WATER POLITICS

Artificial Intelligence (AI) has forever changed the way we search the internet – for most of us it’s made our lives a lot easier. But there’s a hidden cost, if AI is powered by fossil fuels, it could end up killing us instead. Did you know that adding generative AI to Google’s search consumes 4x or 5x more energy per search at minimum? So, will AI make our lives more efficient? Or will it just turn out to be power guzzlers?

Given that our new report shows global ICT accounts for 2-4% of global greenhouse gas emissions, if left unchecked can blow up to 23% in the coming decade, adding AI chatbots will only make it worse. So, let’s get on top of just how power hungry and thirsty AI tools can really be before it’s too late…

Search engines are energy intensive, but adding AI will increase this by 4x or 5x…

Google’s search engine is the internet’s most visited website, facilitating over 8.5bn searches every day. However, it’s also responsible for emitting 1.28mn tonnes of CO2 every day – the same as Laos’s yearly emissions.

Google search emissions every day = Laos’s yearly emissions…

…adding AI to Google = annual emissions of Albania & Macau combined in 2021

But if you think Google searches are bad, according to Martin Bouchard, cofounder of the Canadian data center at QScale, adding generative AI to its search engine will require “at least four or five times more computing per search” at minimum – the annual emissions of Albania and Macau combined in 2021.

The energy jump is so substantial that developers of ChatGPT have intentionally stopped training its model to learn information past 2021 to cut down on computing requirements. And that’s just one AI tool, there’s also AI chatbots…

AI chatbots are extremely thirsty too – ChatGPT drinks 500mL for every conversation…

If someone’s talking about AI it’s most likely chatbots. However, along with its huge computational power, chatbots drink staggering amounts of water to cool themselves down. According to a recent study, ChatGPT drinks 500mL of water for every 20-50 Q&A conversation you have with it. That’s 20x more water compared to 50 Google searches.

If all of ChatGPT’s users had a conversation = it would consume 20 Olympic-sized swimming pools…

…however, equivalent in Google searches = 1000x less water

It might not seem like much, but ChatGPT was the fastest growing app of all time, with over 100mn users in just two months after its launch. If all its users each had a conversation, ChatGPT would consume 50mn liters of water – the same as 20 Olympic-sized swimming pools. However, equivalent in Google searches, 1000x less water would be consumed.

So, imagine when all the other tech companies jump on the AI chatbot hype…

Training/developing AI tools is also power hungry & thirsty…

Each AI tool undergoes rigorous training before it’s launched to the public. Open AI reported that training ChatGPT released 502 metric tons of CO2 – the equivalent of roughly 610 one-way direct flights from New York to Paris.

If all 2,500 AI developers launched chatbots = annual emissions of Macau in 2021 (from training)

If all 2,500 chatbot developers listed on The Manifest, developed their own chatbot that could be an additional 1.3mn metric tons of CO2 emissions (just from training) – the annual emissions of Macau in 2021.

A recent paper also calculated that training GPT-3 consumed 700,000 liters of freshwater – enough to produce 370 BMW cars. Worse still, water consumption would have tripled if training took place in Microsoft’s Asian data centers.

More usage/larger models + more developing/training/upgrading = AI explosion

Whilst utilising AI may lower the number of traditional searches, in reality an AI query requires tons of navigating through far larger amounts of data, burning a lot more energy than conventional software. And let’s face it, just because we’re using AI doesn’t mean we’re going to be searching less.

On top of that, the WSJ reveals there is an “insatiable demand for running large language models right now” as new generative AI models can be anywhere from 10x to 100x bigger than older AI models.

According to CWR’s report, global ICT accounts for 2-4% of global emissions…

…but if left unchecked can blow up to 23% in the coming decade

Moreover, if all tech companies decided to develop/train/upgrade their AI models, the AI explosion could be dire especially if data centers are powered by fossil fuels. According to our latest report, the global ICT sector “is estimated to account for 2-4% of global emissions.” But if left unchecked, “the sector’s share of global emissions can blow up to 23% in the coming decade.”

Clearly the ICT sector/AI developers need to fast track net zero before emissions get uncontrollable.

Carbon neutrality is an intermediate step, ideally we must decarbonise the whole sector

CWR’s report finds that China’s top ICT listco’s lag their NASDAQ counterparts. Indeed, Facebook (now Meta Platforms), Apple, Netflix, and Google (now Alphabet) are at net zero for operational emissions…

Our report highlights that not all ‘net zero’ means the same…

…instead of offsetting emissions through carbon credits like Netflix, tech co’s should aim to operate on carbon free energy 24/7 like Alphabet

However, the report also highlights that not all ‘net zero’ means the same. Rather than offsetting all-scope emissions through carbon credits like Netflix, tech companies should really aim to operate on carbon-free energy 24/7 like Alphabet which is set to achieve this by 2030.

That being said, the report acknowledges that “some emissions cannot be avoided and carbon credits may have to be used – here it is important to ascertain the “quality” of the carbon credit to avoid being accused of greenwashing.” Especially with the recent Verra controversy, in which over 90% of their rainforest offset credits did not represent actual carbon cuts – the Guardian breaks down why.

With the rise of China’s AI chatbots, China ICT needs to fast track transition – ASAP!

Just last month, China’s tech giants launched their AI chatbots – Baidu came out with ‘Ernie Bot’ and SenseTime with ‘SenseChat’. And earlier this month, Tencent released a teaser for the launch of its AI chatbot. There’s no doubt that soon there’ll be a lot more chatbots released as we all adopt a new way of searching.

Before AI explodes, electricity from data centers must be run on RE – see how 5 ICT listco’s can lead the way!

But before this explodes, we must make sure the electricity from data centers is run on renewable energy. According to our report, Scope 2, which is indirect emissions from purchased power and heat, for China’s 5 largest ICT companies listed on the Hong Kong Stock Exchange – Alibaba, Baidu, China Mobile, Tencent & Xiaomi (together the “HSI 5”) were 41MtCO2e – this is 1.2x Hong Kong’s annual GHG emissions.

Cutting these emissions are totally doable – don’t know how? See how 5 ICT listco’s can lead the way!

---

Further readings

• Green Clouds One Day – How does watching online videos exacerbate our climate crisis? Are big tech brands like BAT and FAAG doing enough to source more renewable energy? CWR’s Chan reviews the landscape – it looks like we will see “green clouds” one day

• Greenie Web: Decarbonizing Websites – Greenie Web’s program analyses a website to recommend near instant decarbonization actions. Ian Chew, Founder of Greenie Web expands on the program & talks the future of technological sustainability

• Tis The Season To Be Worried: Our Online Habits– Did you know your social media and streaming are exacerbating the climate crises? CWR’s intern Lam shines a light on how we need to change our online habits & cultivate a sustainable digital lifestyle

• It’s Time to Tweak Yourself to Save the Planet– CWR’s McGregor gives a sneak peek into CWR’s upcoming report that shows how simple tweaks to our habits from transport to food and shopping to being online matter for the climate