The financial results of the world’s biggest three hyperscale cloud companies have seen Amazon, Google and Microsoft all credit growing customer demand for artificial intelligence (AI) services for boosting their respective profit, revenue and market share totals.
The downside to this growth, particularly where Google and Microsoft are concerned, is that it appears to be derailing their carbon reduction commitments, with both firms disclosing sizeable gains in their greenhouse gas (GHG) emissions in their recent annual environmental reports.
Microsoft published its 2024 Environmental sustainability report in May 2024, and revealed that, despite pledging to become a carbon-negative entity by 2030, the company’s GHG emissions for 2023 were 29.1% higher than its 2020 baseline.
The report attributed this rise to a 30.9% increase in the company’s indirect Scope 3 emissions, generated in part by Microsoft’s efforts to expand its global datacentre footprint.
As previously reported by Computer Weekly, Microsoft pledged in December 2023 to invest £2.5bn over the next three years in the UK to double the size of the datacentre footprint needed to meet the demand for its growing portfolio of AI services. “In 2023, we saw our Scope 1 and 2 emissions decrease by 6.3% from our 2020 baseline [and] this area remains on track to meet our goals,” the report said.
“The rise in our Scope 3 emissions primarily comes from the construction of more datacentres and associated embodied carbon in building materials, as well as hardware components such as semiconductors, servers and racks.”
The company went on to say the situation is “unique to our position as a leading cloud supplier that is expanding its datacentres”, but will be relatable to other firms who are struggling to keep tabs on their Scope 3 emissions. “We reflect the challenges the world must overcome to develop and use greener concrete, steel, fuels and chips,” said Microsoft. “These are the biggest drivers of our Scope 3 challenges.”
As it turns out, the challenges Microsoft is facing are far from unique, as Google also cited an increase in datacentre energy consumption as a factor in why its 2023 GHG emissions were up 13% on the previous year.
The disclosure features in the internet search giant’s 86-page 2024 Environmental report, which it published in early July 2024.
Its contents suggests the rise in datacentre energy consumption the company has experienced is linked to the growing demand for compute-intensive AI services.
“AI has been deeply integrated into our products for years, and we’ve invested heavily in improving the efficiency of our AI models and infrastructure,” the report stated.
“While these efforts have helped mitigate some of AI’s environmental footprint, the rapid advancement of AI has brought necessary increased attention to its energy consumption and resource demands.”
To this point, the report states that Google’s total datacentre energy consumption levels grew by 17% during 2023, while maintaining its pledge that its server farms would be powered by 100% renewable energy.
“As Google’s infrastructure continues to power the digital transition, providing numerous economic benefits across the globe, we expect this trend to continue in the future,” the report continued.
At the same time, the company said its growing infrastructure represents an opportunity to drive the “innovations and investments needed to power a low-carbon economy.”
The report added: “Overall, our total GHG emissions increased by 13% – highlighting the challenge of reducing emissions while compute intensity increases and we grow our technical infrastructure investment to support this AI transition.”
Similar to Microsoft, Google is working towards 2030 being the year by which it becomes a confirmed carbon-neutral entity.
While both companies are working towards similar sustainability pledges, they have also been respectively talking up the growing customer demand they are seeing for AI in their financial results.
Surely both firms should have anticipated ahead of time that building out the supporting datacentre infrastructure for their AI endeavours would lead to an uptick in GHG emissions?
“Absolutely they should have, and likely did, expect the increase in admissions from AI,” Stephen Old, head of FinOps at independent software licensing advisory Synyega, told Computer Weekly. “Every metric you look at in computing and emissions points to the fact that AI is going to quickly outweigh emissions from other areas of cloud computing because it uses many times more energy for compute than other forms of computation.”
Remedial action
Both Google and Microsoft have stated a commitment in their financial reports to taking remedial action to ensure their 2030 carbon reduction commitments remain on course.
Microsoft stated in its report that it has launched a company-wide initiative to pinpoint the additional measures it needs to take to bring about a reduction in its Scope 3 emissions, which has led to the development of 80 “discrete and significant” actions it can take.
“[These include] a new requirement for select, high-volume suppliers to use 100% carbon-free electricity for Microsoft goods and services by 2030,” the report stated.
Google, meanwhile, acknowledged in its report that it expects its total GHG emissions to rise again, before dropping towards its emissions reduction target.
The company’s Scope 1 emissions were down 13% on 2022, while its Scope 2 emissions were up 37% year-on-year and its Scope 3 emissions increased 8% over the same period.
“Our total Scope 3 emissions increased by 8% due to increases in emissions generated from goods and services purchased for our operations, upstream emissions from purchased electricity, and emissions related to datacentre construction,” its environmental report stated.
“We expect our Scope 3 emissions will continue to rise in the near term, in part due to increased capital expenditures and expected increases in our technical infrastructure investment to support long-term business growth and initiatives, particularly those related to AI.”
Decarbonisation roadmaps
To mitigate this, the company said it started to work with its largest hardware manufacturing suppliers by spend in 2023 to develop decarbonisation roadmaps that were geared towards lowering its GHG emissions.
It also requested that these same suppliers agree to commit to “achieving a 100% renewable energy match by 2029”, it continued.
“We’re also working directly with suppliers of hotspot commodities – or commodities with disproportionately high emissions – to identify and collaborate on carbon reduction initiatives that support our own emissions reduction target.”
Google’s acknowledgement that its GHG emissions are likely to get worse before they get better also serves to partially highlight the quandary that the cloud providers find themselves in, said Old.
“In terms of mitigations, they are in a catch-22 situation. They know [AI] is a way to lock customers in and make more money, and also that customers are going to ask for it regardless, so they have to offer [AI] or they’ll go somewhere else,” he said. “They also know it’s going to increase emissions. They are simply choosing the money over the emissions.”
Speaking to Computer Weekly, Rich Gibbons, head of IT asset management (ITAM) market development and engagement at Synyega, said with Microsoft’s generative AI [GenAI] offerings, such as Copilot, being keenly adopted across various industries, it may also find it a struggle to keep a lid on its GHG emissions.
“It is unlikely emissions will reduce [for Microsoft], as increased usage of products such as Copilot, Azure OpenAI and ChatGPT will continue to produce more emissions,” he said. “And should use continue to grow, that may well kick-start a new round of datacentre building, too. Perhaps the only real way for organisations such as Microsoft and Google to reduce their emissions will be for the majority of customers to reject these new GenAI services until they are absolutely critical.”
Unsustainable design
Russell Macdonald, chief technologist at HPE, said – as well as needing to balance customers’ AI demands with their environmental commitments – there is another fundamental issue with how the hyperscale cloud giants operate that makes this balancing act even harder. “Hyperscale public cloud was never designed or architected with sustainability in mind,” he told Computer Weekly.
“At the size of hyperscale, cloud providers cannot offer individual customers granular data on the carbon intensity of their use of cloud services in a cost-effective way. The truth is, they don’t know or care, and even if they did know, it would reveal details about how they manage their datacentres and infrastructure that they don’t want customers to know, such as how low their utilisation levels are.”
Hyperscale cloud platforms are built with redundancy in mind, he continued, and are deliberately over-provisioned to provide users with access to scalable and elastic cloud resources, based on what their compute requirements are.
“The bare metal infrastructure that underpins the cloud services we use is therefore poorly utilised – often less than 30% – despite the high levels of automation in cloud datacentres,” he said.
Macdonald also pointed to the fact that both Microsoft and Google emphasised in their respective reports how AI is helping to make other parts of their operations more environmentally friendly. On this point, there is a reference made to how AI is helping to optimise route planning in Google Maps so drivers can reduce their fuel consumption.
“While there is no doubt that AI can play a positive role in combatting climate change, it is a very small proportion of these companies’ total AI investments,” he added.
Emissions goals
What is becoming clear, he continued, is that the “climate moonshot” is even further away for both Microsoft and Google than when their respective 2030 emissions goals were set.
“I’m not singling Microsoft and Google out here specifically as being outliers, it’s purely down to the timing of their respective annual sustainability reports,” said Macdonald. “They are the leading edge of an emerging trend in which demand for AI services – arguably before anyone really knows what to use it for, but [they] don’t want to miss out – completely tears up the rulebook on datacentre and infrastructure requirements.”
Mark Butcher, founder and director of IT sustainability consultancy Posetiv Cloud, told Computer Weekly it’s time the hyperscalers started to accept that the ambitious net-zero targets they have set themselves are “ridiculous”.
“Microsoft are still saying that they will be carbon negative by 2030 when their emissions have increased 30% – it’s just a practical impossibility, and particularly if you consider that they are just at the start of their AI journey and continuing to invest vast sums in datacentres to support AI,” he said. “The problem is only going to get worse. And that’s before you consider the embodied emissions of the physical infrastructure they will be buying, and the water usage to cool the datacentres, and so on.”
With Amazon Web Services (AWS) expected to drop its environmental report later in July 2024, Butcher said it’s feasible that it will report a similar upward trend in its GHG emissions, too.
“The general theme is that there is no way the hyperscalers are going to meet their commitments, and the only way they are going to achieve it is to create more abstraction between the numbers and what is actually happening – effectively getting more and more creative with the numbers,” he said.
“This is hugely damaging to customer environmental programmes because customers are taking the supplier carbon emissions at face value and assuming they are correct, whereas the real numbers may be several orders of magnitude larger – often tens and sometimes hundreds of times bigger than what’s reported in vendor dashboards.”