Amazon's $100B DC spend similar to entire Costa Rica GDP
- Reference: 1755173267
- News link: https://www.theregister.co.uk/2025/08/14/datacenter_investment/
- Source link:
News from a possible future: 'Rampant jellyfish cause AI outage by taking datacenter offline' [1]READ MORE
According to figures from researchers at [2]Omdia , the annual [3]datacenter capex of cloud giant Amazon alone now exceeds $100 billion, making it roughly comparable to the entire GDP of Costa Rica, and greater than that of Luxembourg or Lithuania.
For the other major cloud players, Google comes in with a capex spend of $82 billion, making it larger than Slovenia's economy, while Microsoft manages $75 billion, more than the GDP of Uganda. Meta's $69 billion means it is spending more than the output of Bahrain.
Omdia estimates that global datacenter capex will top $657 billion for the whole of 2025, which means the splurge has almost doubled since 2023 when the total was $330 billion.
These staggering stats are in large part due to companies' continuing desire for ever more compute power to drive AI development, in the expectation it will one day pay off.
[4]
This is despite a [5]report from consultants McKinsey & Company earlier this year that many corporate executives are skeptical about whether the large sums being spent on AI infrastructure today will actually produce any measurable return on investment in the near future.
[6]
[7]
And Meta [8]disclosed as part of its Q2 earnings that the conventional machine learning models powering its recommender systems are driving profits, not the company's generative AI projects.
Yet in a fine example of selling shovels during a gold rush, bit barn operators are happy to oblige clients who keep asking for more AI-capable infrastructure.
[9]
In its latest Cloud and Data Center Market Snapshot, Omdia says that, for the near term at least, orders for AI compute resources continues to outstrip supply. The development of newer and larger models, such as the [10]recently introduced GPT-5, is driving training demand.
At the same time, broader AI adoption is driving inferencing. Omdia claims that much of the population in developed economies is using AI in some capacity, with ChatGPT users alone topping 700 million, with more than 120 million daily visits.
Drilling down into that bit barn investment, Omdia says that IT kit will continue to be the largest item on the datacenter cost sheet over the next several years. Even so, spending on physical infrastructure will grow faster, driven by the need for new solutions in power generation, distribution, and thermal management to cope with all those hot and hungry AI servers.
[11]
Compute density in data hall racks is growing exponentially, calling for "significant and continuous innovation" in cooling and power delivery, while the energy required for all that IT infrastructure is likely to see operators invest in on-site power generation equipment and [12]microgrid-as-a-service . The latter is where a service company provides on-site electricity generation for the bit barn operator.
[13]Datacenter diplomacy: Australia commits to help Vanuatu build bit barns
[14]You've got drought: UK gov suggests you save water by . . . deleting old emails
[15]Colo operators flock to emerging markets to build DCs
[16]Arista pushes Ethernet for AI, downplays effect of tariffs
Looking ahead, Omdia says the ace of datacenter buildout is accelerating, and the capacity of the largest sites is expanding, with multi-gigawatt facilities - equivalent to the entire current capacity of a country like Canada – putting in an appearance.
These are already in the pipeline. Meta has signaled [17]several multi-gigawatt campuses scheduled to come online from 2026, while a recent [18]Deloitte Insights report claims some sites now in the early planning stages could top 5 GW. ®
Get our [19]Tech Resources
[1] https://www.theregister.com/2025/08/12/france_nuclear_reactor_jellyfish_shutdown/
[2] https://omdia.tech.informa.com/
[3] https://www.theregister.com/2025/07/31/amazon_earnings_q2_2025/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aJ4IFtJAbqbT_UXxyh5U5gAAAI8&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[5] https://www.theregister.com/2025/05/01/ai_dc_investment_gamble/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aJ4IFtJAbqbT_UXxyh5U5gAAAI8&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aJ4IFtJAbqbT_UXxyh5U5gAAAI8&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2025/08/01/meta_ai_investments/
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aJ4IFtJAbqbT_UXxyh5U5gAAAI8&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[10] https://www.theregister.com/2025/08/07/openai_gpt_5/
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aJ4IFtJAbqbT_UXxyh5U5gAAAI8&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[12] https://blog.catalystpower.com/more-power-to-you/how-does-microgrid-as-a-service-maas-work
[13] https://www.theregister.com/2025/08/14/nakamal_agreement_australia_vanuatu_datacenters/
[14] https://www.theregister.com/2025/08/12/uk_government_delete_emails_water/
[15] https://www.theregister.com/2025/08/12/emerging_markets_now_the_growth/
[16] https://www.theregister.com/2025/08/06/arista_pushes_ethernet_for_ai/
[17] https://www.theregister.com/2025/07/15/meta_datacenter_build_plan/
[18] https://www.theregister.com/2025/06/26/us_datacenter_power_crunch/
[19] https://whitepapers.theregister.com/
Good
Now someone needs to explain this to Rachel from accounts. Just imagine businesses wanting to spend money here into our economy, growing our economy and providing more tax money to support the govs out of control spending and debt.
I makes me wondering if I should be a new home PC ...
for a few £hundred seem positively parsimonious!
I will not, 12 years old but top spec at the time; the only upgrades have been bigger disks. It runs Debian trixie perfectly.
There area some questionable numbers there
El Reg said: "and the capacity of the largest sites is expanding, with multi-gigawatt facilities - equivalent to the entire current capacity of a country like Canada – putting in an appearance."
I have to question that claim from Omdia, assuming that's what they really meant. Canada's electrical generating capacity ranks 8th in the world by most figures. and 7th in terms of actual electrical energy produced (China, US, India, Russia, Japan, Brazil, Canada, in that order).
Canada in fact has several individual generating plants, nuclear and hydro-electric, which are bigger than the demand from the 5GW AI data centres mentioned in the story as being in the early planning stages. A new nuclear generating plant which is to be somewhere around 10 or 11 GW has already been announced to be built just east of Toronto to meet demand from things like electric cars and electric heat.
A 5GW giant data centre as mentioned in the story would require less than 3 per cent of Canada's current generating capacity, and most of that generating capacity is hydro-electric or nuclear, and so can actually deliver that around the clock, not just under optimal peak conditions as with say wind.
Now if they were going to claim that all AI data centres world wide put together, existing and planned, were to have a total cumulative demand equivalent to that of Canada's total output, then that might be a bit more plausible. However, that would still only amount to about 2 per cent of total electricity produced world wide. That's a lot, but not an inconceivable amount.
The real problem is more likely that the AI companies want to build their AI data centres in places with weak electrical infrastructure. They should instead be taking cues from the aluminum smelting industry and locating where the supply is.
More of the madness
Where does this loop end I wonder?
Is it just throw money at it, because, let's be honest here, these players have too much accumulated cash as it is.
Are these vanity spends? Does a single company need to be in control of, and spending more than a whole country?
Crazy times we live in I tell you!