Discussions around the sustainability of public cloud platforms have become increasingly prominent in recent years. 

Hyperscale providers now publish a growing volume of data, dashboards and metrics designed to demonstrate the environmental efficiency of their infrastructure. 

But, for enterprise IT leaders tasked with making informed decisions, distinguishing between meaningful insight and marketing narrative remains a persistent challenge.

At a surface level, cloud platforms present a compelling sustainability proposition. The ability to consolidate workloads into highly optimised, large-scale data centres offers clear advantages in energy efficiency compared to fragmented on-premise environments. Hyperscalers also benefit from access to renewable energy at scale, alongside the engineering expertise required to continually improve power usage effectiveness (PUE) across their estates.

The difficulty arises when organisations attempt to translate these high-level efficiencies into a clear understanding of their own environmental impact.

In many cases, the data made available by cloud providers is not directly comparable. Methodologies differ, reporting boundaries vary and the level of granularity provided is often insufficient for organisations that want to align cloud usage with their own carbon reporting frameworks. As a result, enterprise IT leaders are frequently left relying on indicative estimates rather than verifiable, auditable data.

This lack of transparency creates a disconnect between perceived and actual sustainability outcomes.

One of the most overlooked aspects of this conversation is that cloud adoption does not eliminate the physical lifecycle of technology – it redistributes it. Servers, storage platforms and network equipment still have to be manufactured, deployed, maintained, refreshed and retired. The environmental impact associated with these stages does not disappear simply because infrastructure is consumed “as-a-service”.

For many organisations, the sustainability conversation becomes disproportionately focused on operational efficiency within the data centre while the upstream and downstream impacts of technology are given far less attention.

In reality, the full lifecycle of digital infrastructure must be considered.

This includes not only how efficiently systems operate in production, but how hardware is sourced, how frequently it is refreshed, how securely it is decommissioned and whether it is reused, redeployed or prematurely discarded. These factors can have a significant bearing on an organisation’s overall environmental footprint, yet they are rarely visible within standard cloud sustainability reporting.

Enterprise strategies therefore need to evolve beyond accepting provider-level claims at face value.

Organisations should look to adopt a more holistic approach, combining the data made available by cloud providers with their own internal governance, asset tracking and lifecycle management processes. This may include:

• Establishing clearer visibility over hardware refresh cycles and associated emissions 

• Integrating lifecycle considerations into cloud migration planning 

• Working with partners who can provide auditable reporting across decommissioning and reuse activities 

• Ensuring sustainability metrics are aligned to recognised standards rather than provider-specific methodologies. 

By doing so, enterprises can move from a position of passive consumption to active accountability.

At the same time, there is a clear role for industry standards in improving consistency and comparability. Without a more unified approach to carbon reporting across cloud platforms, organisations will continue to face challenges in benchmarking providers and making informed decisions. Greater alignment around reporting frameworks, boundaries and measurement methodologies would provide much-needed clarity for enterprise users.

However, it is unlikely this will be resolved solely through standardisation.

Enterprise IT leaders also have a role to play in demanding greater transparency. This means asking more detailed questions of providers, challenging assumptions and ensuring sustainability claims can be substantiated with meaningful data. As sustainability becomes increasingly embedded within procurement and governance processes, the ability to evidence environmental impact will carry greater weight in supplier selection.

There is also a growing skills dimension to consider.

Managing sustainability within modern IT environments is no longer limited to facilities or energy management teams. It increasingly requires a blend of expertise across infrastructure engineering, data security, asset lifecycle management and environmental reporting. Organisations need individuals who understand not only how systems operate, but how decisions made at each stage of the lifecycle influence risk and environmental impact.

This shift is creating new opportunities, but also new responsibilities.

As digital infrastructure continues to scale, the industry must recognise that sustainability is not confined to where workloads run, but how technology is managed from deployment through to end-of-life. The ability to engineer these transition points securely, responsibly and with full accountability will become an increasingly important differentiator.

Ultimately, the challenge for enterprise IT leaders is not simply to choose the “greenest” cloud provider, but to ensure that sustainability is embedded across the entire lifecycle of their technology estate.

Without that broader perspective, there is a risk that organisations optimise for what is visible, while overlooking the impacts that sit just outside the scope of the cloud narrative.

IT infrastructure-related carbon emissions reporting is slowly becoming a global regulatory requirement. 

The International Financial Reporting Standards (IFRS) have incorporated climate-related financial disclosure requirements, including obligations to inventory and report Scope 1, 2, and 3 emissions. 

Individual countries – including Australia, Brazil, China, Hong Kong, Japan, New Zealand, Singapore, and the United Kingdom, as well as the European Union and the State of California (United States) – have enacted these standards into law, although some countries have delayed or minimised Scope 3 reporting. 

Emissions reporting is subject to strict accounting and audit requirements, with fines and reputational risk for those firms whose reporting does not comply with the applicable standards.  

For many companies, emissions associated with IT infrastructure operations account for a significant part or even most of their total emissions inventory. With operations typically spread across owned, colo, and cloud data centres, capturing a complete energy use and emissions inventory depends on the data available from cloud and colocation service providers.

Cloud emissions data 

Cloud service providers have significantly improved their emissions dashboards over the past several years. They allocate Scope 1, location- and market-based Scope 2, and limited Scope 3 emissions related to their public cloud operations (see Table 1). Emissions estimates, with some exceptions, are based on metered resource or energy use and location-specific or regional emissions standards. The data provided should be sufficient to meet customer and regulatory emissions reporting obligations.

Scope 1 emissions

Oracle is the only cloud provider that does not report Scope 1 emissions. Most Scope 1 emissions result from the operation of standby generators. They are typically less than 1% of a data centre’s Scope 1 and Scope 2 operating emissions.  

Where a data centre is supplied by behind-the-meter generation or is required to provide grid support for several hundred hours or more each year (for example, as mandated in Ireland and Texas), Scope 1 direct emissions will become a larger percentage of the reported emissions.  

Scope 1 emissions are more difficult to offset. Offsets derived from direct CO² capture through biological, chemical, and/or mechanical means are required. Environmental attribute credits (EACs) cannot be used. Where a facility depends on a captive fossil-fuel generation plant for power, achieving net-zero emissions will become more difficult and expensive. 

Scope 2 emissions

Operators provide location- and market-based emissions for their public cloud offerings. Each provider uses a slightly different calculation method for energy and resource use and associated emissions factors. Still, each reporting system provides a reasonable approximation of emissions associated with a customer’s operations.  

Location-based emissions

AWS, Google, and Microsoft estimate emissions based on energy use for computing resources and regional emissions factors aggregated from measured or grid-level location-specific emissions. AWS data is limited to specific public cloud offerings.  

Oracle and IBM base emissions estimates on measured customer energy use and regional emissions factors, with Oracle applying a spend-based allocation method for services without energy measurements. IBM Cloud only provides location-based emissions data.

Emissions estimates for cloud services such as Office 365 and Oracle software-as-a-service offerings are typically based on service usage levels combined with regional emissions factors. These estimates have greater uncertainty than those for cloud infrastructure usage. 

Market-based emissions

Market-based emissions reporting is a black box, often reporting an emissions value of zero because many hyperscalers claim they match all their electricity consumption with carbon-free generation. 

However, they do not provide details on the number of EACs applied and associated emissions avoided, which makes it impossible to validate calculations. In many cases, cloud service customers cannot get a reasonable assurance audit of their market-based cloud emissions because the auditor cannot review the EAC details associated with the cloud providers emission data.  

As a result, some operators apply EACs to their cloud service emissions to achieve their net-zero goals. The double application wastes EACs and financial resources because emissions are offset twice. 

Cloud operations in a colocation data centre

All five cloud operators use a mix of owned and wholesale and retail colocation facilities to deliver public cloud services. Emissions calculations for the many colocation facilities are less transparent and more uncertain than for owned facilities. Because cloud providers often do not directly control these data centre operations, nor energy procurement, emissions calculations may be based on regional emission factors and energy-use models. 

Scope 3 emissions

Cloud providers allocate a limited number of Scope 3 categories to their customers. These emissions have little inherent value, as they are typically small and not material to the inventory. None of the providers allocate their full Scope 3 inventory, because public cloud operations are just one of many business line services within supply chains. Much overall Scope 3 inventory is irrelevant to public cloud services.

AWS, Google, and Microsoft report ‘Other Energy-Related Emissions’ associated with energy losses in transmission and distribution systems. These emissions are typically 1% to 8% of a given region’s emissions.

AWS allocates the embedded carbon in equipment and building systems. These emissions are one-time, highly uncertain emissions over which customers have little or no control. It is meaningless for customers to account for these emissions. 

IT Operations in colocation facilities

IT operators face similar emissions accounting challenges in colocation data centres. Most colocation operators are well behind cloud service providers, with few offering an online calculator or data portal to estimate energy use and associated GHG (greenhouse gas) emissions. 

While reliable information on contract terms and conditions is limited, conversations with colocation operators indicate few currently have contract terms for a regulatory-mandated data exchange with tenants covering GHG emissions and other regulatory mandated data. 

Location-based emissions

Most operators have access to data on their IT electricity use and the facility PUE under the terms of the contract. They also know the location of the colocation data centre. This knowledge enables them to apply utility, energy supplier, or regional emission factors to their energy use to calculate location-based emissions. 

Market-based emissions

Getting a usable market-based emissions estimate is more difficult and mirrors the problems with information supplied by the cloud providers. While many colocation providers will attest they match carbon-free EACs to all their electricity consumption, they do not provide the number of EACs, their avoided emissions value, and a reasonable assurance certification of the market-based claim that their tenants can use to audit and certify their GHG emissions inventory. 

The role of IT efficiency in emissions reporting

While accurate emissions accounting is important, the primary purpose of a GHG management programme should be to reduce energy use and associated emissions from IT operations. Uptime Institute survey data reveals that more than 50% of IT operators do not prioritise IT efficiency in their cloud and colocation-based IT operations. This suggests more can be done. 

Discussions with IT operators and consultants who offer cloud and IT optimisation services indicate that many operators can reduce their energy use and associated GHG emissions by up to 40%. These reductions can be achieved through steps such as right-sizing IT assets, including proper balancing of CPU and memory capacity, and using virtualisation and workload management and placement software.

Operators should invest in improving energy efficiency in their IT infrastructure before investing in EACs and offsets to reduce their GHG emissions inventory. 

Next steps 

Emissions reporting by cloud service and colocation providers has improved markedly over the past five years. Several steps remain to improve these processes further. 

The industry needs to develop a standard method for allocating Scope 2 and Scope 3 operational emissions to operators and tenants in colocation facilities. Currently, 3 or 4 allocation methods are in use, reducing the usability and accuracy of data across IT and colocation operators. 

Standard contract language should be developed to manage the exchange of data between public cloud and colocation service providers and their customers. This language should include standard methods to calculate and report energy use and emissions factors. 

Cloud and colocation service providers should publish the emissions factors they use by facility and region, the quantity of carbon-free energy consumed by a given facility or within a region, and the number of EACS and avoided emissions used in market-based calculations. 

IT operators should focus on minimising IT energy use to reduce emissions before embarking on a detailed accounting exercise. They should prioritise investments in real emission reductions.   

Abu Dhabi has taken a step towards global artificial intelligence (AI) with Falcon Perception, a multimodal model that enables machines to efficiently see, read and interpret the physical world.

Developed by the Technology Innovation Institute (TII), the applied research arm of the emirate’s Advanced Technology Research Council, Falcon Perception expands the UAE’s AI ecosystem by combining vision and language capabilities. 

​As global AI competition intensifies, the UAE hopes to position itself among nations capable of advanced multimodal systems at scale, making Falcon Perception central to this ambition. With approximately 600 million parameters, Falcon Perception is notably more compact than many prominent multimodal models, which often use several billion parameters.

“Our goal with Falcon Perception was to challenge the prevailing assumption that vision systems must rely on complex multi-stage architectures. By demonstrating that a single dense transformer can handle perception tasks efficiently, we are opening the door to a new generation of scalable multimodal systems,” said Hakim Hacid, chief researcher at TII’s Artificial Intelligence and Digital Research Centre.

This efficiency-performance balance demonstrates a broader AI trend: rather than increasing parameter counts or requiring extensive compute resources, researchers emphasise model design optimisation, such as efficient transformer variants, to achieve strong results even on resource-constrained hardware.

Multimodal AI is widely seen as the next frontier of artificial intelligence. While large language models (LLMs) have dominated recent advances, the ability for machines to interpret and interact with the physical world is becoming critical as AI expands into robotics, manufacturing and intelligent infrastructure.

Falcon Perception employs a unified transformer-based architecture, enabling end-to-end integration of visual and linguistic features at the model input level. Unlike traditional pipelines that join separately trained computer vision and NLP modules, Falcon Perception processes and reasons across modalities directly in its shared network, reducing inference latency and deployment complexity.

Consequently, the system interprets complex, multi-object visual scenes using natural language prompts. Users can instruct the model to identify, count or segment specific objects in an image, and Falcon Perception returns bounding boxes, segmentation masks or text outputs, even in crowded, intricate environments.

Such capabilities have clear implications for industry. In manufacturing, the model could enable automated inspection and defect detection. In robotics, it enables machines to follow natural-language instructions in dynamic environments. In enterprise settings, it can streamline large-scale document processing and visual data labelling.

For TII, the launch represents not only a technical milestone but also a step in a broader national strategy. Since beginning its AI agenda, the UAE has prioritised building sovereign capabilities, ensuring domestic development, responsible governance and alignment with long-term economic goals for critical technologies.

“Falcon Perception reflects TII’s commitment to advancing AI capabilities that are both cutting-edge and practical. By rethinking how vision and language models are built, we are enabling more efficient multimodal systems that can be deployed across real-world industries while strengthening sovereign AI capabilities,” said Najwa Aaraj, CEO of TII.

TII’s work spans AI safety, evaluation and deployment frameworks, and large-scale research programmes. A flagship outcome of this effort is Falcon, the UAE’s homegrown LLM, first launched by TII in 2023. Falcon quickly gained international attention for its performance and for being released as an open source model, reflecting Abu Dhabi’s belief that openness and governance can coexist.

Falcon is not positioned merely as a technical achievement, but as part of a broader national AI development system. By combining scientific research with agile decision-making at a government level, Abu Dhabi aims to accelerate adoption while maintaining oversight and trust.