Much is written about the infrastructure and electricity needed to feed the AI beast. Recent research by Goldman Sachs estimated the cost of AI at $600 billion. Datacenter energy consumption will more than double by the end the decade – only five years away.
Meta’s nuclear datacenter plan reportedly ruined by bees.
READ MORE.
A report last year from the International Energy Agency said that the world’s bit-barns account for just under 1 percent of global consumption. This rises to 2 to 4 percent for large economies such as the US, China, Europe, and others. Datacenters are responsible for over 20 percent of electricity consumption in hotspots such as Irelandand this figure is still rising.
It’s not just datacenters. The IEA predicts that by 2035 electric vehicles will account for 6 to 8% of global electricity demand, up from 0.5 percent currently.
As access to energy is now seen as a constraint to sustained growth, many of the larger cloud- and datacenter-based companies have taken it upon them to find a reliable power source, and nuclear presents as an attractive option. Atomic generators are a great option for generating power around the clock, unlike renewables. They also emit almost no carbon dioxide, which is great if you want to be carbon-neutral, or even carbon-negative, in the future.
This month, a group including Amazon, Meta and Google backed effortsto ramp up global nuclear capacity, with the aim of doubling it by 2050. Microsoft had earlier been involved in a project to restart generation at Three Mile Island in Pennsylvania by 2028. Amazon also purchased a datacenter on the same site as Susquehanna Nuclear Power Plant, in Pennsylvania.
But the problem with many of these projects is that they’re too little or too late. Some of them are a decade away or more, and huge increases in energy consumption for datacenters are forecasted for the next few year. World Nuclear Association, for instance, acknowledges that a nuclear plant takes at least five to six years to build. Elsa Nightingale, Canalys’ principal ESG analyst, told The Register that
“Undoubtedly, nuclear energy will serve as part of the world’s energy mix for years to come,” is the year in which Canalys was founded. Even with grid expansions, Nightingale says that many electricity grids in the world are not able to meet the astronomical energy demand forecast for AI.
According to a report published last year, Americans could see their electricity bills rise by 70 percent by 2030 if they don’t take action to increase generation and transmission capacity.The report estimated that billions of dollars would need to be invested into energy grids in order to keep up with the demand from datacenters.
The management consultancy Bain & Company forecasted that power consumption will exceed supply in some places within the next few yearand that utility companies were slow to react as they faced flat or shrinking US demand for some time, which led them to prioritize efficiency and not scaling capacity or building new distribution lines.
If datacenters cannot generate enough energy to meet demand, or if AI infrastructure is not prioritized for efficiency, we could be in trouble. Nvidia’s Blackwell GPUs announced last yearconsume over one kilowatt each.
“Do we want to see a future where IT companies are hypothetically competing with hospitals for electricity?” Nightingale asked.
This was one of the scenarios predicted by energy infrastructure company Schneider Electric in earlier this year. The research published by Schneider Electric modeled four possible futures based on the way AI datacenters are managed. One of them points to energy conflicts between AI datacenters and other critical sectors, which could lead to an economic downturn or other negative outcomes.
- Nuclear center must replace roof on 70-year-old lab so it can process radioactive waste
- Amazon, Meta, Google sign pledge to triple nuclear power capacity by 2050
- Energy trio wants to pipe gas from coal mines to keep datacenter lights on
- Our world faces ‘unprecedented’ spike in electricity demand
“The more immediate challenge here is that most of the largest tech firms worldwide have net zero goals which become increasingly untenable as AI emissions rise at the rates forecast,” Nightingale told us.
Do you remember those carbon-neutral and carbon-negative goals that we mentioned earlier? Microsoft a href=”https://www.theregister.com/2024/05/16/microsoft_co2_emissions/”>admitted last year that its CO2 emission has risen by 30 percent since 2020 thanks to the building of more datacenters for AI workloads. Google also admitted that their emissions are up 48 per cent from 2019. AMD president Victor Pengsaid at the Hot Chips Conference last year that the tech industry must urgently focus on efficiency to address this issue. He explained that semiconductor firms should focus more on making infrastructure more efficient.
The solution to this thorny issue is likely to come in a variety of directions, including bringing renewables like wind and solar into the grid to feed it, making energy grids stronger, and datacenters producing some of their power, even if that means gas turbines for the short term
“So, nuclear energy, while an interesting piece of the puzzle, avoids the real problem entirely: AI is not only thirsty for water requiring vast amounts for cooling and arguably more challenging to resolve, but it is also hungry for energy,” said Nightingale.
“Instead of looking to nuclear as a silver bullet, tech companies must first find ways to dramatically reduce AI’s electricity and water demands, before they can credibly claim their AI investments and net zero goals are tenable.” (r)
