As organizations grapple with the dual challenge of digital transformation and environmental responsibility, Matilda Bailey has emerged as a leading voice at the intersection of networking and green technology. With a career dedicated to the nuances of cellular and next-gen wireless solutions, she brings a unique perspective on how the physical infrastructure of our digital world dictates its carbon footprint. In this discussion, we explore the intricate balance between using artificial intelligence to solve climate issues and managing the significant energy demands that these same models impose on global power grids.
We dive deep into the practical applications of machine learning in physical systems, the reliability of various ESG reporting platforms, and the strategic shift toward smaller, more efficient AI models. Bailey also addresses the organizational friction between IT departments and sustainability teams, offering a blueprint for a more integrated, accountable approach to technological growth.
Artificial intelligence is often framed as a double-edged sword that can optimize energy use while simultaneously draining the power grid. How do you see enterprise leaders navigating this paradox where the solution to inefficiency might actually be part of the problem?
The paradox you are describing is the central tension in modern sustainable IT, and it requires a move away from seeing AI as a generic “silver bullet.” For senior leaders, the challenge is no longer about whether to use AI, but how to deploy it in a way where its environmental benefits clearly outweigh the operational footprint of the hardware itself. We see large language models consuming massive quantities of electricity and water for cooling, yet these same systems can identify efficiencies in a power grid that a human analyst might never spot. The most successful organizations I work with avoid pursuing AI for its own sake; instead, they target measurable operational problems where they can quantify actual improvements in energy savings or resource consumption. It is about being a practical strategist—if the energy required to train a model is greater than the carbon it saves over its lifetime, then that model is a net loss for the planet.
When we look at the specific areas where AI can have the most immediate impact, such as energy grid optimization and building management, what are the core requirements for a model to move beyond a simple demonstration and provide real-world value?
To move a project from a flashy demo to something that actually impacts the bottom line and the environment, you need to hit three specific markers: a measurable physical output, data that is clean enough to trust, and a decision-maker who is ready to act on the insights. In the context of energy grids and HVAC systems, we are looking for very specific data points like kilowatt-hours or equipment loads. AI creates the most value when it is applied to these physical systems where machine learning models can continuously adjust cooling infrastructure based on real-time environmental conditions. Without that link to a physical outcome—like reducing a building’s energy draw or optimizing a data center’s cooling cycle—the AI is just crunching numbers in a vacuum. If you don’t have a person or an automated system willing to pull the lever based on what the model says, the entire exercise becomes a drain on resources rather than a tool for conservation.
Predictive maintenance is frequently cited as a mature use case for sustainability. Could you walk us through the ripple effect that preventing a single piece of industrial equipment from failing has on an organization’s overall carbon footprint?
Predictive maintenance is a fascinating area because its environmental benefits are often a byproduct of its financial logic. When we use machine learning to analyze telemetry data from industrial assets, we can detect tiny anomalies long before a catastrophic failure occurs, which allows us to extend the life of that asset significantly. By replacing parts only when necessary rather than on a fixed, often premature schedule, a company drastically reduces the waste associated with manufacturing and transporting new components. Think about the manufacturing overhead and the transport emissions involved in an emergency repair versus a planned, optimized maintenance window. Businesses end up saving costs and carbon simultaneously because they aren’t stuck in a cycle of “break and replace,” which is one of the most resource-intensive ways to operate.
As supply chains become more complex, how is AI being used to refine logistics and inventory management, and why are these specific areas considered so measurable in terms of sustainability?
The beauty of applying AI to logistics and supply chain management is that the results are almost immediately visible in the data, specifically through metrics like fuel consumption and total mileage. By using better demand forecasting, companies can avoid the trap of overproduction, which is a massive source of hidden waste in almost every industry. When you have a model that accurately predicts how much inventory is needed and where, you can optimize transportation routes to ensure that trucks aren’t moving half-empty or taking redundant paths. These reductions in excess inventory and mileage translate directly into lower emissions, making it one of the most quantifiable areas of sustainable technology. It turns the nebulous goal of “being greener” into a concrete set of numbers that a CFO can understand and appreciate.
There is a lot of buzz surrounding AI-powered ESG reporting tools, yet some experts warn that these platforms might be overpromising. What are the risks of relying too heavily on generative AI to handle complex sustainability disclosures and Scope 3 emissions data?
The danger with using generative AI for ESG reporting is the “hallucination” factor where a model can produce confident, beautifully formatted estimates based on incomplete or untrustworthy data. Scope 3 emissions, which involve the entire value chain, are notoriously difficult to track, and pointing a model at inconsistent data from dozens of different suppliers is a recipe for disaster. While natural language processing is excellent for automating the collection of data from invoices or shipping records, it should never be the final word on a disclosure. I always tell my clients that a simple spreadsheet validated by a diligent compliance officer is infinitely more reliable than a summary generated by a power-hungry LLM-based system. If the underlying data is flawed, the AI will simply help you be wrong faster and with more misplaced confidence.
With so many platforms on the market—from IBM Envizi and Microsoft Cloud for Sustainability to specialized tools like Persefoni and Watershed—how should an enterprise evaluate which software fits its specific decarbonization goals?
The market is currently flooded with high-quality options, so the choice really comes down to the existing ecosystem of the company and the specific depth of reporting required. For instance, if an organization is already heavily invested in the Salesforce ecosystem, Net Zero Cloud is a logical step because it integrates emissions data with existing operational and customer information. On the other hand, platforms like Persefoni or Watershed are specifically built for high-stakes carbon accounting and managing complex Scope 1, 2, and 3 emissions across global supply chains. Some tools like Diligent ESG focus heavily on the governance and board reporting side, while Pulsora and IBM Envizi lean into performance analytics and automated data collection from various business systems. The key is to look for a tool that doesn’t just store data but offers reduction scenario modeling, helping the company plan for a future where they actually hit their climate targets.
We often hear about the energy intensity of training large AI models, but what about the day-to-day operational costs like water usage and the lifecycle of the hardware itself?
The environmental cost of AI goes far beyond the electricity bill; it involves a massive amount of water for cooling data centers and a significant amount of “embodied carbon” in the specialized chips and servers required to run these models. As data centers expand to meet AI demand, they place a tremendous strain on local water supplies and power grids, which is something sustainability leaders have to account for in their reporting. We also have to consider the full lifecycle of the computing equipment, which is often replaced more frequently than standard hardware to keep up with the processing requirements of modern AI. To truly measure the impact, an organization needs to look at marginal carbon intensity and water usage effectiveness, rather than just basic power usage metrics. If we ignore the resources consumed to build and cool the machines, we are only seeing half of the environmental picture.
Given the high resource cost of frontier models, why is there a growing movement toward using “small models” or classical optimization for sustainability tasks?
The reality is that most enterprise sustainability use cases simply do not require the massive scale of a frontier LLM; they can be handled much more efficiently by a small, specialized model or even classical optimization techniques. If you are trying to track carbon emissions or optimize a building’s HVAC system, using a massive general-purpose model is like using a sledgehammer to hang a picture frame—it’s overkill and incredibly wasteful. Smaller models can produce comparable, and often more accurate, results for specific tasks while consuming a fraction of the energy. By right-sizing the technology to the task, organizations can achieve their sustainability goals without adding unnecessary weight to their own carbon footprint. It is a more disciplined approach to AI that prioritizes efficiency and precision over the novelty of using the latest, largest model.
There seems to be an accountability gap in many organizations where IT manages the technology, facilities pays the electric bill, and the sustainability team reports the emissions. How can companies bridge this gap to ensure AI’s environmental costs are actually managed?
This organizational fragmentation is one of the biggest hurdles to sustainable tech because when no single group fully owns the environmental cost, the costs tend to spiral out of control. You have a situation where IT enables the access, various departments pay for the licensing, and facilities covers the power bill, leaving the sustainability team to report on emissions they have very little power to change. To bridge this, organizations need to start calculating a “net environmental impact,” which compares the emissions a specific AI project saves against the energy and resources it took to build and run that project. We need to move toward a model of shared accountability where the cost of the carbon is factored directly into the department’s budget. Ultimately, the simplest metric we have is whether anyone actually acts on the model’s output—because a model that generates insights that are never used is just pure carbon cost.
What is your forecast for the role of AI in corporate sustainability over the next five years?
I believe we are moving toward a period of “radical transparency” where AI will no longer be used just to estimate emissions, but to provide real-time, granular verification of environmental impact across entire global networks. We will see a shift away from massive, energy-intensive general models toward a decentralized network of highly specialized, “green” AI agents that are optimized for low-power environments and run during periods of high renewable energy availability. Scheduling compute-intensive processing to align with when the sun is shining or the wind is blowing will become a standard operational procedure for responsible firms. Furthermore, as regulatory pressure increases, the ability of AI to audit complex Scope 3 data will become a non-negotiable requirement for staying in business. The companies that thrive will be those that view AI not as an infinite resource, but as a precious tool that must be managed with the same level of fiscal and environmental discipline as any other part of their infrastructure.
