ESG Focus: Do Data Centres Threaten ESG Targets?

ESG Focus | Jun 24 2024

This story features GOODMAN GROUP, and other companies. For more info SHARE ANALYSIS: GMG

FNArena’s dedicated ESG Focus news section zooms in on matters Environmental, Social & Governance (ESG) that are increasingly guiding investors preferences and decisions globally. For more news updates, past and future:

Do Data Centres Threaten ESG Targets?

Will Investors, data centre developers and operators be forced to tackle the challenges and opportunities arising from increased cooling demand and energy use in our generative Ai world?

-Data centre growth collides with energy and water use
-The state of play for Australian companies
-Data centre cooling technological trends
-Nvidia efficiency versus ESG challenges

By Danielle Ecuyer

A top down look at the data centre industry

The expansion in data centre developments to accommodate processing and data storage for generative Ai is a rubber hits the road moment in terms of energy and water usage.

Just when the world is grappling to reduce both emissions from energy production and tackle increasing water scarcity, demand for both resources is growing rapidly due to a surge in demand for large (hyperscalers) and smaller data centres (co-location providers) developments.

ESG reporting requirements and actual reports on water/energy use and efficiency remain scant but are likely to increase around these two topics, with some players currently providing more information than others.

In the US, Google was sued to provide water usage information at its The Dalles, Oregon based data centres. The lawsuit ultimately revealed water consumption had tripled over the previous five years.

Google’s data centres were using more than 25% of all the city’s water usage.

Increasingly, data centre developers and operators will need a social license to run the centres, Jarden concluded in a recent analysis of water usage efficiency in this sector.

Depending on the company and the sovereign state in which data centres operate, disclosure varies greatly.

In Australia, Jarden highlights a 2023 UTS study disclosed some 5% of sustainability experts considered Australian data centre operators had provided insufficient information.

Although, it was highlighted Australian data centre developers kept water and energy usage high on the list of priorities and domestically closed-loop water cooling systems tend to be used.

In contrast, US hyperscalers normally employ open loop systems, which means used water is not recycled. 

Microsoft reported Ai consumes 1.8l to 12l of water per KWh of energy use for its data centres. Microsoft has increased water consumption by 34% between 2021 and 2022.

Jarden observes a single text Ai request uses ten times the energy in comparison with a typical Google search.

Is the use of water and energy also inextricably linked?

In the US hyperscalers have preferred to develop data centres in states with strong renewable energy production and large open spaces such as Arizona. Think solar energy.

With that comes the problems of increased water challenges in drier states and higher temperatures which exacerbates the cooling requirements.

One solution from Google is to adopt air cooling (air conditioning) powered by 80% renewable energy instead of water. This option is only possible up to a certain temperature.

Types of cooling available

Wilsons’ recent deep dive into domestic data centres included a description of the varying types of server cooling systems.

Air conditioning or air cooling is the traditional method and most frequently used to cool data centres. This is unlikely to change in the short-to-medium term, according to Wilsons. Notably, air cooling is the least efficient and most expensive to operate, just walk past your office server to gauge how intensive it is.

Direct-to-chip is considered a transitory cooling method, which uses liquid within the server to cool it, however, the costs remain elevated, and it has been typically used at smaller scales.

Liquid immersion cooling is the submersion of computer components in a “thermally conductive, but not electrically conductive liquid”. Due to the efficiency benefits of this method, it is up to -50% less costly to run versus air cooling, and is considered the best potential option for high-density computing infrastructure or Gen Ai.

The industry is expected to transition to liquid immersion from direct-to-chip.

Is water usage a financial risk for data centres?

The UTS study confirmed there is a lot more work for data centre developers and operators to do in terms of disclosure, as well as increased awareness from sustainability experts to understand the implications of increased data centre developments.

Jarden discussed the varying disclosures for Australian listed companies.

Goodman Group ((GMG)) currently doesn’t disclose water usage efficiency (WUE) but it does reveal it seeks to make its data centres as efficient as possible and employs closed-loop water cooling systems, which obviate the need for constant water inputs. The company also notes it uses sub-metering to track water usage.

NextDC  ((NXT)) provides WUE statistics and reports reuse of water for its cooling systems. The company has achieved a reduction in WUE to 1.73l per kwh of energy from 2.07l per kwh. Most interestingly, Next DC has stated it views water use as a potential “financial risk” due to the increasing costs of a warming world and the rising costs of water rights.

Macquarie Technology ((MAQ)) does not report water use; however, it points to the use of the best cooling technologies available where possible.

Telstra ((TLS)), TPG Telecom ((TPG)) and Seek ((SEK)) are also exposed to the issue.

Telstra is both a data centre owner and a user with TPG and Seek both users.

Telstra highlights data centres are one of the largest energy consumption sources for the telco and although total water consumption is provided, the data centre usage is not isolated. Nor does the company highlight the type of cooling systems used. Telstra did note a new trial of a water cooling system to reduce energy consumption.

TPG is reputably expanding data centre use, but makes no mention of energy or water use metrics.

Seek is also growing data centre usage and only reports the related energy consumption.

Ultimately, Jarden concludes water usage is an “under reported risk” and as demand for centres grows, so too will the issue of water intensity at data centres, as well as reporting requirements, as investors seek out a better ESG-adjusted outlook.

Liquid cooling technologies a beacon of hope for data centres

According to Dell’Oro Groupliquid cooling market revenue is expected to reach US$2bn by 2027, on a 60% compound average growth rate between 2020 and 2027.

Liquid cooling includes submersion technologies, as outlined, and ASX-listed DUG Technology ((DUG)) has approved and pending patents in liquid immersion cooling technology.

The company has been employing submersion liquid cooling for over ten years in its High-Performance Computing operations.

Analysts at Wilsons point out DUG is essentially a global seismic data analytics company for the oil and gas industry, but it is exploring options and growth potential for immersion cooling technology into data centres.

Chip suppliers supporting the power density (usage) problem

There are two sides to the water and energy ESG equation, and one that is often overlooked is efficiency.

Wilsons pointed to examples of efficiency gains as outlined by Jensen Huang at Nvidia’s March GTC conference:

“If you were to train a GPT model, a 1.8 trillion parameter model, it took about 3-5 months with 25,000 Amperes.

With Nvidia’s Hopper chips it would take ~8,000 GPUs, would consume 15MWs and would take about 90days.

This would allow you to train a ground-breaking AI model.

If you used the Blackwell chips, it would only take ~2000 GPUs (instead of 8,000) also over 90-days but only consume ~4MWs (instead of 15MWs)”

Nvidia’s technological development has allowed for around a -75% power reduction in energy use for processing with the new Blackwell GPUs.

A fun fact, Wilsons highlights is, Moore’s Law states in eight years there can be a 40x to 60x improvement and yet this Law is below what Blackwell will achieve.

Tremendous food for thought on the pace of technological innovation in the GenAi picks and shovels space, if nothing else!

Ultimately, the Nvidia example suggests part of the ESG solution for data centres will come via improved technologies that lower both energy, cooling and water usage.

Hyperscalers face real ESG challenges, is this a red flag for investors?

Microsoft’s recent 2024 Environmental Sustainability Report pointed to a -6% fall in Scope 1 and Scope 2 emissions since 2020, whereas Scope 3 emissions rose 30% which directly relates to data centre developments.

The company’s water consumption also increased by 87% over the period.

Microsoft’s target is to be carbon negative, water positive with zero waste, as well as protecting more land than it uses by 2030.

Amazon’s goal is to power all its energy use with renewables by 2030. Alphabet is aiming for net zero-emissions across all its businesses and value chain by 2030.

Morgan Stanley questions whether these tech behemoths can reach their 2030 targets, which poses a risk to ESG and Sustainability funds that own them.

For more reading on Generative Ai:

FNArena’s dedicated ESG Focus news section zooms in on matters Environmental, Social & Governance (ESG) that are increasingly guiding investors preferences and decisions globally. For more news updates, past and future:

Find out why FNArena subscribers like the service so much: “Your Feedback (Thank You)” – Warning this story contains unashamedly positive feedback on the service provided.

FNArena is proud about its track record and past achievements: Ten Years On

Share on FacebookTweet about this on TwitterShare on LinkedIn

Click to view our Glossary of Financial Terms