is something that couple of non startup companies do and do it well. What is the problem that they solve? The article is quite light on what they actually do.
More WeWork than Theranos. Theranos was a fraud.
WeWork was a legit real estate business with a SaaS multiple on valuations.
Arbitrage most likely. They take money from big players, and probably outsource the different parts of construction, then they hold the capex on their balance sheets instead of the big players
[flagged]
You registered just to post this. Reconsider your choices.
[dead]
Chat, why do we hear so little about banks no longer funding mega data centers? Noise or signal?
I think it’s lagging indicator. There were signs of saturation in the end of 2025. From 2025 to March 2026 average developer ai spent gone from $20/mo to at least $200. In my company it’s $1000-1500/mo to anthropic only.
In many cases this 20x+ increase over 3 mo.
This is going in other industries too. Claude cowork is just an example and beginning of the trend.
Show me any product that / company that charge you 20x and customers are happy , well its anthropic
I dont know what world you/I are living in. I do ask claude to enumerate/explain concepts i am not familiar with. I never approached the free tier limit (is there one?). At work, we have a webpage which ia basically a chat to different models, sometimes i use it
Would I be paying 20€ to ask those questions? I dunno, i dont feel any particular need. Would I be paying 200€?! Are you insane hell no
People aren't paying $200 to chat, they're paying to have ClaudeCode or Cowork or Claude for Chrome do the work instead.
The question should really be what is the reservation price of existing buyers.
At some point the price will rise. But the value has to have risen for existing buyers to be ok with that - but can they perceive the value? Hmmm difficult to tell. Benchmarks are not an objective way to measure that.
In the long run google is most likely to acquire a serious cost advantage given their level of vertical integration.
We've been experimenting with claude code handling jira tickets and opening PR -- we're starting with Opus. It costs about $1 per PR that gets merged-- how much does it cost to have a software engineer do that PR? That's your price sensitivity. It will only get cheaper as models get more efficient and people get better at using them, though.
You’re operating in a micro perspective.
Managers of firms care about impact of financials. They don’t care about the metrics you are measuring / gaming. Ultimately all ‘progress’ has to show in the cash flows.
Are you taking more cost reduction projects and more revenue-generating projects? Are you actually delivering? Are customers perceiving you to be as trusted as before? Etc. are the only things that matter. ‘Show me the money’.
To me this is akin to the discussion re. Scrum, agile etc. Who cares? Show me the money.
Eh? Is the average developer paying 200$ a month?
I mean, I imagine some top% vibe coding bros paying 200$. I would require some serious evidence that the average developer pays for it.
I certainly wouldn't. It's moderately useful, but not 200$/mo useful.
My work pays for the $200 sub for me. At work in a typical 5h slot I tend to use 35~40% of the usage limit, so the $100 sub wouldn’t let me keep my current workflow and AI usage but I am more often than not using 2x the $100 sub. We’ll see if they keep paying for this or not, because we are not yet sure if using AI like this is sustainable and worth the tech debt of moving faster
But that is your employer paying a license (probably in bulk, I have no idea how much B2B sales cost for AI). My employer pays licenses for a lot of things that I probably wouldn't pay personally.
The person I was replying to said the average developer pays 200$. I call bullshit on that one.
People forgot the price of full msdn back in the day or any "SDK", Low code shit etc.
And I doubt someone can have the cognitive load to follow 10 Claude max. Let alone 1.
The "average" in the YC microcosm, I assume.