Stopping the great AI energy squeeze will need more than data centres - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
FT商学院

Stopping the great AI energy squeeze will need more than data centres

Innovative solutions require joined up government, which is currently in short supply

Last month, the corporate behemoth that is Amazon celebrated its 20th birthday in Ireland. It should have been a joyous moment. After all Amazon, like other tech giants, has invested heavily in the country over the past two decades, partly due to its low tax regime, supporting heady growth.

But in reality these birthday celebrations had a sour tinge. One reason is that European courts ruled last month that €13bn of tax breaks given to Apple were unlawful. On a recent visit, I was told that local business leaders fear this could undermine future investment.

Another, more immediate, spoiler is energy. Amazon Web Services is currently rolling out €30bn of investments in Europe amid a boom in artificial intelligence, according to Neil Morris, its Irish head. But none of that bonanza is going to Ireland, because Amazon officials worry about future energy constraints. Indeed, there are reports that the company has already been rerouting some cloud activity because of this.

And while the Irish government has pledged to expand the grid, mostly via wind farms, this is not happening fast enough to meet demand. The water infrastructure is creaking too. Yes, you read that right: an (in)famously wet and windy country is struggling to sustain tech with water and wind power.

There are at least four sobering lessons here. First, this saga shows that our popular discourse around tech innovation is, at best, limited and, at worst, delusional. More specifically, in modern culture we tend to talk about the internet and AI as if it they were a purely disembodied thing (like a “cloud”).

As a consequence, politicians and voters often overlook the unglamorous physical infrastructure that makes this “thing” work, such as data centres, power lines and undersea cables. But this oft-ignored hardware is essential to the operation of our modern digital economy, and we urgently need to pay it more respect and attention.

Second, we need to realise this infrastructure is also increasingly under strain. In recent years the energy consumption of data centres has been fairly stable, because rising levels of internet usage were offset by rising energy efficiency. However, this is now changing fast: AI queries use around 10 times more energy than existing search engines. Thus the electricity consumption of data centres will at least double by 2026, according to the International Energy Agency — and in the US they are expected to consume nine per cent of all electricity by 2030. In Ireland the usage has already exploded to over a fifth of the grid — more than households.

Third, the scramble by companies and governments to work out how — or if — they can find this additional electricity has produced one unexpected blessing: tech has become a key driver of the energy transition.

Yes, surging electricity usage is raising emissions. But companies such as Google, Microsoft and Apple are investing heavily in hydro, wind and solar power and battery innovation. Microsoft even recently announced a deal with the Constellation utility group to invest $1.6bn to restart the Three Mile Island nuclear power station in Pennsylvania to meet AI electricity demand. Constellation’s market capitalisation has since jumped above $80bn because investors expect more such deals.

Meanwhile OpenAI’s Sam Altman and Microsoft co-founder Bill Gates are extolling the joys of small modular reactors. They and others in the tech sector hope that such moves will eventually reduce the energy squeeze, particularly if future versions of AI use less energy. If so, current fears about electricity supply might turn out to be misguided — just as predictions of a global famine were upended by the 1960s’ green revolution. Tech itself can solve tech woes — or so they hope.

However, the fourth lesson is that such an innovative energy solution couldn’t work without joined-up government policy. Sadly, that is in short supply. After all, you need planning permission to build data centres, which often means government intervention. Just look at how Angela Rayner, the UK deputy prime minister, is wading into a local fight in Abbots Langley in Hertfordshire, where locals want to block new digital investment.

You also need government involvement to create connected electricity grids. One huge impediment to the rollout of renewable energy in the US, for example, is that it is scandalously hard to get the permits needed to build transmission lines to connect renewable energy resources in the American heartlands to power-hungry places such as California.

And if the energy squeeze intensifies, we will also need government to adjudicate the future distribution of scarce electricity resources and to tackle questions such as whether households ought to get priority over business if the grid crumbles, and whether the state or Big Tech should fund innovation.

Libertarians — and many techies — might argue that market forces (ie prices) should determine the answers. But that vision is politically toxic, as Irish leaders know only too well. So brace yourself for energy battles across the industrialised world. It is not just AI’s existential future risks that we should worry about now.

gillian.tett@ft.com

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。

谷歌的卡武克丘奥卢:将抽象的AI思维转化为用户友好的产品

这家科技巨头的首席AI架构师兼DeepMind首席技术官讨论了Gemini 3大型语言模型以及向通用人工智能目标迈进的最新进展。

美国石油巨头埃克森美孚告诉特朗普:委内瑞拉“无法投资”

首席执行官达伦•伍兹反击总统要求仓促重返这个动荡国家的呼吁。
41分钟前

英伟达在中国的前景与亚洲的数据中心融资

黄仁勋表示,H200在市场上很有竞争力。但它不会一直保持竞争力。

东南亚在关税战中的韧性出乎意料

东盟国家应借助其韧性推进国内改革和贸易改革。

通用汽车披露因电动车转型与中国业务重组造成70亿美元冲击

总部位于底特律的汽车制造商称其“主动下调了电动车产能”。

中国AI芯片“三小龙”的真实火力如何?

三家新近上市的半导体制造商均处于亏损状态,且其销售额都不到最大竞争对手的十分之一。
设置字号×
最小
较小
默认
较大
最大
分享×