Stopping the great AI energy squeeze will need more than data centres - FT中文網
登錄×
電子郵件/用戶名
密碼
記住我
請輸入郵箱和密碼進行綁定操作:
請輸入手機號碼,透過簡訊驗證(目前僅支援中國大陸地區的手機號):
請您閱讀我們的用戶註冊協議私隱權保護政策,點擊下方按鈕即視爲您接受。
FT商學院

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

版權聲明:本文版權歸FT中文網所有,未經允許任何單位或個人不得轉載,複製或以任何其他方式使用本文全部或部分,侵權必究。

川普和海湖莊園的力量

這位前房地產開發商非常瞭解如何將建築和空間有效地用作宣傳。

爲2024年的世界感到高興的十個理由

從巴黎聖母院的修復到《抑制熱情》的大結局,這一年其實並不算太糟。

2025年德國大選:主要的競選承諾是什麼?

各大政黨提出了截然不同的計劃,以重振歐洲最大經濟體的命運。

「市場恐慌」:巴西財政赤字導致貨幣跌至新低

總統在面臨其第三個任期內的最大挑戰。

川普過渡團隊尋求在「第一天」讓美國退出世衛組織

美國的迅速退出將使全球衛生機構失去主要資金來源,並削弱其應對緊急情況的能力。

谷歌推動重新確立人工智慧領域的領先地位,提振了投資者信心

在經歷了過山車般的一年之後,人工智慧和量子計算領域的一系列突破帶來了轉機。
設置字型大小×
最小
較小
默認
較大
最大
分享×