European datacenter energy consumption set to triple by end of decade
- Reference: 1729852326
- News link: https://www.theregister.co.uk/2024/10/25/eu_dc_power/
- Source link:
"In the last two decades, no technology has driven the need for accelerated power infrastructure development in Europe more than AI," analysts wrote in a McKinsey report [1]published on Thursday.
At the current rate of adoption, the consultancy estimates that European datacenter power consumption will reach 150 terawatt hours (TWh) by 2030 – up from about 62TWh today. If this actually pans out, bit barns could suck back up to five percent of the continent's electrical energy supply.
[2]
McKinsey estimates that European utilities will need to add approximately 25 gigawatts of generation capacity to power all the power-hungry GPUs being pressed into service training and running generative AI models and services – and most of it is going to need to be of the "green" variety.
[3]
[4]
That last detail, McKinsey analysts note, is something of a problem. "The datacenter industry faces a big challenge to decarbonize its footprint and reach net-zero targets on a 2030–40 timeline."
Some, including former Google CEO Eric Schmidt, have [5]argued that we shouldn't even be worrying about AI's energy appetite because the tech itself is our best shot at addressing growing the increasing prevalence of greenhouse gases.
[6]
"We're not going to hit the climate goals anyway because we're not organized to do it," Schmidt conceded while speaking at a recent AI summit in Washington DC.
Barring an AI climate salvation, McKinsey notes that renewable energy credits remain among the most popular ways cloud and datacenter operators go about offsetting emissions generated by their facilities.
But the consultancy also feels that such schemes have " minimal impact on long-term emissions in power systems and rarely incentivizes the development of new projects or the generation of clean energy."
[7]
One way to reduce AI’s impact is to place datacenters close to locations that produce clean energy. That approach can work best when these facilities house hardware used to train models – a process that requires tremendous amounts of computational power, but doesn't need to be physically close to population centers.
However, as the mix of AI workloads shifts toward inference – the actual running of the models – this will change and latency requirements will dictate the facilities be built near cities.
Long term, McKinsey is tracking a number of potential remedies – ranging from carbon capture to onsite power generation using small modular reactors. As we've [8]previously reported , such reactors have garnered a lot of attention from cloud providers and hyperscalers in recent months despite the fact none have actually been deployed commercially.
Nuclear isn't the only option for onsite generation – it's just one of the cleaner and more energy-dense items on the menu. Diesel generators, fuel cells, and facility-scale battery backups are commonly employed to provide supplemental power and smooth out fluctuations in grid power.
[9]Fujitsu delivers GPU optimization tech it touts as a server-saver
[10]AI's energy appetite has Taiwan reconsidering the nuclear option
[11]Tech giants set to pay through the nose for nuclear power that's still years away
[12]Amazon makes $500M bet on itty-bitty nuclear reactors to fuel cloud empire
While datacenter power consumption is expected to rise, a separate McKinsey [13]report also published Thursday suggests overall European energy demand may fall short of previous predictions. The firm believes that as much of 40 percent of the 460TWh in forecast growth may not materialize.
Growing populations and gross domestic profit were expected to drive power consumption across the continent by upwards of seven percent by 2030, according to McKinsey. Despite these trends, analysts observe demand has actually slowed – in large part because of energy efficiency gains, a shift to a services-oriented economy, milder winters, and high energy costs which have driven a trend toward de-industrialization. ®
Get our [14]Tech Resources
[1] https://www.mckinsey.com/industries/electric-power-and-natural-gas/our-insights/the-role-of-power-in-unlocking-the-european-ai-revolution
[2] 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=2ZxvApIV9VxBt4bCF0GpRYwAAAIA&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] 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=44ZxvApIV9VxBt4bCF0GpRYwAAAIA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%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=3&c=33ZxvApIV9VxBt4bCF0GpRYwAAAIA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.theregister.com/2024/10/08/eric_schmidt_speech/
[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=44ZxvApIV9VxBt4bCF0GpRYwAAAIA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] 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=33ZxvApIV9VxBt4bCF0GpRYwAAAIA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2024/10/18/big_tech_nuclear_power/
[9] https://www.theregister.com/2024/10/23/fujitsu_gpu_middleware/
[10] https://www.theregister.com/2024/10/22/taiwans_ai_chip_boom_sparks/
[11] https://www.theregister.com/2024/10/18/big_tech_nuclear_power/
[12] https://www.theregister.com/2024/10/16/amazon_nuclear_smr/
[13] https://www.mckinsey.com/industries/electric-power-and-natural-gas/our-insights/electricity-demand-in-europe-growing-or-going
[14] https://whitepapers.theregister.com/
Re: Sack the intern!
Funny you should mention Ireland as they were estimating the drain on their grid by data centres doubling over the next decade, and to counter this, they were insisting that any new data centre has to include on-prem green power to reduce the impact to the wider grid.
Think that was from a BBC article - but could have been from these very pages.
Either way, we're seeing this now: That Data centres are increasing base load on the grid meaning there's less capacity for the rest of society, and our wonderful UK Government has now suggested that Data Centres are 'essential services' so would be a priority for power should there be a short fall.
It's just one element of a tapestry of decisions that call into question the motivation for making all things electric when we're told the existing grid can cope, yet we have also been warned of rolling power outages - and yes, there have been power outages in this area over the past few years - and they've become more frequent over that time, too. So something isn't right: Either there's a lot more thefts hitting substations, more building work hitting mains power, or the grid is close to capacity already and yet more demand is being added.
Re: Sack the intern!
I'm sure the detail of any Irish on-prem green power requirements means "not very green at all". Solar's going to be a poor resource for any DC at these latitudes, on shore wind offers greater capacity factor but is still painfully intermittent. SMR aren't coming to the rescue any time soon, so it'll be fossil gas, and some Excel spreadsheets to show that somebody somewhere else has done something that offsets it all, honest.
As for the grid being close to capacity, it is in some areas, not in others. There's also a lot of stress put on the grid by renewables that fade in and fade out as nature sees fit. Been some very squeaky bum moments for National Grid in the UK caused by all those crappy solar farms that the morons of government think are wonderful.
Green?
good luck with that, Germany has just decommissioned a wind farm, because there is coal under it
Double that estimate
Don't forget all these electric cars we're being pushed towards.
Don't be surprised....
When the generators can't produce enough electrons.
- The price will rise especially to the domestic users
- We get power outages
But AI... this .... AI that ... AI Shit seems to rule.
You have been warned people.
Sack the intern!
"One way to reduce AI’s impact is to place datacenters close to locations that produce clean energy. "
Hardly. That only minimises grid losses, other than that very modest gain it has no bearing on the NET impact of any DC.
The only way a DC's power impact can be significantly reduced is if the operator builds, or pays for the building of new low emission generating capacity that wouldn't otherwise have been built, and whose construction doesn't supplant or block any other public grid renewable projects . Otherwise, even with an over-the-fence power connection all that's happening is that the DC uses clean power that would otherwise have been used by the rest of the economy. In much of Europe, gas CCGT is the swing power source, so if (say) Microsoft contract to buy power from an Irish windfarm that was being built anyway, then instead of that new output supporting extant industry, services and consumers, chances are their demand has to be met by the grid's balancing mechanisms, most likely from EirGrid calling power from the UK grid, where it will be coming from CCGT. And because the UK gas grid tops up with loss-heavy LNG, the net emissions will be considerable. Pretty much the same scenario across most of Europe.
Note: Not sure if it's the Reg intern, or the McKinsey intern - I'll plump for the latter.