Scroll Top

AI companies face massive energy headaches as datacenter build outs accelerate

WHY THIS MATTERS IN BRIEF

AI training is hugely energy intensive and not only is the grid struggling but companies building new datacenters are having issues securing the power they need.

 

Love the Exponential Future? Join our XPotential Community, future proof yourself with courses from XPotential University, read about exponential tech and trendsconnect, watch a keynote, or browse my blog.

Energy constraints are set to become the IT industry’s next major bottleneck, reckons Facebook co-founder and CEO Mark Zuckerberg. In a recent interview, he said he thinks that the long-running GPU drought is basically over, and Artificial Intelligence (AI) growth and development, such as increasingly massive training costs, won’t be reined in by capital constraints in the near term. Instead, energy issues are going to become the next major pinch point.

 

RELATED
The end of fossil fuels, Asia and Europe draw up a Global Energy Grid

 

Zuckerberg tells interviewer Dwarkesh Patel that with the GPU drought now over, businesses will be tempted to “invest a lot of money in building out these things,” like data centers. However, before capital becomes a question, industry is going to be confronted by energy constraints, reckons Zuckerberg.

 

 

The Future of Energy 2050, with Futurist Matthew Griffin

 

Providing some support to his assertion, the Facebook boss observes that a lot of new data centers are consuming 50 to 100 megawatts, with particularly large ones that “might be 150 megawatts.” It seems only a question of time before data centers start requiring 300 or 500 megawatts, or even pushing to a Gigawatt like Elon Musks’ Grok data center which could soon have over 300,000 Nvidia Blackwell B200 GPUs.

How long will the exponential AI training and energy curve keep going – Zuckerberg asks rhetorically. This question brings the topic of energy production into focus. The Facebook CEO highlights that energy production is probably worth investing in soon.

 

RELATED
Futurist Keynote Series: The Future of AI, Comms, DeepFakes, and Space!

 

Making a new power station is not a trivial task. Taking into account regulations (particularly with nuclear energy), powerline planning, and construction it can take many years from drawing up a plan to power flowing to the grid. In a timely article, the WSJ looks closely at the US state of Georgia and how it is coping with the opening of a multitude of tech and industrial businesses.

 

Energy will determine AI’s future not compute

 

Many of the new businesses “hoover up huge amounts of electricity,” notes the WSJ. At the same time, states are trying to lessen reliance on fossil fuel generated electricity, making attracting key businesses while assuring them of energy supplies a tightrope walk. Making things more complicated is the question of demand and supply affecting consumer prices and supplies.

Going forward, flexibility seems key, utilising diverse power sources from renewables like solar and wind, employing batteries for spike management, using nuclear where available, such as in Amazon’s case, and pragmatically using fossil fuel resources as required.

 

RELATED
Scientists discovered a way to read information directly from black holes

 

Interestingly, and relevant to the Zuckerberg interview above, the WSJ notes that Facebook is collaborating with a Georgia co-op and solar developer called Silicon Ranch, to help power its data centers.

Related Posts

Leave a comment

You have Successfully Subscribed!

Pin It on Pinterest

Share This