AI energy demand through 2030: Navigating the wicked problem of infrastructure and sustainability

Insights

• AI’s growth is constrained by limits in energy supply, with demand for electricity increasing quickly due to massive AI workloads.
• The next 1,000 days will be critical for organizations to continue to be competitive. AI-enabled energy management can unlock capacity sooner than building new generation and grid upgrades.
• Sustainable AI is an engineering and operating-model discipline, not just procurement. Sustainability leaders should work to right-size AI models; optimize inference and workflows; and measure energy and carbon per workload to keep business value from scaling linearly with energy use.

AI has moved from an incremental efficiency tool to a general-purpose engine of enterprise modernization and innovation.

But in all of this, enterprises are not entering a simple AI cycle that continues to repeat itself with energy requirements going up and down in sync. This era is defined by a broader modernization cycle in which electricity becomes the primary backbone for new ways of living and working. This includes remote interaction, automated decisioning, advanced manufacturing, robots, and autonomous mobility.

With this framing, energy demand is the signature of a more digital, more automated, more electrified economy.

This acceleration is compute-driven, as generative AI and the agents that work with them flip the economics of digital work. Simple prompts can yield complex outputs at high speed, creating vast opportunity for enterprises. But they do this work at a significantly higher energy intensity than prior generations of software.

Scientists have estimated that the power requirements of data centers in North America increased from 2,688 megawatts at the end of 2022 to 5,341 megawatts at the end of 2023, partly driven by the demands of generative AI. Globally, the electricity consumption of data centers rose to 460 terawatt-hours in 2022. This would have made data centers the eleventh largest electricity consumer in the world, between the nations of Saudi Arabia (371 terawatt-hours) and France (463 terawatt-hours), according to the Organization for Economic Co-operation and Development.

At the same time, enterprise adoption is broadening. Infosys Knowledge Institute research shows that 42% of enterprise-scale businesses have already deployed AI and another 40% are exploring or experimenting, meaning demand pressure is not hypothetical, but diffusing across industries.

In Infosys Knowledge Institute’s Tech Navigator: A path to the agentic-first enterprise report, we positioned this moment as the start of an era where enterprises build digital and cloud foundations to augment and amplify human potential, unlock efficiency at scale, and reinvision customer journeys and internal processes.

At Schneider Electric, the AI narrative is one of increasingly broad AI adoption, and where the transition to AI is accelerating electrification, energy load, and infrastructure demand. Working out how to navigate an enterprise’s AI journey while contending with these factors is the heart of Schneider Electric’s Research Institute, particularly its "Time to Power’ research.” Led by Vincent Petit, this research examines the generational electrification and infrastructure modernization cycle underway in the US, and why the next 1,000 days represent a decisive window for enterprises to act to generate AI business value for years to come.

Rising challenges: AI workload growth and grid capacity constraints

AI agents are materially more energy intensive than traditional digital workloads. The training and operation of large models that they run on raises new questions about carbon accounting, clean electricity procurement, water use (for cooling), and indirect second-order effects of technology shifts.

There is also an execution gap within enterprises. Many organizations still orient around fragmented experiments rather than integrated products and platforms, with a lack of data, MLOps and LLMOps, security, governance, and change management.

Without those foundations, leaders face a double bind. They cannot scale AI safely and economically, but they also cannot pause adoption because competitors will deploy AI to move faster, serve customers better, and compress cost curves.

Why AI energy demand creates complex infrastructure and sustainability challenges

This compute-demand, AI-first imperative, is a wicked problem for enterprises. The real difficulty is that AI’s growth rate is now coming up against the slowest-moving part of the enterprise - physical infrastructure. The risk is not merely higher energy costs. Rather, it is the hard constraints of data center growth, reshoring and advanced manufacturing, and the broader electrification of mobility and buildings. In fact, six wicked problems intertwine in a frightening risk landscape for big enterprise:

1. Conflicting objectives: Upgrading energy and grid capacity pushes toward long-term resilience, while competing with AI-first firms pushes toward immediate speed, scale, and cost advantage. Maximizing both at once is very difficult.  
2. Different timescales and constraints: AI adoption cycles move in months, while energy infrastructure moves in years. In this way, decisions must be made under structural delay.
3. Fragmented authority: Enterprises don’t control the grid, permitting, utility investment, or broader energy mix, and yet their competitiveness depends on those outcomes.
4. Interdependent causes: More AI drives more compute demand, which drives more power, cooling, and water demand, which hits grid limits, which then constrains data center expansion and AI rollout.
5. Solutions are either better or worse, not completely true or false: Choices like shifting workloads, demand-response, efficiency engineering, or new procurement can improve the situation, but won’t definitively solve it.
6. Every intervention has side effects: Migrating workloads may reduce peak load but can increase latency, cost, compliance risk, or vendor concentration.

Even though net zero is not a headline item in 2026, it is imperative that enterprises fulfil their decarbonization commitments.

The next 1,000 days are critical because three years is a meaningful window of time to deploy demand-side solutions, but an insufficient period to rebuild large parts of the grid if action is delayed.

Practical learnings: Schneider Electric’s approach to sustainable AI energy management

The resolution to this wicked problem is to treat AI, energy, and infrastructure as a single overarching, strategic system. With this comprehensive perspective, organizations can work smarter to unlock capacity while new supply and grid upgrades come online.

The most decisive lever is not building more generation and wires, but improving how electricity is used. Because load profiles in many markets are peak-driven and not flat, the grid is materially underutilized. Deploying AI-enabled solutions on the demand side, especially across buildings and industrial facilities, will flatten peaks and raise utilization, effectively creating virtual capacity faster than physical buildouts.

A central insight from our research at the Schneider Electric Sustainability Research Institute is that peak load could be reduced by up to 60% with demand-side deployment at scale, buying roughly a decade of time to modernize the grid.

In parallel, there are a number of other actions organizations can take now. This includes a portfolio of compute efficiency techniques that decouple business value creation from energy growth; selecting fit-for-purpose model sizes; using smaller or open models where appropriate; and applying quantization, pruning, and knowledge distillation to reduce storage and compute needs.

These technical choices matter at scale because they change the unit economics of every inference call and every automated workflow.

Operationally, this also requires reimagining experiences and processes to focus investment on high-leverage journeys such as employee productivity, customer service, sales enablement, software delivery, and knowledge work.

Strategic imperatives: Building sustainable AI infrastructure for future growth

There are six strategic imperatives to address the six wicked AI energy problems, enabling sustainability executives to become valuable contributors to thought leadership and strategic know-how as enterprises embrace agentic capabilities:

1. Elevate infrastructure as a board-level AI narrative: Treat power availability, grid constraints, and cooling, water as first-class constraints in AI strategy. Executives should build scenario plans for capacity, price volatility, and permitting timelines.
2. Prioritize virtual capacity in the next 1,000 days: Deploy AI-enabled energy management, peak shaving, and flexibility programs across buildings and industrial sites to reduce peaks and raise utilization, buying time for physical grid modernization. Focus first on the highest-peak, highest-cost sites and scale with standardized playbooks.
3. Engineer for sustainable AI economics: Institutionalize model selection and optimization; right-sizing; open- versus closed-models; retrieval-augmented approaches; quantization, pruning, and distillation; and track energy- and carbon-per-workload. Also optimize for cost-per-outcome, not for cost-per-token.
4. Shift the workforce up the stack: Make augmentation explicit. Deploy copilots and in-the-flow tools, redesign roles, and create targeted upskilling so productivity gains translate into higher-value work such as design thinking, domain expertise, customer empathy, and creative problem solving.
5. Measure what matters: Establish a small set of key performance indicators reviewed by executive leadership. These include value realized, model quality and safety, time-to-deploy, energy and carbon intensity, and resilience (uptime and recovery). These KPIs should steer investment and prevent low-value experimentation.
6. Partner across the ecosystem: Infrastructure constraints cannot be solved by any single enterprise. Proactively collaborate with utilities, hyperscalers, equipment providers, and policymakers on demand response, clean energy procurement, and grid modernization. Do not allow the dream of agentic AI at scale to become a bottleneck for growth.

AI has become a modernization force, with many business growth characteristics. However, its limiting factor is increasingly physical: electricity, grid capacity, and the infrastructure required to deploy compute at scale.

The winning response is not to slow AI adoption, but to accelerate smart capacity - demand-side flexibility and compute efficiency - while building AI-first operating capabilities that enable organizations to scale responsibly and sustainably.