跳到正文
The Guardian Technology· Jillian Ambrose Energy correspondent·· 15 小時前AI 評分68

IEA 報告警告全球數據中心碳排放到2030年將增加70%

Datacentres to create 70% more carbon emissions by 2030, IEA report warns

AI 導讀

IEA 預測,全球數據中心到2030年的碳排放將增加70%,用電量將增至目前的兩倍,AI 擴張佔用電增幅的四分之三。報告估算,AI 數據中心每年將額外消耗350 terrawatt-hours電力,超過英國全年用電量;到2030年,美國數據中心約60%用電將來自天然氣,中國 AI 數據中心約一半將來自煤電。

正文

The world’s datacentres will create 70% more carbon emissions within the next four years as the US and China turn to fossil fuels to power their AI growth agendas, according to the world’s energy watchdog.

The International Energy Agency (IEA) warned that datacentres will consume twice as much electricity by 2030 compared with today, with the global expansion of artificial intelligence accounting for three-quarters of this growth.

The Paris-based agency found that the breakneck pace of the industry’s growth means that within four years AI datacentres will consume an extra 350 terrawatt-hours of electricity every year, or more than the annual electricity consumption of Great Britain.

Its influential annual report warned that the world’s energy future was less certain after the impact of the Middle East war, which it has described as the greatest energy market disruption in history.

“I don’t see any credible pathway for the energy world to return to the way it was before,” said the IEA’s executive director Fatih Birol.

Instead, it will be reshaped by the policy decisions governments make after the crisis to meet the increasing demand for energy to power a growing global population, the rising need for air-conditioning and a booming AI economy, the IEA said.

A datacentre under construction at the Google Storey County campus in Sparks, Nevada, US.

A datacentre under construction at the Google Storey County campus in Sparks, Nevada, US. Photograph: Daniel Cole/Reuters

Datacentre developers have emerged as the world’s biggest corporate buyers of renewable energy to help fuel their servers which have a connection to local power grids. Despite this, the report showed that the appetite for fossil fuels is expected to grow even faster, particularly in the US and China.

By the end of the decade, around half of the electricity needed to power China’s AI datacentres will come from coal power plants, according to the IEA. The rest will come from renewable energy, with smaller amounts from nuclear reactors and very little from gas plants.

In contrast, the US is expected to rely on gas to provide about 60% of the electricity for its datacentres by 2030. There were plans for about 180GW of new gas-fired power plants linked to US datacentres in August this year, the agency said. That is almost as much new gas capacity as the total now installed across the European Union.

As a result of the growth in fossil fuel generation for datacentres, emissions associated with this sector were on track to rise by 70% by 2030, the IEA said.

The appetite for gas power among US datacentre developers is more than 16 times higher than in Canada, where the IEA identified plans for 11GW of gas generation to power their servers. In the rest of the world, AI developers have put forward plans for new gas plants with a combined capacity of just 4GW, the agency said.

This is because renewable energy is expected to play a greater role in supplying datacentres outside the US and China. In the EU, the IEA expects datacentres to rely on renewables for about 70% of their electricity by 2030 – followed by nuclear power, gas generation, and a very small amount of coal power.

Workers at TikTok’s datacentre in Hamar, Norway, finish construction of a huge steel building with exterior spiral staircases.

Workers at TikTok’s datacentre in Hamar, Norway. Photograph: Victoria Klesty/Reuters

The agency based its forecasts for the notoriously opaque industry on data showing the annual global shipment of AI servers, which have climbed eighteen-fold since 2020. It also used space-based imagery to track the buildout of 93 large “AI factories” around the world.

Its analysis suggests that the world’s datacentre capacity has climbed tenfold in the last 24 months to 17GW. By the time the AI factories were completed, their combined capacity was likely to be about 50GW, the IEA said.

The world’s existing stock of datacentres, including those with AI-optimised servers, now stands at 120GW, with 80% of this capacity in China, the US and the EU. A further 120GW will be built by the end of the decade under the IEA scenario often considered its central base case, of which three-quarters will be AI datacentres.

來源:The Guardian Technology · theguardian.com