Asset-Light, Impact-Heavy: Inside Salesforce’s Approach to Reduce the Environmental Impact of AI Growth
Ente: Salesforce.org
Paese: US
Descrizione
0% Key Takeaways Salesforce embeds sustainability in its procurement strategy by setting ambitious expectations for suppliers and partnering across the value chain. Deploying right-sized AI models and zero-copy architecture eliminates massive amounts of energy and data waste. AI-driven tools can provide near-instant visibility into emissions and water usage to enable proactive environmental action. Driven by the rapid acceleration of AI, global data center demand is expected to more than triple by 2030 , heavily straining power grids and increasing data center water consumption by an estimated 170%. This significant environmental impact is compelling technology leaders to overhaul their energy strategies and ask tough questions. “Beyond today’s energy use, the concern is the projected ‘hockey stick’ growth in demand and whether it will outpace clean energy,” said Sunya Norman, SVP of Impact at Salesforce, who leads the company’s AI sustainability strategy. As AI infrastructure, models, and measurement standards are evolving rapidly, the industry is still learning what it will take to scale AI more sustainably. Salesforce has developed a preliminary approach to AI sustainability that continues to evolve alongside the technology itself — informed by collaboration with cloud providers, researchers, and customers. Given that sustainability is a core value at Salesforce and the company has committed to achieving science-based targets, working to reduce its environmental impact is an important priority. While Salesforce doesn’t directly operate data centers, it’s leveraging its position in the value chain to help reduce impact across layers of the stack, including procurement, compute efficiency, data intelligence, data architecture, and water stewardship. 1. Working together to build a sustainable value chain Salesforce hosts its cloud infrastructure, Hyperforce, on public providers like Amazon Web Services (AWS) and Google Cloud Platform (GCP). Because Salesforce doesn’t operate the physical facilities, emissions from these processes are considered indirect, or scope 3. This means decarbonizing the work depends on supplier influence rather than direct operational control. “This isn’t only an environmental issue. It’s also a business and community one. Companies need to be thinking now about how to manage demand, improve efficiency and transparency, and support the transition to clean energy so innovation can scale responsibly,” Norman said. “We see our role as helping bring the ecosystem together, working alongside partners across the supply chain so we can collectively make progress that none of us could make alone.” That orientation drives a shared responsibility model that uses several tools, from contractual mechanisms to direct collaboration with providers. As a major cloud customer, Salesforce works closely with its partners to encourage greater efficiency at both the facility and compute levels, using its procurement relationships to help advance broader industry progress. The collaboration tends to find natural alignment. Public cloud providers carry their own ambitious sustainability commitments, which gives both sides reasons to work through specifics as a team. “We use these deep relationships to drive change together,” said Amanda von Almen, Senior Director of Sustainability Intelligence and Decarbonization at Salesforce. “We mutually align and work together on shared goals such as data centers powered by 100% renewable energy, reducing water use, and using efficient hardware. 2. Right-sizing AI models to the task at hand The prevailing enterprise instinct is to throw massive, energy-intensive AI models at every problem. Salesforce counters this with “smart demand” — strictly calibrating model size to the specific task. Doing this reduces the load on individual data centers, as compute demand is the upstream constraint that drives downstream energy use in data centers. In collaboration with Hugging Face and academic partners, we subsequently parlayed this work into creating the AI Energy Score , a standardized Energy Star-style rating system that measures the power consumption of AI models during inference. Real-time rankings are displayed on a public leaderboard ; the goal of this metric is to provide transparency that helps organizations choose more sustainable and cost-effective models. “Where possible, we use right-sized LLMs to avoid wasting money and emissions on oversized models,” said Eric Gertsman, Director of Tech Sustainability at Salesforce. Within the Agentforce Trust Layer, Salesforce has fine-tuned smaller, more efficient models — ranging from 44 to 135 million parameters — for targeted tasks like data masking. These precision models can operate up to 99% more efficiently than massive frontier large language models (LLMs). Additionally, Salesforce’s new Hyperforce cloud architecture drives an estimated 40% increase in operational efficiency due to super-efficient data center facilitie
Settori: educazione, workforce development, environment, nonprofit, CSR
Vai al bando originale
Registrati gratis su Bandolo per trovare bandi compatibili con la tua azienda.