Salesforce Warns AI Data Centres Face Power Crisis by 2027

​​​​​​​Salesforce has published research warning that AI data centres could consume 3% of global electricity by 2030, as the cloud-based customer relationship management company outlines mounting environmental challenges facing the artificial intelligence sector.

The company's recently-published AI Sustainability Outlook examines how AI impacts environmental resources, with particular focus on energy consumption and water usage across data centre operations. Sunya Norman, SVP Impact at Salesforce, says the company's position in agentic AI creates responsibilities around trust and sustainability.

"As a leader in agentic AI, it is Salesforce's imperative to ensure that AI is trusted, reliable and sustainable," Sunya says. "In this outlook, we explore the current landscape, our efforts and evolving insights so far, and our preliminary path forward. We're early in the journey, but the future is being shaped now."

The research coincides with warnings from Gartner that 40% of AI data centre projects could encounter power bottlenecks by 2027, potentially disrupting AI growth trajectories and increasing operational costs. These infrastructure constraints could affect power reliability across affected regions.

Energy consumption drives environmental concerns for AI infrastructure

Data centres supporting AI workloads require sophisticated cooling systems to prevent hardware overheating, contributing to increased electricity consumption compared to traditional computing infrastructure. Fossil fuels currently provide 56% of global data centre energy, according to Salesforce's research.

The International Energy Agency reports that without transition to clean power sources, data centre infrastructure could become the fastest-growing contributor to global emissions. This scenario would reverse progress made in other sectors working to reduce carbon footprints.

Water consumption presents another resource challenge, with one fifth of data centre water sourcing coming from watersheds experiencing moderate to high water stress. AI cooling systems require substantial water volumes to maintain operational temperatures across server farms.

Mining operations for critical minerals including lithium and copper support AI compute infrastructure development. These extraction processes involve energy and water-intensive techniques that generate pollutants and contribute to environmental degradation in affected regions.

Salesforce Agentforce tackles sustainability challenges through automation

Despite environmental concerns, AI applications offer potential solutions for sustainability challenges. The International Energy Agency estimates AI solutions could reduce energy-related emissions by 5% by 2035 through optimisation of complex systems and acceleration of discovery processes.

Salesforce's Agentforce platform demonstrates practical applications for sustainability initiatives. Good360 uses the technology to build resource matching agents that automate donation routing during natural disasters, ensuring aid reaches priority areas efficiently.

Stephane Moulec, Chief Technology Officer at Good360, says the technology addresses waste in disaster response systems. "Globally, a significant amount of goods that could be matched to disaster survivors end up going to the landfill. Good360 is here to change that," Moulec says.

Rare has developed an AI-powered regenerative agriculture coach using Agentforce, providing farmers with real-time guidance based on crop conditions, weather data, and agroecological factors. The platform's initial phase targets 5,000 farmers whilst reducing staff time requirements by 40%.

Groundswell applies AI technology to community solar scaling, helping approximately 30,000 households reduce electricity costs through energy burden reductions of 50%.

Three-pillar approach guides Salesforce AI sustainability strategy

Salesforce structures its sustainability approach around smart demand, efficiency improvements and clean supply chain management. The smart demand pillar emphasises selecting appropriate AI tools for specific projects rather than applying uniform solutions across different use cases.

The company recommends avoiding unnecessary AI usage by implementing simpler algorithms where complex models provide minimal additional value. Transparency with customers regarding data centre power consumption patterns helps create flexibility in energy usage timing.

Efficiency improvements focus on developing smaller AI models that can outperform larger general-purpose systems in specific applications. Purpose-built agents often achieve superior speed and efficiency compared to comprehensive models across targeted tasks.

Quantisation, distillation, and pruning techniques enable model streamlining without quality reduction whilst decreasing compute requirements. Deploying smaller models on edge devices reduces pressure on centralised data centres, contributing to energy savings across AI infrastructure.

Clean supply chain management involves sourcing sustainable suppliers and embedding environmental considerations across procurement processes. This approach extends sustainability impact beyond individual company operations into broader supply networks.

"By sharing our progress openly, we aim to spark transparency and inspire collective action. A sustainable future with AI is within reach, and I'm optimistic about what we can achieve together," Sunya says.