Enterprise AI customers have incredible influence over how the technology evolves in terms of energy.
That’s the thesis of Sustainable AI Group, a new research and advisory firm founded by Sasha Luccioni, formerly head of AI and climate at Hugging Face, and Boriz Gamazaychikov, former head of AI sustainability at Salesforce.
These corporations — those that are rapidly embedding AI into everything from HR workflows to customer support — number in the thousands. Most aren’t training their own models or running their own GPU clusters; they’re buying access to models via APIs and off-the-shelf applications. That fundamentally limits their direct control, because they can’t choose the GPU or even the data center where a model runs, Gamazaychikov said, but it doesn’t mean they’re powerless.
Enterprise companies are the ones whose budgets will ultimately justify the trillions of dollars now flowing into AI infrastructure, he explained, and the question is whether those buyers will use that leverage to demand more sustainable, transparent systems. SAIG will exist to help them direct that leverage.
To begin, the firm will help companies map where they’re actually using AI, conduct an initial estimate or baseline of the environmental impact of that usage, and set up capabilities for ongoing measurement and tracking internally.
But it’s not just an advisory company. SAIG will conduct research into AI’s environmental footprint as models evolve, grounded in Luccioni’s work on measuring the climate impact of AI systems. For example, the group plans to build tools that can automatically direct different use cases to models that are more efficient and fit-for-purpose than the one-size-fits-all frontier option.
Gamazaychikov said one initial goal is to enable enterprise companies to use their procurement processes and contracts to shift the AI industry toward more efficient energy strategies. These companies, he argued, should be baking environmental criteria into RFPs and master service agreements, asking providers what they measure, how and where their models are trained and deployed, and what the upstream energy and infrastructure impacts look like. But so far, that’s largely not happening.
One key clause to include up front, he said, is a requirement for disclosures around off-grid data centers, particularly those that are gas-powered. Though likely to be a small part of the market in the long-term, behind-the-meter generation could be a particularly risky part of the supply chain for enterprise companies, Gamazaychikov said, even if they eventually plan to get grid-connected.
The enterprise impact
Even as tech giants like Microsoft appear to be wavering on some climate commitments in favor of speed to power, the enterprises racing to adopt AI are running into a different kind of pressure: further downstream, in the form of employees and their own customers.
“I’ve heard from non-tech companies that this is one of the first topics on sustainability that they’ve heard brought up at town halls, and I’ve heard that story multiple times,” Gamazaychikov said.
Large corporations, he added, are under pressure from investors and leadership to embed AI into their daily work. For many of these companies, their sustainability programs peaked in the early 2020s, when they rushed to announce net-zero plans and science-based targets. Then generative AI arrived and “took the air out of the room,” shifting executive attention elsewhere.
Despite that change, Gamazaychikov said those targets are actually a good example of how enterprise companies can flex their leverage. The cascade from large buyers to their suppliers pushed thousands of companies to set their own targets, and helped drive the renewable energy market, he said.
AI could follow a similar arc: If enterprise buyers begin insisting on more transparency and optionality — asking for clarity on model‑level and data‑center‑level impacts, and rewarding better practices — they can shift the trajectory of the industry.
“The long-term impact would be for the AI industry to not just have one answer for everybody — scale and growth at all costs,” Gamazaychikov said. Instead, buyers could force a more nuanced system where providers offer more sustainable, less costly options alongside the most general, energy-hungry models.
Challenges with the current political climate aside, he added, the focus on the AI boom, and the rise of community opposition to data centers, presents an opportunity for sustainability teams to regain relevance inside their companies.
The challenge, however, is that the enterprises with the most leverage are often furthest from the consequences. Most large buyers are several steps upstream from data center protests, debates over how to add capacity, and utility decisions around how to serve new large loads. The goal of SAIG, Gamazaychikov said, is to make those connections clear, so that decisions made in contract negotiations or model selection can, eventually, improve conditions downstream.
