AI’s Energy Reckoning Has a Sustainability Upside. The People Running It Know.

At NVIDIA GTC 2026, the CSOs of Amazon, PepsiCo, and Equinix made the case that AI is both the problem and the most credible tool for solving it.

The sustainability conversation at major tech conferences tends to follow a predictable script. Commitments are announced, carbon targets are cited, and everyone applauds. The panel that took place on day two of NVIDIA GTC last week was different, not because the panelists were more bullish on their own progress, but because they were more honest about the actual complexity of the moment.

CSO Insights: Driving AI-Optimized Energy Efficiency and Data Center Performance brought together Kara Hurst, Chief Sustainability Officer at Amazon; Jim Andrew, CSO at PepsiCo; Christopher Wellise, VP of Global Sustainability at Equinix; and Tenika Versey Walker, Global Head of Sustainable Futures at NVIDIA. The conversation covered energy infrastructure, water scarcity, digital twins, and community investment. What it really covered was the central tension of this intelligence era: the same technology that is stressing global energy systems is also, in very practical ways, the most powerful tool available for addressing that stress.

As I have advocated many times, this tension deserves to be taken seriously rather than resolved too quickly in either direction.

From Grid Hog to Grid Asset

Wellise opened with a framing that should probably be the standard for how the industry talks about data centers right now. The question is not whether AI-driven infrastructure is increasing energy demand. It is. The question is whether that infrastructure can be repositioned from a passive consumer of grid capacity to an active participant in grid management.

The shift Wellise described involves bringing new energy technologies directly to site locations, including small modular reactors and hydrogen fuel cells, and rethinking the relationship between data centers and utilities as a genuine partnership rather than a procurement transaction. Equinix has reached 96% coverage of renewable energy across its portfolio. More interesting, in my opinion, is the heat recovery work. Ten thousand homes in the Helsinki metropolitan area are being heated by Equinix data centers. The Paris Aquatic Center, used for the 2024 Olympics, was heated the same way. These are not PR anecdotes. They are proof points that the infrastructure itself can be redesigned to give back to the systems it draws from.

The digital twin work Equinix is doing internally is equally instructive. By building virtual models of data center environments and simulating how cooling systems respond to different conditions, including ambient temperature, weather patterns, and hot and cold aisle configurations, the company has been able to squeeze 9% additional efficiency out of already highly optimized facilities. On infrastructure operating at this scale, 9% is not marginal. And it was AI that found it.

Water Is the Next Conversation You Need to Be Having

Kara Hurst made the point directly: if water is not on your sustainability agenda, it is time to put it there. She is right, and the data she cited makes the stakes clear. Only one in four people across the world has reliable access to safe water. That is the backdrop against which Amazon is operating a data center business that, like all large-scale compute infrastructure, has material water requirements for cooling.

The response Amazon has built is worth walking through carefully. AWS has published a water use efficiency metric of 0.15 liters per kilowatt hour across both legacy and new data centers. That number is 17% better than the 2023 figure and 40% better than 2021. In water-scarce markets including Bahrain, India, Mexico, and South Africa, Amazon is operating data centers with zero water for cooling, a constraint that has driven real innovation in cooling technology.

The water positive goal for AWS by 2030 is now 53% achieved. For India specifically, the target has been pulled forward to 2027 across all Amazon operations, a recognition that timelines need to adjust to local conditions rather than sitting at a uniform global pace. The partnership with Water.org and the network of over 40 water replenishment projects represent a model in which corporate water use creates an obligation to invest in the water systems of the communities where operations sit.

The AI angle here is direct. Amazon built an internal tool called FlowMS specifically to detect water leaks using AI-based analysis. One deployment of that tool identified a leak that was invisible to the human eye and prevented the loss of 9 million gallons of water. Given that much of the world’s water infrastructure is aging and under-monitored, the ability to deploy this kind of detection at scale matters well beyond corporate operations.

Amazon also launched a center for water and AI innovation in partnership with academics and an NGO. The framing is correct. Water scarcity is a systems problem, and AI’s ability to model, monitor, and optimize complex systems is exactly what that problem requires.

The Farm as a Test Case for AI at Scale

Jim Andrew’s contribution to the conversation came from a very different domain, and it grounded the more abstract infrastructure discussion in something concrete.

PepsiCo sources approximately 45 billion potatoes per year and works with around 300,000 farmers across its supply chain. The company is now deploying AI-powered drone scouting that can detect crop diseases up to seven days faster than conventional methods. For a farmer, seven days is the difference between a contained problem and a spread one. The economic impact compounds: earlier detection reduces agrochemical use by up to 10%, and agrochemicals are among the highest-energy inputs in any farming operation. Sustainability benefit and cost benefit land in the same place.

The AI agronomist tool Andrew described is perhaps the clearest illustration of what AI can do when it is put in the hands of people who are not AI professionals. A farmer takes a photograph of a leaf with a problem. The image is geotagged via satellite. An immediate diagnosis is returned along with a set of recommendations. Andrew described the visible moment when a farmer moves from uncertainty to certainty, and said it happens in seconds. That shift in decision-making speed, multiplied across hundreds of thousands of farms, is where aggregate impact is built.

The digital twin work PepsiCo is doing at the operational level tells a similar story. Working with NVIDIA and Siemens, the company is building physics-accurate simulations of its plants and warehouses before making changes to physical infrastructure. The results in existing facilities include a 20% increase in throughput and 10 to 15% reductions in capital expenditure. For new facilities currently under construction, the simulation is happening before the first brick is laid. The complexity of running both food and beverage operations within a single facility, optimizing for worker safety, routing efficiency, and emissions simultaneously, is the kind of systems problem that only high-fidelity simulation can resolve ahead of implementation.

The Honest Part

What gave this panel its credibility was the willingness to hold the tension without resolving it prematurely. Wellise acknowledged plainly that AI is stretching the grid. Hurst addressed the anxiety that exists within the sustainability community about embracing a technology that has real energy and water costs. Andrew noted that at PepsiCo, AI represents approximately 0.01% of the company’s total carbon footprint, and yet it generates a disproportionate share of stakeholder concern.

That disproportion matters. The risk is that anxiety about AI’s energy footprint leads sustainability practitioners to underinvest in the tool at exactly the moment when it has the most to offer. The counter-argument, made by each panelist in different ways, is that AI’s efficiency benefits across logistics, agriculture, water management, grid optimization, and materials science are likely to outstrip its energy costs by a significant margin. That case needs to be made with specifics, not reassurances, and this panel made it with specifics.

Hurst’s closing observation was the most useful. Amazon operates on the principle that it is always day one. For AI and sustainability, she suggested it is not even day one yet. It is the moment before day one, when the coffee is brewing. That framing asks for humility, for continued investment, and for genuine collaboration across sectors and institutions. Given what is actually at stake in terms of water systems, agricultural resilience, and grid stability, that is probably the right posture.