News: 1744626728

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Dot com era crash on the cards for AI datacenter spending? It's a 'risk'

(2025/04/14)


Interview Those who ignore history are destined to repeat mistakes of the past and, with signs of an inflating bit barn spending bubble, comparisons are being made with the infamous dotcom bust a quarter of a century ago.

Rather more recently, Synergy Research Group said in January that mergers and acquisitions of datacenter businesses throughout 2024 [1]swelled to record highs of $73 billion , and yet more investors are lining up for a piece of the action.

Microsoft puts $1B US datacenter builds on hold amid AI, tariff uncertainty [2]READ MORE

Having lived through the early days of the internet frenzy, Fabrice Coquio, senior veep at Digital Realty, which bills itself as the world's largest provider of cloud and carrier-neutral datacenter, colocation and interconnection services, is perhaps better placed than most to venture an opinion. Is there a bubble?

"I have been in this industry for 25 years, so I've seen some ups and downs. At the moment, definitely that's on the very bullish side, particularly because of what people believe will be required for AI," he tells The Register .

Grabbing a box of Kleenex tissues, he quips that back at the turn of the millennium, if investors were told the internet was inside they would have rushed to buy it. "Today I am telling you there is AI inside. So buy it."

[3]

"Is there a bubble? Potentially? I see the risk, because when some of the traditional investments in real estate – like housing, logistics and so on – are not that important, people are looking to invest their amazing capacity of available funds in new segments, and they say, 'Oh, why not datacenters?'"

[4]

[5]

He adds: "In the UK, in France, in Germany, you've got people coming from nowhere having no experiences… that have no idea about what AI and datacenters are really and still investing in them.

"It's the expression of a typical bubble. At the same time, is the driver of AI a big thing? Yes… [with] AI [there] is a sense of incredible productivity for companies and then for individuals. And this might change drastically the way we work, we operate, and we deliver something in a more efficient way.

[6]

[7]Microsoft , [8]AWS and [9]Google have all sunk billions of dollars into AI, throwing money at promising startups to try to corner the market, and spending on [10]infrastructure to train models or sell the software to customers. So far, it looks as though the market is [11]top heavy and shareholders are starting to get twitchy.

Coquio says: "Is it going to be quick? I don't think so. I've seen the time needed for cloud. We've got more than 10 years of experience of cloud and cloud is now, today, mature."

The arrival in January of Chinese startup [12]Deepseek indicates that the AI industry – one of the major causes of accelerating datacenter builds – may be built on quicksand. Deepseek made unverifiable claims about the cost of training its models and their performance. The stock market for American corporations operating in AI [13]tanked as investors asked if buying expensive GPUs was the best - or most cost effective - way to create a large language model.

[14]

Unperturbed, the biggest funders keep on believing, because at this point they've invested so much, and keeping faith is better than the alternative. Even Alibaba, which [15]recently warned the datacenter market is overheating with too many new sites being constructed, is still going to plough billions into new bit barns.

According to McKinsey, the "future demand" for dataceter capacity is likely to hinge upon multiple factors that are "still hard to accurately determine".

It [16]says : "The pace of adoption of advanced-AI use cases will certainly count, but so too will the mix of different types of chips deployed and their associated power consumption, as well as the balance between cloud and edge computing for AI workloads and the typical compute, storage, and network needs of AI workloads."

To cover its back, McKinsey has painted different potential scenarios. It mostly expects global demand for bit barn capacity to expand between 19 to 22 percent annually from 2023 to 2030 "to reach an annual demand of 171 to 219 gigawatts (GW)."

"A less likely yet still possible scenario sees demand rising by 27 percent to reach 298 GW. This contrasts with the current demand of 60 GW, raising the potential for a significant supply deficit. To avoid a deficit, at least twice the data center capacity built since 2000 would have to be built in less than a quarter of the time."

Manoj Sukumaran, senior principal analyst for data center compute and networking at Omdia, is forecasting the server market to hit $290 billin in 2025, up from $230 billion last year. "This growth will be driven by the expansion of accelerated computing for AI."

His colleague, Alan Howard, who covers datacenters and colocation services at Omdia, agrees, saying the hyperscalers and the broader colocation data center industry "will continue to grow aggressively reasonably unhampered.

"There is a growing sector of property developers chasing GW+ campus projects and that will be a market to watch for any kind of oversaturation. This will take time as master-planned energy campuses are typically a multi-year infrastructure project."

Over at Synergy Research Group, chief analyst John Dinsdale, doesn't buy into any comparisons with the dot com era or that the AI and AI DC sectors are too inflated.

"With AI most of the applications and monetization are real. The leading cloud AI segments are growing by over 100 percent per year and generating some serious money. The dotcom bubble was caused by a frenzy of unrealistic plans and expectations that suckered in investors - many of which should have known better. It was an extreme example of herd mentality causing irrational exuberance."

Offering something of a different viewpoint is John David Lovelock, distinguished VP analyst at Gartner, who told us a few months back that the business value from AI remains elusive for many businesses.

"Our expectations for what generative AI can and will do are starting to come down."

Digital Realty does have [17]skin in this game and Coquio agrees AI is an important driver for the datacenter industry, but he says it's not the only one with connectivity, content distribution, satellite comms, digital media in general and, of course, the cloud all playing a part.

"We are constantly deploying new cloud capacities for Microsoft, Oracle, French companies, German companies, almost every day. AI is another layer. We do serve normal enterprises which are just doing the same thing that they were doing 25 years ago, removing on-prem equipment and putting that in [our] datacenters, so there are constant requirements which are still the same."

That's not to say Digital Realty isn't feeling the impact of wider changes. "We are generalist, meaning that we provide one rack up to 50/60, megawatts for a single purchase holder. There are 20 companies in the world representing, at the moment, 60 percent of the bookings [across the industry.]"

This score of hyperscaler customers is made up of 17 US businesses and three in China, including [18]AWS , [19]Microsoft , [20]Meta , Alibaba and so on.

"Most of them have completed their time to market initiative to be in multiple areas where they needed multiple subcontractors for colocation services. They are now engaged for many years already in reducing the number of partners, either because they build more and more themselves, or when they need absolutely a collaboration provider, providing connectivity, concentration of customers and so on, they reduce that to the bare minimum," Coquio tells The Register .

[21]Datacenters are hotter than ever, and we're not talking about rack temperatures

[22]Alibaba exec warns of overheating AI infrastructure market

[23]Amazon to cough $75B on capex in 2024, more next year

[24]AI spending spree continues as Microsoft commits $80B for 2025

With such power hypercalers are demanding that colo providers dealing in commodity services cut prices. Only the biggest will survive, perhaps, or they will get niche to make better profits in countries that are already more developed.

"There might be some surprises in the US, in Europe, not really in Greece, the market is just developing, but in the flat markets. So England, Germany, France, Holland. There might be some movements at a certain point," he says, adding: "There will be some projects not being really built."

With the high-flying datacenter industry's head in the clouds, maybe there's more than just over-hyped AI tech that could cause it to come crashing back down to Earth. What goes up, typically comes down. The thing that matters is how rapidly and when, so all those involved can at least try to be prepared for the bumpy landing. ®

Get our [25]Tech Resources



[1] https://www.theregister.com/2025/01/29/exploding_datacenter_market/

[2] https://www.theregister.com/2025/04/09/microsoft_puts_more_datacenter_builds/

[3] 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=2Z_0xC4Ob-PiwZXnJL85OwAAAAEM&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[4] 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=44Z_0xC4Ob-PiwZXnJL85OwAAAAEM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] 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=33Z_0xC4Ob-PiwZXnJL85OwAAAAEM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[6] 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=44Z_0xC4Ob-PiwZXnJL85OwAAAAEM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[7] https://www.theregister.com/2025/04/07/microsofts_ai_strategy/

[8] https://www.theregister.com/2025/01/08/amazons_latest_investment_is_11b/

[9] https://www.theregister.com/2024/04/17/google_deepmind_funding/

[10] https://www.theregister.com/2025/04/11/ai_revolution_driving_datacenter_network/

[11] https://www.theregister.com/2024/11/29/microsoft_preps_big_guns_for/

[12] https://www.theregister.com/2025/01/26/deepseek_r1_ai_cot/

[13] https://www.theregister.com/2025/01/27/tech_stocks_tank_as_us/

[14] 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=33Z_0xC4Ob-PiwZXnJL85OwAAAAEM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[15] https://www.theregister.com/2025/03/25/alibaba_ai_infrastructure/

[16] https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/ai-power-expanding-data-center-capacity-to-meet-growing-demand

[17] https://www.digitalrealty.com/resources/articles/private-ai-exchange-ecosystem

[18] https://www.theregister.com/2024/11/01/amazon_75b_capex/

[19] https://www.theregister.com/2025/01/06/ai_spending_spree_continues_as/

[20] https://www.theregister.com/2025/01/24/meta_ai_spending/

[21] https://www.theregister.com/2025/01/29/exploding_datacenter_market/

[22] https://www.theregister.com/2025/03/25/alibaba_ai_infrastructure/

[23] https://www.theregister.com/2024/11/01/amazon_75b_capex/

[24] https://www.theregister.com/2025/01/06/ai_spending_spree_continues_as/

[25] https://whitepapers.theregister.com/



Business case

Anonymous Coward

When projects cannot point to a business case for their investments, I would hesitate to invest.

During the dot.com bubble, it was not so much the Internet that didn't deliver, but 99% of the investment opportunities were unable to come up with a way to earn money. They didn't known it when they started, and they didn't find it when they had to deliver.

I see the same now. Massive amounts of money are invested into AI projects that have no clue who will pay them for what products. Some will find a market and become the next Google or Amazon. the other 999 (or 9999+) will vanish. Most of the big data centers will go the way of SUN when their gear is not required anymore.

Re: Business case

Like a badger

" I would hesitate to invest"

Worth noting that many early stage investors are not in it for the long term, and really, honestly don't care about the long term viability (or otherwise). All they want is to invest a shed-load of cash in a startup run by a handful of geeks, and then to offload that as soon as possible in exchange for the three to ten shed-loads of cash. Obviously this requires a bullish market, and a good supply of credulous mid-stage investors who haven't spotted that most AI investments will go to the wall.

Easy money! No business plan required, no credible product required, no customers required. Which makes me wonder if Liz Holmes is out of clink yet, because her expertise in this area would be useful? Maybe the Orange Felon can issue a presidential pardon - come on MAGA loons, communicate with his Orangeness on "Truth Social" and make it so!

Actual use cases?

heyrick

There are some use cases for "AI", like my camera's intelligence where it mostly correctly recognises things and sets up the shot accordingly. Or the photo editor where I can remove things in a way that isn't too horrible - including entire people. It's not perfect but it can do a better job that I could in a fraction of a second.

For the LLMs, it's a toy mostly. Various images from textual prompts that might sort of resemble what you envisaged. It's a quick way to make little images to drop into blog posts to jazz things up a bit. Or a chat bot that sort of gives an answer in the ballpark of correct, if not actually correct. But, then, a few minutes trawling StackOverflow would result in much the same.

It's a fun thing to play with, but I use free offerings with the implied restrictions because fun doesn't justify paying because it isn't a necessity. Therefore, what's the actual business case to justify the vast amounts being spent on hardware and electricity?

I feel like Max Bialystock

zimzam

Yes… [with] AI [there] is a sense of incredible productivity for companies and then for individuals

You keep saying that but you don't tell me how!

Apples and Oranges

SparkE

Although there are a lot of parallels, I wouldn’t compare Ai to the Dot Com Bubble.

During the Dot Com Bubble era, The power was in the hands of a much broader demographic of people. It was much more democratized. People were using the internet to create.

Ai tech is ruled and powered by Big Businesses and their goal is deception (machines pretending to be human) and manipulation to enable monetization. Including replacing human workers with Ai whenever possible. Also, no one was forcing people to adopt the internet or use a specific site or tech back then. Adoption was organic.

In this Ai “situation” we are finding ourselves in, much of the public is actively rejecting the use of Ai. So much so that companies like Google are seeing that not many people were willing to pay for their Ai as a stand alone product. So rather than let the product-that-no-one-wants die on the vine through free market natural selection, they force it on its Google Workspace Subscribers and raise their subscription rates. It’s also forced on the free tier so that Google can essentially use Ai to Hoover-Up more data stored in user’s Spreadsheets and stored documents. This is the way of things now. Demand isn’t coming from the consumer, it’s coming from the supplier.

I am all for using Ai trained on vetted data to use for scientific research and medical diagnosing, etc. But training Ai on garbage internet data in an effort to imitate humans and manipulate them accordingly to make a profit is a bubble that I hope pops before dividing the world into two groups; the ultra rich and the ultra poor.

Re: Apples and Oranges

Will Godfrey

While I agree with you overall, I would note we already have the ultra rich and the ultra poor. However, I think the number of the former is shrinking while the number of latter is definitely expanding.

Re: Apples and Oranges

ecofeco

Market hype is market hype. Bubbles are bubbles.

Often it's wrong to compares apples to oranges, but bubbles in the apple market or orange market are still bubbles. Splitting hairs does not change that.

As for using it for oppression and deception and dependency, yeah, we do not need more of that from the tech douche bros

Anonymous Coward

There's going to be a massive boom in European data centres, but not because of AI, it's because of Trump.

Europe is slowly waking up to the realisation that it's entirely dependent on US organisations for cloud compute, a situation that will have to be undone rapidly as Trump can, and will, shut off whomever he wants. Are you running your infrastructure on AWS and you're in a fight with Trump? All it takes is for the US administration to issue sanctions against your organisation and insist on enforcement through the courts for your entire infrastructure to come crashing down.

Expect over the next 3-6 months to see european governments and organisations ban purchasing critical supporting services from US based suppliers. It's a risk that can no longer be hidden.

WHOCOULDAKNOWED?!

ecofeco

Many of us have seen this movie before. Many times. We know how it ends.

Spoiler alert for the sweet summer children born yesterday: it's the same ending every time.

System going down in 5 minutes.