The conversation around artificial intelligence (AI) so often revolves around scale, speed and sheer computational power. Yet, as the world confronts urgent climate, energy and equity challenges, a different question is being asked among forward-looking business and technology leaders: what if “bigger” is the wrong goal for AI? What if the future demands not just more powerful AI, but more purposeful, frugal and accessible AI? This theme sits at the heart of the latest episode of the Cambridge Executive Business Insights: Rethinking AI podcast hosted by Jaideep Prabhu, Professor of Marketing at Cambridge Judge Business School. Joined by Philippe Rambach, Chief AI Officer at Schneider Electric, the episode explores how AI can be a catalyst for a smarter, leaner and more sustainable energy future, without falling into the trap of technological excess.
Frugality as innovation: setting the stage
For decades, “innovation” in both business and technology was synonymous with doing more: more capacity, bigger models, greater consumption. But the limits of “more” are now painfully visible. From ballooning data centre demand to the staggeringly energy-intensive training of the latest large language models (LLMs), there’s a growing recognition that the planet cannot afford unbounded technology escalation.
It’s why concepts like frugal innovation (rooted in doing better with fewer resources) and frugal AI are emerging from the theoretical sidelines to the mainstream. As Jaideep Prabhu notes, frugal innovation originally thrived in resource-constrained environments of emerging markets, but now influences Western enterprises searching for sustainable advantage. In the case of Schneider Electric, frugal AI is not simply a slogan but a guiding design principle, one which directly shapes how millions of people consume, optimise and conserve energy.
Deploying AI at scale with purpose, not excess
When Philippe Rambach assumed the newly created role of Chief AI Officer at Schneider Electric, it was a recognition not of AI’s technological mystique, but of its potential as a business transformer. Schneider, with its €40 billion revenues and 140,000 employees worldwide, occupies a pivotal junction of energy management and industrial automation. Its core mission? Helping its customers become radically more energy efficient, lowering costs, reducing carbon, and accelerating the transition to renewables.
The approach to AI at Schneider is markedly pragmatic. Philippe Rambach characterises his primary focus as “deploying AI at scale”, both for internal operational efficiency and, crucially, to supercharge value for customers. As he explains, “We always start from the business value” and then build tailored, agile teams combining AI specialists and domain experts, with the explicit goal of deployment beyond pilots, which impacts thousands of customers and tens of thousands of employees.
But this is not technology for its own sake. As Philippe makes clear, “We want to deploy an AI that impacts the way we work, that impacts our customer, that brings true value. Not tech for tech, AI for AI”.
What is frugal AI?
At the heart of Schneider’s AI philosophy is a simple proposition: always start from the need and then select the most efficient solution, even if it isn’t technically “AI” at all. The label is less important than the outcome.
For instance, as Rambach recounts, there are moments when partners ask for advanced AI but, upon careful analysis, a basic linear regression suffices: “A simple linear regression will do it. So, we won’t use AI? No, you will not use AI because you know what? A simple linear regression will do it”.
That is frugal AI in action: choosing the most energy-efficient, cost-effective, and suitable method for the challenge, whilst resisting the siren song of ever-larger, resource-hungry models unless truly justified.
This does not mean eschewing state-of-the-art methods when needed – there are problems only generative or foundational models can tackle, but it means not defaulting to “bigger is better” thinking.
The paradox of AI and energy
The potential of AI to accelerate the energy transition is often celebrated: smart grids, advanced optimisation and real-time demand management can cut emissions and unlock unprecedented efficiency. Yet, as Jaideep Prabhu points out, there’s a catch: the exponential rise of AI workloads and data centre energy demand can threaten to overwhelm national grids and undercut climate progress.
Philippe Rambach offers a nuanced perspective. In relation to global energy consumption, 120,000+ terawatt-hours annually, data centre usage (around 500 terawatt-hours) is still relatively modest, and of that, AI constitutes perhaps 10%, a number that, even if tripled in the coming years, remains a small portion of total demand.
But the real issue is what AI is used for. “The AI used to generate jokes for your neighbours is just consuming energy and does not help the planet. But the AI used to help the planet does work… If we are able to reduce by a few percent the 120,000 terawatt by using a few terawatt of AI, the math works”.
In other words, AI’s social and environmental purpose must outweigh its own carbon cost. Used intelligently, AI can be a lever for system-wide savings that far exceed its energy appetite. But trivial, wasteful or misapplied AI quickly becomes counterproductive.
Smart grids, smarter demand: where AI’s impact is greatest
For all the hyperbole about data centres and foundational models, some of the most profound opportunities for AI in the energy transition are deeply practical, local and immediate.
Decarbonising the grid, by integrating more solar, wind and distributed renewables, demands AI for real-time balancing and optimisation. Yet, as Philippe notes, the most undervalued area is demand-side optimisation: “the energy that impacts less the planet is the energy that we do not use or do not need”.
Applying AI to optimise building heating and cooling is a case in point. With tiny, embedded models in smart thermostats, Schneider can “save 10 to 15% of energy” in real time, at minimal computational cost. At scale, this simple intervention multiplies across millions of homes, factories and offices.
Similarly, AI applied to demand forecasting, supply prediction and grid operations can “avoid peak demand,” thereby reducing the need to start up the dirtiest forms of generation at the worst times. By helping shift or shave energy demand away from carbon-intensive peaks, AI multiplies the impact of every renewable watt deployed.
Philippe argues that such approaches are “heavily underestimated”, perhaps because, unlike grandiose foundation models, these interventions are humble, local and invisible. Yet they hold the key to the 10-20% demand reductions necessary for real climate progress.
Measuring AI’s environmental footprint: the next frontier
For the architects of AI strategy, measuring progress isn’t just about predictive accuracy or return on investment: it must include return on carbon. Companies, including Schneider Electric, are still developing the tools to continuously monitor and optimise the carbon footprint of AI models in operation. While it is currently feasible to do case-by-case checks (for instance, verifying that a model which emits 1kg of carbon saves 500kg elsewhere), there is a universal need for greater transparency from hyperscalers and cloud providers. Continuous monitoring must become standard, so organisations can make informed trade-offs.
Without robust measurement, technology leaders risk falling back on guesswork or rationalising unsustainable choices. Industry-wide progress on this front is both a technical and strategic necessity.
Frugal AI by design: principles for action
As organisations increasingly build custom AI solutions, using everything from proprietary LLMs to open-source agentic approaches, how can they ensure frugality is engineered in from the outset?
Rambach distils the core principle with clarity: “Never start from, oh, we have all these great technologies, what can we use? Start from, what do we want for us, for our employees, for our customers?”. From there, select the simplest, fastest, most economical way to achieve that value, whether that means a large model, a small model, or no AI at all.
This careful, need-first approach accomplishes several benefits:
- Lower upfront and operational carbon/energy costs
- Easier explainability, auditability and trust (smaller, simpler models are more transparent)
- Faster, less risky deployment
- Stronger fit between technology and actual business requirements
It also resists the vendor-driven temptation to adopt “the latest model” simply for its buzz factor. Incremental, context-driven improvement is a mantra throughout Schneider’s AI journey: “You can use AI to progressively improve what you do, use less energy. You don’t need to redo everything from day one”.
The power of systemic adoption and ecosystem partnerships
If the energy transition is to succeed, AI-enabled efficiency and optimisation must scale from isolated use-cases to systemic change, spanning the built environment, industry, grids and individual consumers.
Philippe Rambach offers a compelling example from the Nordics: a region-wide estate of commercial buildings and shopping centres saw carbon emissions and costs fall by 30% by using AI to forecast and optimise renewable energy use, consumption and battery storage every 15 minutes. Crucially, this is not simply a technical achievement, it requires partnership across operators, utilities, and technology providers.
The same logic applies more broadly: no single company, even one with 140,000 employees, can unilaterally drive change. Technology partners, electricians, system integrators, panels builders, policy makers and consumers all form part of a dynamic ecosystem, where training, trust and mutually reinforcing incentives are vital. As Philippe Rambach puts it, “alone we can do nothing”.
As the energy system decarbonises with more renewables, varied sources and geographic expansion, those partnerships must also extend between continents and countries, including highly populous developing nations where energy demand and carbon emissions are still rising. The imperative, as Philippe notes, is to “give energy to everybody at a cost acceptable for the planet”.
AI and the complicated case of data centres
Concerns about AI’s environmental impact are most acute in the context of hyperscale data centres, where worries about power supply, grid stability and cooling are mounting worldwide. Yet, as Philippe observes, there is cause for measured optimism. Data centres run on electricity, which is a form of energy that can be decarbonised.
Continuous innovation – from more efficient power electronics and increased voltage (to reduce losses), through to evolving cooling techniques and infrastructure upgrades – means the per unit energy efficiency of data centres keeps improving. AI itself can help further, optimising everything from workload balancing to predictive maintenance and dynamic cooling management.
In future, the vast battery banks that safeguard data centre reliability could even play a balancing role, supporting grids with intermittent renewables by storing and dispatching excess power when needed. These innovations are embryonic but point the way to systemic impact.
Nevertheless, the proliferation of ever-larger LLMs and more powerful foundational models should be tempered with realism. As Rambach says, “There are use cases who need very large models… but it is not needed everywhere”. The key is to “choose the right technology for the right need”, and avoid treating every problem as a job for the biggest new model.
Electricity as cornerstone: decarbonising at the source
AI’s full potential in reshaping energy systems can only be realised if the foundational energy source, that is, electricity itself, is decarbonised. Whether from renewables (solar, wind), new nuclear (including small modular reactors), or as-yet-unproven energy vectors, the trend is clear: only electricity offers a credible path to zero-carbon abundance.
As Philippe Rambach frames it, “there is no path to burn oil and not get carbon. Nobody has any direction to make that happen. That’s why I’m a big fan of electricity”. This perspective isn’t naïve optimism: it reflects a recognition that, while no single solution is sufficient, the path for electricity is at least visible and actionable.
The human factor: learning, adopting and adapting
All technology transformation is ultimately human, and requires learning, upskilling, and cultural change. Philippe stresses the importance of continuous training: “Learn, learn, learn and train. You need to understand a bit what it is. It’s a technology that you cannot just watch from behind”.
Adoption at scale demands that organisations and their people see AI as a tool of incremental benefit, rather than a threat or a black box. This philosophy of seeking benefit, moving stepwise, and empowering teams, resonates with the best traditions of frugal innovation.
And for policymakers and business leaders looking to the next three to five years, success will be measured by the ubiquity of these solutions: “if everybody was using the solution, which is a big if… from a technological point of view, we have the solution to the transition. The question is deployment. How do we go faster?”.
Frugal AI: towards a smarter, more equitable future
The lesson from Schneider Electric’s experience, and the insights of Philippe Rambach, is not that technological ambition is misplaced, but that it must be matched with purpose, context and humility. AI is a lever, not a panacea. It can only drive the energy transition if its energy and carbon costs are carefully weighed, its interventions targeted, and its benefits democratised through the ecosystem.
At a more philosophical level, this approach is grounded in a belief that value creation is maximised not by maximising inputs (energy, computation, data), but by maximising outputs, that is, the outcomes that matter for people and the planet. The future belongs to those who design not the biggest systems, but the most effective, frugal and meaningful ones.
To delve further into the principles, strategies and practical examples of frugal AI in the energy sector, listen to the full episode of the Cambridge Executive Business Insights: Rethinking AI podcast, available wherever you get your podcasts.
