Does Cheaper AI Mean More AI Usage?

The article explores whether AI follows the Jevons paradox, arguing that while efficiency gains can drive adoption, AI’s growth depends more on usability, demand, and industry constraints.

In 1865, William Stanley Jevons warned that Britain’s economic dominance was doomed because its coal reserves would inevitably be exhausted. His reasoning was counterintuitive: making coal use more efficient would not conserve it, but rather increase its consumption. This idea, now known as the Jevons paradox, has been used to explain why efficiency gains in energy use often lead to higher overall demand. Recently, Microsoft CEO Satya Nadella invoked this principle to suggest that AI will become even more widespread as it becomes cheaper and more efficient. His argument is that AI will follow the same trajectory as coal or lighting: lower costs will drive higher adoption, leading to an explosion in usage.

But does AI really fit the Jevons paradox? Unlike coal, AI is not an essential physical resource, nor does it face the same straightforward supply constraints. While energy and compute power are necessary for AI to function, its spread depends on far more than just cost and efficiency. Adoption has been slow, and businesses still struggle to integrate AI into their workflows. The Jevons paradox may not be the best lens through which to understand AI’s future. Instead, AI’s trajectory is likely shaped by a more complex mix of technological bottlenecks, market demand, and network effects.

The Jevons Paradox and Its Limits

The Jevons paradox originally applied to energy sources like coal. The logic was simple: when a resource becomes cheaper to use, people use more of it. This principle has been observed in other domains, such as lighting and transportation. Economist William Nordhaus found that lighting efficiency has improved dramatically over time—from ancient oil lamps producing 0.06 lumens per watt to modern LEDs producing over 100 lumens per watt. Yet instead of using the same amount of light at lower cost, society has vastly expanded its use of artificial lighting. Similarly, improvements in fuel efficiency have often led to increased driving rather than reduced fuel consumption.

However, the Jevons paradox does not always apply. In many cases, efficiency gains do lead to net savings. Home insulation, for example, generally reduces energy use, despite some rebound effects such as people keeping their homes warmer. The key question is whether increased efficiency leads to proportional increases in demand. If the demand for a good is already saturated or constrained by other factors, then improving efficiency will not necessarily lead to runaway consumption.

Does AI Fit the Jevons Paradox?

For Jevons’ logic to hold in AI, one would expect three things to be true:

1. AI’s adoption is primarily constrained by cost.

2. Demand for AI is elastic, meaning that as costs fall, businesses and individuals will use significantly more of it.

3. AI is a general-purpose technology that will integrate deeply into all sectors, much like electricity or coal.

On the first point, AI’s biggest constraint is not just cost but usability. While AI models are expensive to train and run, many businesses are not adopting them simply because they are difficult to integrate into existing workflows. According to the U.S. Census Bureau, only 5% of American firms currently use AI, with another 7% planning to adopt it in the future. This is not the pattern one would expect if AI were like coal or lighting, where falling costs would lead to widespread, inevitable adoption. Many businesses do not see an obvious use case for AI, suggesting that cost efficiency alone will not drive exponential growth.

On the second point, demand elasticity, the situation is unclear. Some applications of AI, such as customer service chatbots or automated data processing, may see increased use as costs decline. But for other industries, AI adoption is not just a matter of affordability. Healthcare, for instance, faces significant regulatory hurdles before AI can be widely implemented in diagnostics or treatment. Legal and financial sectors also have constraints due to liability risks and concerns about accuracy. If AI adoption is not driven purely by price but also by trust, regulation, and complexity, then efficiency gains will not necessarily result in exponential growth.

On the third point, whether AI is a fundamental technology like electricity or just a powerful tool for certain industries remains an open question. Electricity became indispensable because it powered everything from manufacturing to households. Coal was the backbone of the industrial economy. AI, by contrast, is still largely a supplementary tool—useful in some contexts but not yet essential. If AI is to follow the Jevons paradox, it would need to become a core necessity rather than an optional advantage.

Alternative Explanations for AI Growth

If the Jevons paradox does not fully explain AI’s trajectory, what does? One alternative framework is that AI’s growth will depend on network effects rather than pure efficiency gains. Technologies like the internet did not take off simply because they became cheaper, but because their usefulness increased as more people adopted them. AI may follow a similar path, where its value increases as more industries find ways to integrate it.

Another possibility is that AI will grow in the same way as high-end enterprise software: steadily but not explosively. Large-scale adoption will depend on solving real business problems, not just on making AI models cheaper to run. In this view, AI is more like enterprise cloud computing—an important but incremental shift rather than a sudden explosion in demand.

Conclusion

Satya Nadella’s argument that AI follows the Jevons paradox is compelling on the surface, but it assumes that AI adoption is primarily a function of efficiency and cost. In reality, AI adoption is constrained by usability, trust, and industry-specific challenges. Unlike coal or lighting, AI is not a universally necessary resource, nor is its demand purely driven by price. While AI usage will undoubtedly grow, its expansion is likely to be shaped more by network effects, industry integration, and regulatory barriers than by simple cost reductions. Nadella may be right that AI will continue to scale, but if Microsoft’s AI investments succeed, it will be for reasons beyond the Jevons paradox.