Will AI ever be 1.5C compatible?
As the Artificial Intelligence (AI) industry rapidly expands, with billions in government and private investments, concerns over its environmental impact are intensifying.
Two months into 2025, US President Donald Trump launched the ‘Stargate Project,’ a $500bn initiative to develop AI infrastructure in the US.
Meanwhile, European Commission President Ursula von der Leyen unveiled ‘InvestAI,’ aiming to mobilise €200bn for AI investment, including a €20bn European fund for AI gigafactories. The UK has also launched its own AI plan.
McKinsey estimates the long-term AI opportunity at $4.4trn in added productivity growth from corporate applications, with 92% of companies planning to increase AI investments over the next three years.
“Increased investment in AI has further accelerated the growth in demand for data processing capabilities and capacity, leading to the further expansion of data centres globally,” explains Carolina Pinto, strategic intelligence analyst at GlobalData, a UK-based insights platform.
Pinto highlights that these data centres consume significant amounts of energy and water, making them a fast-growing source greenhouse gas (GHG) emissions.
Standard workloads on conventional servers use around 10 to 18 kilowatts (kW) of electricity, whereas AI-enabled racks consume up to 60 kW. This is equivalent to the energy needed to power about 600 standard LED light bulbs. AI developers even suggest that 120 kW racks will be needed to develop the most advanced AI models.
The AI boom is already hindering the progress made by major tech companies on their climate commitments. Generative AI is a particular sticking point, requiring some 33 times as much energy as task-specific software.
"Increased investment in AI has further accelerated the growth in demand for data processing capabilities"
Businesses that seek to adopt AI must ensure that their data centre and AI services suppliers are not only transitioning to renewable energy use, but are also investing in energy efficiency technologies and more sustainable cooling alternatives.
Carolina Pinto, strategic intelligence analyst at GlobalData, a UK-based insights platform
Amount Microsoft's emissions have risen since 2020, largely attributed to demands of AI processing.
Amount of Ireland's electricity consumption expected to be accounted for by data centres within three years.
Big tech’s big carbon conundrum
Google aims to achieve net-zero emissions by 2030 but acknowledged in its 2023 sustainability report that it is no longer “maintaining operational carbon neutrality.” This shift is largely attributed to a significant increase in emissions, driven by the growing use of AI.
Microsoft, with an even more ambitious target to become carbon negative by 2030, faces similar challenges. Its emissions have risen by 29% since 2020, primarily due to the expansion of AI-optimised data centres, as revealed in its May sustainability report.
As AI use increases, big businesses are grappling with how to manage data centre energy consumption without undermining their emissions goals. Nuclear energy is gaining traction as a potential solution.
Amazon, for example, has signed multiple agreements to develop small modular nuclear reactors (SMRs) in Washington and Virginia to support its push for carbon-free energy. Similarly, Microsoft secured a 20-year Power Purchase Agreement (PPA) with energy firm Constellation to revive the Three Mile Island nuclear plant in Pennsylvania after its 2019 closure.
Additionally, Microsoft and Google have launched a new innovation hub in Denmark focused on reducing data centre emissions. The hub is exploring renewable alternatives to diesel generation and heat reuse. Another focus is addressing Scope 3 emissions in material supply chains, including concrete, steel and aluminium.
However, Pinto stresses that even if the largest AI companies transition all their data centres to renewable energy, the global push for electrification already is placing immense strain on existing energy supplies and infrastructure.
And the growing demand for data centres is only exacerbating this pressure. Last year, for instance, South Dublin County Council denied Google permission to build a new data centre, citing the inability of the region’s energy grid to handle the increased demand. According to Ireland’s national Energy and Climate Plan, data centres could consume 31% of the country’s electricity within three years, surpassing homes at 28%.
Pinto adds: “Businesses that seek to adopt AI must ensure that their data centre and AI services suppliers are not only transitioning to renewable energy use, but are also investing in energy efficiency technologies and more sustainable cooling alternatives.”
Once we have better data, we can explore regulatory frameworks for different AI applications — but that is further down the line. Right now, the priority is gathering the necessary information.
Salesforce’s sustainable AI lead Boris Gamazychikov
The push for energy-efficient AI models
Energy efficiency presents both a challenge and a solution in the AI debate. Businesses seek to adopt more energy-efficient AI, but a lack of data and transparency in data centre supply chains makes it difficult to accurately measure the carbon footprint of AI models.
Research from Capgemini found that while nearly half of business executives believe that the use of Gen AI is increasing their emissions, only one in eight businesses are currently able to effectively track this.
Capgemini’s head of global sustainability services and corporate responsibility Cyril Garcia emphasises that the main issue today is the lack of transparency. Many organisations cannot accurately calculate the carbon footprint of their AI usage because they struggle to access clear emissions data from providers.
Garcia adds: “Strong data governance, and cross-functional collaboration are key to making sure AI is leveraged sustainably.”
Several major tech companies, including Capgemini and Salesforce, are developing initiatives to improve the energy efficiency of both the AI they use internally, and the AI integrated into their products and services.
Garcia explains that Capgemini is helping clients enhance performance through tools that are focused on energy-efficient design and development. This includes optimising Large Language Models (LLMs) during training, monitoring carbon impact during inference, and implementing measures to reduce environmental effects.
“As an industry, we need clear, collaborative standards to enhance transparency while fostering innovation,” says Garcia.
Several major tech companies are developing initiatives to improve the energy efficiency of both the AI they use internally, and the AI integrated into their products and services.
Strong data governance, and cross-functional collaboration are key to making sure AI is leveraged sustainably.
Cyril Garcia, head of global sustainability services and corporate responsibility, Capgemini
AI, accountability and the path to sustainability
Salesforce’s sustainable AI lead Boris Gamazychikov also emphasises that transparency is the foundation of effective AI regulation.
To address this challenge, Salesforce launched the AI Energy Score at the AI Action Summit in Paris earlier this year. The AI Energy Score is a tool that evaluates energy consumption for AI models during both model training and inference.
It aims to drive a market preference for energy-efficient AI and promote sustainable development. The platform includes a five-star rating system and a label generator, allowing AI developers to create standardised labels displaying their models’ energy scores.
Additionally, it enables developers—whether working with open source or proprietary models — to submit their models for assessment.
Salesforce has also integrated AI across its products and internal operations, leveraging its in-house AI research team, which has been developing AI solutions for over a decade. The company doesn’t operate its own data centres and instead procures cloud services from other companies.
For its own AI models, Salesforce is focusing on developing smaller, domain-specific models tailored to customer needs. Rather than building large, multi-purpose models, Salesforce prioritises efficiency by designing AI that excels in specific tasks, making it more energy-efficient.
Gamazychikov informs edie that Salesforce is finalising its emissions calculations and does not expect a significant increase. However, it acknowledges industry trends and is exploring ways to reduce its impact.
Salesforce has joined the Coalition for Sustainable AI, a global multi-stakeholder initiative launched at the AI Action Summit in Paris. The coalition aims to ensure that AI development, including both hardware and software, aligns with environmental sustainability.
Its mission is twofold: to prevent AI from exacerbating environmental harm and to leverage its potential to accelerate decarbonisation and conserve and enhance nature.
Other members of the coalition include Capgemini, Altair, Ericsson, IBM, Lenovo, L’Oréal Group, Nvidia, Philips, Schneider Electric, TotalEnergies and Veolia.
As part of the Coalition, these organisations have outlined plans to gather data across the entire lifecycle of the AI they deploy.
“Ultimately, everything comes down to transparency and understanding the current landscape,” says Gamazychikov.
He adds: “Once we have better data, we can explore regulatory frameworks for different AI applications—but that is further down the line. Right now, the priority is gathering the necessary information.”
Gamazychikov has urged edie readers to assess whether their company procures AI from any providers, and request disclosure of energy consumption data.
He concludes: “Together, this would send a strong signal that transparency and sustainability are important.”