Beyond tokens and model pricing
Every executive sponsor of an AI initiative eventually asks some version of the same question: What is AI actually costing us and is the value it creates worth it?
Over the past two years, organizations have become increasingly sophisticated in answering this question. AI FinOps has emerged as a discipline for tracking model usage, token consumption, licenses, infrastructure costs and the growing portfolio of AI investments across the enterprise. Much like cloud FinOps before it, the goal is to create visibility into AI costs and determine whether the value created justifies the spend.
Yet beneath this recent layer of financial governance sits another ledger that most organizations rarely consider. While organizations may scrutinize AI spending down to the token, few can connect that same AI activity to its environmental impact and weigh that impact against the value it creates.
A day of prompts, visualized
Consider a day with AI at a typical medium-to-large sized enterprise. By 8:00 a.m., account managers are drafting client emails with an AI assistant. Developers have coding agents refactoring applications, each task spawning dozens of model calls. An automated contract analysis system classified 5,000 documents overnight. Finance is summarizing quarter-end close notes, HR is screening position descriptions, and someone in marketing has asked a reasoning model to stress-test a pricing strategy, generating thousands of hidden “thinking” tokens before a visible answer appears.
The ability of AI to perform these tasks feels weightless, even magical. But the reality behind the scenes is surprisingly physical.
Each prompt travels across a sprawling network infrastructure to GPU racks in a data center no employee will ever see, perhaps thousands of miles away where a large, 100 MW data center may be drawing up to two million liters of cooling water per day for on-site cooling and other indirect uses (Congressional Research Service Report). There, the processing activity initiated by the prompt comprises thousands of calculations, each one consuming electricity, generating CO2 emissions and releasing heat that must be removed (often requiring water consumption) to keep the equipment running before the response is completed and sent back to the user.
Viewed this way, a prompt is less like typing a message and more like activating a complex industrial process. Yet the impacts associated with that process remain largely invisible and disconnected from the activity that creates them. The environmental cost of millions of inferences generated over the course of a workday, a work week or a fiscal year is out of sight, and therefore out of mind.
These impacts are not irrelevant simply because they are invisible. And even if the enterprise chooses to ignore them, there’s no guarantee that others will.
Why this matters now
For many years, organizations could reasonably treat the environmental consequences of computing as somebody else’s responsibility. Technology providers managed the infrastructure, utilities generated the power and sustainability teams focused on more direct impacts, like visible resource use within the enterprise and its supply chain. Several factors are changing this:
- Scale – AI is driving unprecedented growth in compute demand. According to a 2026 report by United Nations University Institute for Water, Environment and Health, global data center electricity consumption is expected to more than double by 2030, reaching approximately 945 terawatt-hours annually. The report estimates that AI will account for 40% of that demand. The water footprint associated with this amount of electricity over the same period is projected to reach 9.3 trillion liters annually, enough to meet the annual water needs of approximately 1.3 billion people in Sub-Saharan Africa.
- Public awareness – For decades, the infrastructure behind the digital economy was largely invisible to the public — until AI. Today, proposals for new data centers increasingly attract public scrutiny and, in some cases, community opposition driven by concerns over energy demand, water usage and strain on local infrastructure. Organizations that appear indifferent to this public mood may find themselves in a spotlight they don’t crave.
- Stakeholder expectations – Investors, customers, employees and regulators are increasingly interested in understanding the broader consequences of technology. The conversation is expanding beyond productivity to include questions about impact, transparency and responsible use.
- Sustainability commitments – Many organizations have public sustainability and responsible-business commitments and have set emissions targets. As AI becomes embedded in operations, leaders are increasingly evaluating how AI adoption affects these goals and obligations. Impact visibility gives leaders another lens through which to evaluate AI investments and additional information with which to supplement an existing environmental ledger.
By contrast, those that proactively address AI’s resource impacts can turn a potential liability into a source of credibility and differentiation.
The problem with the “invisible” ledger
If understanding AI’s environmental footprint were straightforward, most organizations would already be measuring it.
In many respects, AI’s environmental visibility today is where cost visibility was during the early years of cloud adoption. It took a few years to develop tagging, ownership, budget alerts and other governance measures to bring proliferating cloud costs under control.
AI’s environmental footprint is remarkably similar. Multiple models, cloud platforms and AI-enabled applications generate impacts beyond most organizations’ visibility, while limited provider transparency makes them difficult to measure. As use of AI accelerates, organizations may find themselves expanding their environmental footprint without a clear understanding of which activities, applications or providers are driving the increase.
The good news is that we know this problem is solvable. Much as cloud costs gave rise to the discipline of FinOps with transparency, governance and accountability, AI’s environmental impact requires the same. For many companies, it means following a road already traveled.
“Visibility enables better management decisions. An AI application that helps optimize a complex process, shorten a supply chain or reduce material waste may more than justify its resource impact. But not every AI use case creates the same value. Understanding both the resource demands and the benefits gives leaders a stronger basis for deciding when and where AI is worth the investment.” – Steve Wang, Sustainability Audit Leader, Protiviti
It’s time for an executive conversation
The way to think about AI’s environmental impact is as the next key piece of information to inform better governance and decision-making. The first set of questions focused on risk. The second focused on cost. The next one is impact.
Organizations that begin asking these questions now will be better positioned than those that wait until stakeholder pressure, reporting requirements or public scrutiny force the issue. Start here:
- Boards: How does our AI strategy align with our corporate responsibility commitments, and what visibility do we have into its broader impacts?
- CEOs: As AI becomes central to growth and innovation, how do we ensure adoption aligns with the organization’s values and public commitments?
- CIOs: What information can our cloud and AI providers provide regarding energy use, emissions, and water and resource consumption associated with AI workloads?
- CFOs: If stakeholders ask us to account for AI-related environmental impacts, what data can we currently provide and where are the gaps?
- Chief sustainability officers: Should AI’s sustainability impacts be assessed and managed as a distinct issue, or integrated within broader sustainability programs, and do existing reporting frameworks adequately capture those impacts?
- Procurement leaders: Are environmental considerations and disclosure requirements included in AI vendor evaluations?
- Executive teams: In which cases is AI creating enough value to justify its resource consumption, and would better information change the way we design, source or use it?
A more complete measure of AI value
As organizations become more sophisticated in understanding AI’s financial costs and work to optimize them, through task-appropriate model selection, application design, prompt efficiency, workload optimization and procurement decisions, they may start to realize that financial and environmental efficiencies overlap. But they should not assume that a lower AI bill automatically translates into a better footprint. Environmental impacts require their own measurement, reporting and governance disciplines to build trust in the conclusions.
What matters is having enough visibility to understand the tradeoff. Ultimately, making AI’s environmental ledger visible gives leaders a better answer to the question “Is the value worth it?” As AI becomes more deeply embedded in the enterprise, organizations that can answer that question will be better positioned to make informed decisions about where and how AI is deployed.
Learn about Protiviti’s AI services: https://www.protiviti.com/us-en/artificial-intelligence-services.
Mattie Newman, a Manager with Protiviti’s sustainability practice, contributed to this content.
