Digital Newsletter: Cloudy with a Chance of Hidden Emissions

Background

Accounting for between 3 and 4% of global greenhouse gas (GHG) emissions in 2020 [1], the digital sector shows no signs of slowing down in its consumption of energy and resources. Alphabet (Google) and Microsoft, two of the five largest companies in the sector, recently made headlines by announcing significant increases in their carbon footprints for fiscal year 2023 (+13% and +20%, respectively [2][3]) despite its stated ambitions to be “net zero” or “carbon negative” by 2030.

Every component of the global digital system follows the same trend: the number of terminals The number of new units manufactured increases every year [4]. The networks are consuming more and more electricity (despite efficiency gains) to support the strong growth in data traffic [5]. The data centers They now account for about 2% of global electricity consumption—roughly the same as countries such as France or Germany—with consumption having doubled between 2015 and 2022 and expected to double again by 2026 [6] with the continued growth of energy-intensive technologies such as artificial intelligence (AI) and blockchain.

Cloud service providers, such as Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP), have continued to grow and gain ground over “traditional” data centers since our latest article about them. Given their size and influence, they should be key players in the decarbonization of the digital sector—but is that actually the case? How has their climate strategy evolved over the past two years?

Furthermore, with the proliferation of climate reporting regulations around the world, calculating the GHG emissions of cloud services is becoming a key consideration for businesses. The hyperscalers Do they provide enough information to help their customers assess their emissions related to cloud usage? This article aims to help companies see through the claims made by hyperscalers and take a critical look at their cloud-based carbon footprint calculators.

Introduction

Companies' IT needs—both in terms of computing and storage—are increasingly reliant on services cloud. In 2023, approximately 75% of the workloads computers were running in the cloud [7], and that figure is expected to reach 90% over the next three years. In terms of storage, the solutions cloud are storing more and more data over time. A recent study conducted by Boavizta [8] confirms these trends: the vast majority of companies surveyed, regardless of size, expect an exponential increase in their use of (workloads and storage) in the public cloud by more than 10% per year.

The cloud market is still dominated by large-scale service providers (commonly referred to as “hyperscalers”), such as the “Big Three”—AWS (Amazon), Azure (Microsoft), and GCP (Google).

In 2022, Carbone 4 analyzed the climate strategies and communications of these three suppliers. The article concluded that the companies’ climate claims lacked maturity and highlighted the shortcomings of the carbon footprint calculation tools made available to customers.

These tools were not transparent enough to enable those clients to comply with current regulations, let alone allow them to anticipate the CSRD (Corporate Sustainability Reporting Standard).

As their influence in the digital sector continues to grow, have hyperscalers raised their standards for climate action, or are the outlooks still stormy?

"Net-Zero" or "Carbon-Neutral" Companies, Again and Again—Despite Insufficient Progress in Reducing Emissions

The carbon neutrality goals of the leading cloud service providers have changed little: Amazon still aims to achieve “net-zero across all its operations by 2040,” and Microsoft remains committed to being “carbon neutral by 2030”; despite the lack of a clear scientific definition of what it means for a company to be “carbon neutral” [9] and the misleading nature of this terminology. Google has abandoned its corporate claim of being “carbon neutral since 2007” in favor of a stated goal of “net-zero by 2030.” Other major providers, such as Oracle and Alibaba Cloud, are aiming for “carbon neutrality” by 2025 and 2030, respectively.

These claims of carbon neutrality are being viewed less and less favorably by customers: the aforementioned Boavizta study shows that 88% of AWS, Azure, and GCP users surveyed believe that the goal of a “carbon-neutral cloud” is illusory or lacks sufficient transparency.

The Net Zero Initiative defines the three ways in which a company can contribute to global carbon neutrality : reduce one's own emissions, help others reduce their emissions, and increase carbon sinks. Since the potential for developing carbon sinks is limited [10], and the decarbonizing effect of digital companies on the rest of the economy remains to be proven [11], The top priority for hyperscalers should be to effectively reduce their own emissions.

However, None of the three largest cloud service providers has yet seen significant reductions in its emissions (except in 2020, when the declines can be considered largely due to the Covid-19 pandemic). In 2022, Amazon and Microsoft reported reductions of 0.7% and 0.5%, respectively—about one-tenth of the annual reductions of 5% to 7% required to meet the Paris Agreement targets—while Google’s emissions increased slightly. In 2023, while Amazon recorded a more pronounced decline (3%), Google and Microsoft reported sharp increases (13% and 20%, respectively), due to investments in equipment and infrastructure related to the rise of generative AI.

To make matters worse, if we examine the carbon footprints of the “Big Three” using the method location-based for energy consumption, All of them rose in 2023, in some cases dramatically: +2% for Amazon, +12% for Google, and +22% for Microsoft [12].

The cuts called for in recent years are almost exclusively attributable to reductions in Scope 2 emissions, unsurprisingly. Indeed, as noted earlier [13], the hyperscalers’ main strategy for reducing emissions in recent years has been to purchase electricity labeled as “renewable” through contracts such as power purchase agreements (PPAs) Power Purchase Agreement) or non-aggregated certificates of origin (GO).

It should be noted that hyperscalers also lack transparency regarding the nature of these “green electricity” contracts, as they do not disclose the proportion of electricity coming from each type of contract (GO, PPA, etc.), even though this information is required by the CSRD.

Reporting Scope 2 emissions using the method market-based enables hyperscalers to combine zero Scope 2 emissions for each kilowatt-hour (kWh) of electricity purchased through PPAs or GO certificates.

This raises several issues:

• PPAs and GO certificates do not guarantee spatial or temporal consistency between the “green” electricity purchased and the electricity consumed by the customer [14]. These systems can promote the development of new renewable energy generation facilities under the right conditions [15], but the electricity physically consumed by hyperscalers’ data centers still relies on all the other power generation facilities in the local grid. In a country like Ireland, where the average carbon intensity of electricity is nearly 400 gCO2e/kWh [16], reporting GHG emissions from electricity consumption as zero is misleading.
  • The Net Zero Initiative recently published a guide on accounting for energy purchases [17], which explains how to rigorously account for the benefits of renewable energy purchases based on contract types.
• The increase in electricity consumption and relying solely on the decarbonization of electricity can lead to significant competition for use : There is no guarantee that there will be enough low-carbon electricity for all sectors, as the IPCC states that global climate goals cannot be achieved without efforts to reduce energy consumption.
• Some countries, such as France, have long required emissions to be accounted for using the location-based. The Corporate Sustainability Reporting Directive (CSRD)[18] adds a requirement for all companies reporting their emissions in the'European Union' to use the method location-based. More generally, compliance with the GHG Protocol also requires the reporting of emissions location-based even a value market-based is given for Scope 2 emissions.

Even if we disregard the questionable nature of these Scope 2 emissions reductions claimed by major cloud service providers using the method market-based, another problem arises: this reduction lever has reached its limit. In fact, Scope 2 emissions now account for only a few percent of the total carbon footprint reported by hyperscalers.

With nearly 100% of electricity consumption covered by purchases of “green” electricity, Scope 2 emissions are no longer reducible (since they are close to zero) and can no longer shoulder the entire burden of the hyperscalers' decarbonization strategy, as shown in the figure below.

Among the “Big Three,” only Google still has room to reduce its Scope 2 emissions, having rightly opted for a more accurate accounting method: emissions related to electricity consumption are counted as zero only when the purchased green electricity meets spatial consistency and temporal consistency in terms of time[19].

To achieve their decarbonization goals, Hyperscalers need to focus on strategies other than purchasing green electricity, since their physical impact on reducing emissions is debatable and their limitations are already becoming apparent. Efforts to reduce consumption (for example, questioning whether it is necessary for the industry to invest so heavily in AI), neglected for a long time, should be incorporated into hyperscalers' strategies if they truly wish to make progress toward their emissions reduction goals.

Beyond their own decarbonization efforts, cloud providers have another responsibility : to provide their clients with precise tools so they can report the emissions associated with their cloud service purchases. Current regulatory developments are moving in this direction. In Europe, for example, the CSRD requires reporting companies to specifically disclose the carbon footprint of their purchases of cloud services as a subset of their “purchases of goods and services.”

But how detailed are the hyperscalers’ “carbon calculators” at present? Are they accurate enough to enable companies to complete their CSRD reports?

The accuracy of carbon calculators has improved slightly

Like any purchased service, expenses related to cloud services are based on physical flows that emit greenhouse gases and must be included in the carbon footprint calculation of any company that reports its emissions.

The figure below shows the main sources of emissions from cloud services and how to account for them, highlighting the relevant GHG Protocol category based on the reporting company’s perspective:

The largest cloud providers all offer “carbon calculators” to help their customers estimate their emissions from using cloud services. In 2022, Carbone 4 highlighted the shortcomings of these tools: a lack of transparency regarding the methodology used to calculate emissions and metrics, and an insufficient scope of coverage. Two years later, the overall quality of these calculators has improved slightly:

Nevertheless, most of these tools remain largely inadequate overall, often providing only broadcasts market-based for electricity consumption and excluding major sources of emissions.

In particular, omission by Azure and AWS upstream energy (emissions related to the construction and maintenance of power generation facilities, electricity transmission, losses, etc.), combined with accounting for market-based, virtually eliminates all emissions related to electricity consumption.

By also excluding the manufacturing of IT equipment (and all Scope 3 emissions, for that matter), the carbon calculators from AWS and Oracle report near-zero emissions, thereby creating the illusion of a service with no environmental impact. This makes it very difficult for customers to comply with the CSRD reporting requirements and to gain an understanding of the carbon footprint associated with their cloud usage.

Finally, AWS and Azure provide metrics on emissions avoided based on these near-zero emissions and on the assumption that customers would have hosted their operations on on-premises architectures (on-premises) less efficient and using "regular" electricity if they hadn't used their services. These misleading indicators disregard the best practices of the WBSCD and the NZI regarding the transparency and credibility of avoided emissions claims. They also disregard the best practices described in The Net Zero Initiative's sector-specific guide on avoided emissions for digital solutions.

For example, while cloud services can, in theory, offer efficiency gains compared to an equivalent solution for a constant level of IT activity, on-premises, they may also be responsible for significant rebound effects, which are not taken into account in these indicators.

On the other hand, GCP provides (as it did two years ago) emission factors location-based in real time for electricity consumption and now includes important factors such as server manufacturing.

The OVHcloud calculator covers a broad scope and is very transparent about the methodology used to calculate emissions, attributing nearly every component of OVHcloud’s carbon footprint to its customers and enabling them to understand their carbon dependencies.

At the same time, nonprofit organizations are launching initiatives to estimate the energy consumption required for certain cloud services when providers do not clearly disclose this information, such as the tool Boavizta Cloud Scanner for AWS.

Would you like to stay informed about our work on this topic?

Click here

The advice and best practices provided for reducing emissions remain half-hearted when it comes to energy conservation

AWS, Azure, and GCP offer best-practice frameworks for a more sustainable use of their services. The underlying guidelines primarily address effectiveness - the amount of energy required for a given service: use the appropriate services to meet a given need, define data retention policies, and optimize software and architecture to run the workloads, etc.

With regard to the carbon intensity of the electricity consumed, GCP emphasizes the importance of choosing regions with low-carbon electricity in location-based to host activities, and highlights mechanisms such as planning for workloads depending on the wind and sunlight.

However, one lever for reduction is missing: the moderation. While there are many guidelines on the a way to meet a specific need using the most efficient architecture, none of them mention the the need to first consider whether it is necessary. The conclusion remains the same: hyperscalers are not on track to achieve the 5 to 7 percent annual reduction in emissions required by the Paris Agreement, and it is unlikely they will succeed without rethinking their business models to include incentives for energy efficiency.

Conclusion

Without controlling the rebound effect, cloud data centers lose their relevance in terms of reducing emissions associated with organizations’ IT operations, despite the shared use of servers and building optimization.

From a customer’s perspective, the key factors to keep in mind during a cloud migration aimed at reducing emissions are highlighted in The Net Zero Initiative Guide for the Digital Sector :

• Choose a region with a reliable access to low-carbon electricity – in other words, with a emission factor location-based low electricity costs – to host its activities,
• Reevaluate your needs and limit the growth of your organization’s IT operations (volumes of stored data, desired latency, required computing power, etc.) – set absolute emissions reduction targets rather than intensity-based targets might help.

On the supplier side, Amazon and Microsoft must drastically improve the quality of their carbon calculators to enable their customers to comply with the CSRD and better manage their emissions, and all hyperscalers must consider implementing business models that are more sustainable and less prone to rebound effects.

Cloud service customers can also play a role, by contacting their suppliers and requesting greater transparency and more details about the emissions associated with their purchases. In the meantime, when the scope covered by carbon calculators is insufficient, emissions must be re-estimated using assumptions that are more representative of the actual carbon dependencies at play.