Governments Can Advance 'Greener AI' Through the Power of Procurement

In 2026, public officials are under alarming pressure to integrate AI systems into public services and agency operations. At the same time, communities and governments alike are asking that AI and data center decision-making take into account the harmful environmental externalities of AI, and rightly so: each Microsoft Copilot contract or bespoke large language model sits atop computing infrastructure that is devouring ever more land, power, and water.

To address these competing demands, I worked with Data & Society and the GovAI Coalition to develop a guide that helps government agencies consider environmental sustainability in AI adoption. The result, “Greening AI in the public sector: a handbook for procurement,” is based on 10 months of fieldwork and 5 months of co-design interviews and focus groups with government staff and green software practitioners. The guidebook was born through Data & Society’s partnership with the Coalition, which collectively represents over 900 local, state, county, and federal agencies. Increasingly, we have been hearing from members who are concerned about the environmental impacts of AI infrastructures but lack the information and resources that would help them take action. This handbook is that resource.

The guide is designed to respond to conditions on the ground, such as staffing and budgetary limitations, a de-prioritization of environmental sustainability, and a predatory corporate marketplace. Within these constraints, it proposes low-lift actions that can still make an impact alongside more advanced considerations The guide covers the basics of greener computing, situates this technology within common procurement pathways, and walks readers through vendor engagement and evaluation. In response to a consistent thread that we heard, the handbook also explicates greener software’s associations with cost savings, performance benefits, and customer autonomy, helping agency workers make a strong argument for this shift even in settings where sustainability is not the most tractable concern.

While the handbook is focused on the public sector, any organization seeking to make greener AI adoption decisions can apply the lessons it provides. Moreover, the overview of sustainable computing and government procurement contexts provides insight for anyone interested in how public officials can and should make decisions about technology in the public interest. As I conclude my research, I am left with five takeaways about the broader implications of “greening AI” in the public sector.

There are advantages of greener computing that can be leveraged without specific expertise or significant effort.

Implementing sustainability into a tech procurement process does not need to be difficult. The world of greener tech is moving fast, and there are readily-adoptable technical resources that enable developers to assess and report information about their products.

Our handbook synthesizes existing measurement schemes to recommend the most straightforward actions — asking a vendor to share average kilowatt hours per token, for example, or to calculate a value within an industry scoring system — and also provides a comprehensive collection of additional resources for further engagement. Some of the actions that we suggest are as simple as adding a “yes or no” question to a vendor solicitation; in an agency where subject knowledge or leadership buy-in are low, these are manageable changes. While a single checkbox on a form will not individually transform the marketplace, these steps, in aggregate, can add up.

The public sector can exercise significant market pressure.

Individually and collectively, towns, cities, and states wield enormous buying power. In fact, commercial suppliers reported to me that an uptick in customer inquiries about sustainability has prompted them to quantify and improve upon their performance in this arena. Yet, many officials with whom I spoke confessed an exasperation that corporate agendas are determining their agencies’ priorities and decision-making. As we see from suppliers’ responses, capitulating to a perception of powerlessness is a mistake. The market pressure that localities may exert has the potential to direct the private sector — not the other way around. Government actors have both the mandate and the obligation to use this responsibility for good.

We need principled governance systems in addition to efficient hardware systems.

Substantial work is being done to develop low-impact computing. Still, society needs to intervene far upstream from making tweaks to kilowatt hours and parameter counts. There is a growing movement toward “frugal AI” — an orientation toward the technology that advocates “using AI as much as necessary, but as little as possible.” The first international guidance on the subject defines frugal AI as a service for which:

• the necessity of using an AI system rather than another less resource-intensive solution to achieve the same objective has been demonstrated;

• best practices are adopted by the producer, supplier, and customer to reduce the environmental impact of the service using an AI algorithm;

• uses and needs aim to remain within planetary boundaries and have been previously questioned.

Such an approach looks at the impacts of AI’s physical infrastructure as well as the social and ecological effects of its particular application. For AI systems that touch public services, these effects can have life-or-death repercussions. A robust process of problem definition and solution justification foregrounds community needs and demands accountability to the public. As a white paper from the Frugal AI Hub at Cambridge University’s Judge Business School summarizes it, AI should be “Fair, Few, Frugal.”

During my interviews, officials from diverse localities and agencies described an amorphic pressure to adopt AI systems to be more competitive, “AI-ready,” or upskilled; in effect, because everyone else is doing it. This perceived imperative belies the dread and dissatisfaction with the technology that many of my interviewees expressed. The adoption of this technology is not inevitable. Frugal AI is government strategy in other parts of the world; there is no reason for us to cede our imaginations in the US.

“Greener AI” doesn’t address the full ecological and planetary threat posed by Big Tech.

One worry about conversations on “greener AI” is that they have the potential to locate the problem in the wrong place. It is a mistake to think of the environmental consequences of AI as restricted to data centers. For one, these harms also include how AI is being used in service of other familiar climate villains; notably, to turbocharge fossil fuel extraction. More fundamentally, however, is the broader picture: that the sale and specter of this technology are effectuating a consolidation of corporate power and individual wealth heretofore unknown in human history.

Big Tech is the darling industry du jour, but the sector is not exceptional among its corporate leviathan forebears. It is the inheritor of a decades-long trend of oligopolistic neoliberal governance that has achieved its most extreme form in the hands of a putatively corrupt Trump administration. The attention on AI should also include how its market function is calcifying the growth-obsessed, extractive, polluting economic system that generated and continues to fuel climate change. A rising tide lifts all boats, and the fortunes of the most environmentally-harmful sectors — oil, mining, steel, industrial agriculture — are rising along with Big Tech.

While AI, as an asset, helps consolidate markets, it also produces its own class of super-polluting billionaires. The enormous environmental footprint of Amazon Web Services, for instance, does not account for the fact that Jeff Bezos himself expels more carbon in 90 minutes than you will in your lifetime. Elon Musk’s two private jets emit the equivalent of 5,437 years’ worth of emissions for someone in the world’s poorest 50%. And this is to say nothing of the sheer market maleficence of these tech oligarchs’ wealth: billionaires are invested disproportionately in the dirtiest industries, and an average billionaire’s investment emissions equate to almost 400,000 years of consumption emissions by an average person.

Procuring more efficient computing technology is an important step but will do little toward genuine climate progress unless it is coupled with a dramatic restructuring of corporate influence in the public sector. Massive sums of public money are going toward contracts with what are now the largest companies in the world. This is, in effect, a mass transfer of wealth from people to corporations without consent. It is an outrage that should serve as an urgent call to action for everyone.

The public has more power to shape the industry’s trajectory than we think.

My interviews and focus groups revealed just how many hardworking public officials are earnestly grappling with the AI industry’s intrusion into government services. They are searching for reliable information amid a glut of hype, for actionable examples from other localities and, frankly, for justifications to push back. Time and again, I heard how local opposition reversed a particular AI procurement decision; how social media blowback to an AI announcement embarrassed a public official or changed agency discourse; or how a community engagement process enabled intentional AI policy-setting. AI fervor is certainly gripping government agencies in the same way it’s hounding those in the private sector, but there are skeptics and dissenters everywhere who just need support. Public advocacy helps empower officials who might otherwise be drowned out in internal government discussions.

For this reason, residents command the greatest influence at the local, regional, and state levels. The federal government under the Trump administration has displayed its intent to weaken the public sector while entrenching the control of private actors, but officials in other tiers of government are particularly receptive and susceptible to community involvement. Communities can and should be asking their town and city offices about their responsible AI strategies. They should be requesting information about AI contracts under consideration or in place. They should hold officials to account for partnerships with major tech companies and refuse to let their taxpayer money line billionaires’ pockets.

My immersion in the world of greener AI opened my eyes to approaches to technology that maximize public good while minimizing planetary harm. It is a discourse that we desperately need, not just in government settings but throughout nonprofit and private sectors and beyond. I hope, though, that the concept will be just one provocation in a larger social reckoning over how the public sector orients itself toward this industry and its AI project. We have more power than we might realize to leverage the civic and public sectors to fight corporate interests and demand a sustainable future.

“Greening AI in the public sector: a handbook for procurement” was supported by the Internet Society Foundation, and was written for Dr. Tamara Kneese's Climate, Technology, and Justice program at Data & Society. The author wishes to express gratitude to Dr. Kneese and Dr. Meg Young for their partnership on this project. The opinions expressed here are solely those of the author and do not necessarily reflect the views, positions, or policies of the Data & Society Research Institute, the GovAI Coalition, or the author’s other affiliated professional organizations.