INCONCRETO NEWS

AI’s Energy Debate in 2026: Between Efficiency Claims and Environmental Reality

News

04 May, 2026

Carla ABIVEN

AI’s Energy Debate in 2026: Between Efficiency Claims and Environmental Reality

The environmental cost of artificial intelligence has quickly evolved from a niche technical concern into a mainstream debate shaping policy, investment, and public perception. At the center of this discussion is Sam Altman, CEO of OpenAI, who recently pushed back against criticisms of AI’s growing resource consumption during an onstage interview at the India AI Impact Summit.

Altman’s remarks highlight a broader tension in the AI industry: how to balance rapid innovation with mounting scrutiny over energy and water usage. His argument—that concerns about AI’s resource demands are often exaggerated or misunderstood—has sparked both agreement and backlash, reflecting the complexity of the issue in 2026.

The Water Controversy: Myth vs Reality

One of the more viral claims about AI systems like ChatGPT is that they consume gallons of water per query. In 2023, Pengfei Li of the Rochester Institute of Technology calculated that“every 20 to 50 questions to a large language model like ChatGPT require the equivalent of a 500 ml bottle of water, mostly for cooling the servers that generate responses”. In 2026, Altman dismissed this as “totally fake”, pointing to a significant shift in data center cooling technologies. He said earlier practices have changed and that such water usage is no longer part of current data centre operations, pushing back against growing scrutiny over AI’s environmental impact. Traditionally, many facilities relied on evaporative cooling systems, which can indeed consume large amounts of water. However, major AI companies have increasingly transitioned toward closed-loop or air-based cooling systems.

This shift is real and measurable. Hyperscale data center operators—companies like Microsoft, Google, and Amazon—have invested heavily in reducing water intensity per computation. OpenAI’s own infrastructure, including its large-scale facility in Texas, reportedly uses a recirculating cooling system that minimizes waste.

Yet critics argue that Altman’s framing downplays the broader picture. Even if individual queries require negligible water, the aggregate impact of billions of daily interactions is substantial. Reports from firms like Xylem and Global Water Intelligence suggest that AI-related water consumption could increase by over 100% by 2050, driven not just by cooling but also by semiconductor manufacturing and electricity generation.

In other words, the controversy isn’t about whether one ChatGPT prompt uses a glass of water—it’s about the systemic footprint of an industry scaling at unprecedented speed.

Energy Use: A More Honest Concern

Unlike his dismissal of water concerns, Altman acknowledged that AI’s electricity consumption is a legitimate issue. His estimate—around 0.34 watt-hours per ChatGPT query—frames AI as relatively efficient on a per-task basis. For comparison, that’s roughly the energy used by an oven in one second.

However, this figure is both useful and misleading. It provides a snapshot of inference (the act of answering a query), but excludes the enormous energy required to train large models like GPT-5 and beyond. Training runs can consume megawatt-hours to gigawatt-hours of electricity, depending on scale.

Moreover, energy use varies dramatically depending on the task. A simple text response is far less resource-intensive than generating high-resolution images or running multimodal reasoning across video and audio inputs—capabilities that are becoming standard in 2026.

Industry analysts emphasize that AI’s total electricity demand—not per-query efficiency—is what matters. Global data center electricity usage is projected to double or even triple by 2030, with AI as a primary driver.

The “Human Comparison” Fallacy

Perhaps the most controversial part of Altman’s remarks was his comparison between AI systems and human intelligence. He argued that training a human requires decades of energy consumption—food, education, and societal infrastructure—suggesting that AI may already be competitive in terms of energy efficiency.

This analogy is provocative but flawed.

Humans are biological systems embedded in ecosystems, not engineered tools optimized for task efficiency. Comparing the lifetime energy cost of a human to the operational cost of an AI system conflates fundamentally different categories. Critics argue that a more appropriate comparison would be between AI systems and existing digital tools, not human existence itself.

Still, Altman’s point resonates in one respect: once trained, AI systems can scale knowledge and productivity at near-zero marginal cost per user. This is where AI’s efficiency advantage becomes compelling—and economically transformative.

The Industry’s Defensive Posture

Altman’s comments reflect a broader shift in tone across the AI industry. In 2023 and 2024, companies largely emphasized AI’s transformative potential with limited discussion of environmental costs. By 2025 and 2026, that narrative has changed.

Amazon, Apple, Facebook, Google, and Microsoft have committed billions of dollars to renewable projects to cut their carbon footprint and support sustainable infrastructure. They buy huge amounts of clean electricity through renewable power purchase agreements (PPAs), invest directly in new projects, and push their suppliers to adopt greener practices. This shift is reshaping how the tech industry interacts with the planet’s resources.

Tech firms are now actively defending their sustainability credentials, publishing transparency reports, and investing in renewable energy. OpenAI, for example, has aligned with partners pushing for increased use of nuclear, solar, and wind power—an approach Altman explicitly endorsed.

This pivot is partly reactive. Governments in the European Union and parts of Asia are considering regulations requiring disclosure of AI systems’ environmental impact. Meanwhile, institutional investors are pressuring companies to meet ESG (Environmental, Social, Governance) targets.

Regulating Scale: Policy Meets Infrastructure

As of April 2026, the AI resource debate is entering a truly moment. Policymakers are no longer asking whether AI should be regulated, but how. Environmental impact is becoming a key part of that conversation.

Several potential solutions are gaining traction:

Energy diversification: Expanding nuclear and renewables to meet AI demand
Efficiency standards: Benchmarking models based on energy per task
Geographic optimization: Locating data centers in regions with sustainable resources
Hardware innovation: Developing chips optimized for lower power consumption

At the same time, researchers are exploring fundamentally new approaches to AI that could reduce reliance on brute-force computation, such as neuromorphic computing and more efficient training paradigms.

This shift toward regulation is also being accelerated by the sheer scale of projected resource demand, which is forcing policymakers to think systemically rather than incrementally. Recent research by Xylem and Global Water Intelligence shows that AI-driven water demand could rise by roughly 129% by 2050, adding around 30 trillion liters annually across the value chain—from chip manufacturing to electricity generation and data center operations. Crucially, more than half of that increase is expected to come indirectly from power generation, underscoring how tightly energy and water challenges are intertwined.

This is why energy diversification is no longer just a climate issue but a resource strategy. Renewable sources like wind and solar dramatically reduce water usage compared to thermoelectric power, which relies heavily on water for cooling. At the same time, the push for efficiency standards is gaining momentum, with researchers calling for transparent, standardized metrics to measure AI energy and water intensity per task—something still largely absent from corporate disclosures.

Geography is also emerging as a decisive factor. Data centers are increasingly being located in regions with abundant renewable energy or cooler climates, where both electricity and cooling demands can be optimized. However, tensions remain, as many existing hubs are in water-stressed areas, raising concerns about local resource competition.

Finally, the next frontier lies in technological breakthroughs. Beyond incremental hardware improvements, approaches like neuromorphic computing and more efficient model architectures aim to fundamentally reduce computational intensity. Without such advances, even the most aggressive efficiency gains may struggle to keep pace with AI’s exponential growth trajectory.

The debate sparked by Altman’s comments ultimately reflects a deeper question: what kind of technological future do we want?

AI has the potential to accelerate scientific discovery, improve healthcare, and drive economic growth. But these benefits come with real environmental costs that cannot be ignored.

Altman is right to challenge exaggerated claims and highlight efficiency gains. However, critics are equally justified in pointing out that aggregate impact—not per-query metrics—should guide the conversation.

In 2026, the AI industry stands at a crossroads. Its leaders must do more than defend their technologies—they must demonstrate that innovation and sustainability can scale together.

The outcome of this debate will shape not just the future of AI, but the broader relationship between technology and the planet.

INCONCRETO’s Strategic Vision

At INCONCRETO, we interpret the current debate around AI’s resource consumption as a shift in scale, not in principle. The discussion sparked by Sam Altman reflects a broader industry move: from questioning impact to managing perception.

The emphasis on per-query efficiency—highlighted by OpenAI—signals progress, but it does not address the core dynamic: exponential usage growth. As AI integrates across all sectors, total demand for compute, and therefore energy, continues to accelerate beyond efficiency gains.

This creates a structural reality: AI is becoming resource-bound.

According to the International Energy Agency, data center electricity demand is set to rise sharply this decade, driven largely by AI workloads. In this context, energy is no longer a constraint at the margins—it is becoming a central factor in scaling.

The implication is clear. The future of AI will not be determined solely by model performance, but by the ability to align compute with energy, infrastructure, and geography.

Efficiency matters. But at scale, total consumption defines impact.

AI is no longer just a technological expansion.

It is an infrastructural one.