A prominent Montreal AI researcher and a former Salesforce executive are launching a new firm to advise big businesses on how to use AI more sustainably. The move comes amid concerns about the environmental impact of the AI era and the data centres being built to power it.

“The AI industry is obviously on an unsustainable trajectory,” said Boris Gamazaychikov, CEO of the new Sustainable AI Group (SAIG). A former head of AI sustainability at Salesforce, he’s teaming up with Sasha Luccioni, who recently left open source giant Hugging Face and led widely recognized work measuring the emissions of AI models. 

Montreal-based SAIG will help clients analyze the efficiency of the AI tools they’re using, and collect the right data to evaluate their technology. It could conduct experiments on their behalf, or work with them to develop internal policies.

One way for businesses to help save money and the planet is by getting smarter about the AI tools they’re using. The most advanced or “frontier” models require the most energy for developers to train, and also typically cost the most for clients to use. But such powerful systems might not be the best fit for every task for which a business might use AI. 

Take a brand that wants to gauge buyer sentiment by analyzing comments on its website. “You don’t need to use a frontier LLM for that,” said Luccioni, chief scientist at SAIG. She added that the firm could instead use a smaller, open source system running on its own hardware—closer to free, and much more energy-efficient.

Tech giants, financial investors and energy companies are piling into data centres, seeking to capitalize on the massive demand for processing power to train and run the most advanced AI models. 

That need will continue to grow as AI developers try to make breakthroughs by continually increasing the amount of data and compute they use. Some businesses are also promoting “tokenmaxxing,” whereby staff are encouraged to use AI as much as possible in an attempt to boost their productivity and generate new products. The proliferation of AI agents and of systems that take multiple steps to reason through problems are also increasing compute demand.

While Silicon Valley firms and their shareholders were once big proponents of the environmental, social and governance (ESG) movement, many have backtracked on sustainability commitments as political pressures have changed and the AI frenzy has heightened. Many large tech firms, while remaining major buyers of renewable power, are also turning to burning natural gas to power their compute ambitions.

Gamazaychikov said “very few” firms are driving the AI buildout. “In reaction to that, we are seeing a growing backlash,” he said. He cited proposed data centres stalling in the U.S. due to community opposition, with bipartisan support for the pushback.

Local policymakers are starting to question the benefits of data centres, while governments are setting sustainability requirements for their own AI procurement, Luccioni said. Meanwhile, tech workers are pushing their employers to live up to ESG commitments. “You have both external and internal pressure,” she said. 

The SAIG co-founders say businesses that buy from those tech firms can also help encourage sustainable AI adoption and development. “Enterprises are looking to deploy AI more responsibly, and they’re going to have a lot more power as this industry matures,” Gamazaychikov said.

SAIG will help clients do that. For example, a customer might engage the firm to measure the carbon footprint of its meeting transcription software. SAIG could develop a testing methodology to account for different AI models and lengths of recordings, then test open source alternatives to get a sense of the cost and accuracy of each. “We help them make the most-informed choice,” Luccioni said. With client approval, SAIG also plans to publish its research findings.

The co-founders say SAIG is already seeing interest from businesses in Canada, the U.S., France and the U.K. It’s initially targeting firms that have strategies to become more sustainable and to rapidly adopt AI, but haven’t yet joined the two sides together. “We speak both languages,” Luccioni said.