Traffic in the cloud: the induced demand of AI

By Ray Adrian Macalalag

If we go back to ten or fifteen years ago, was there something that you wish was more efficient? Probably it was about queuing in a passenger terminal three times: buying a ticket, checking in, and paying for the terminal fee. This still happens today in my occasional commutes between Iloilo and Bacolod, where passengers must navigate fragmented systems across different ports. Systems and interoperability vary from one institution to another. We also have minor inconveniences that became so difficult, particularly thinking. The term “thought process” has become more common in conversations. It arises even in difficult ones because people have become more skeptical if we are still able to think clearly or if we have totally hooked ourselves to artificial intelligence.

The adoption of artificial intelligence is regarded as one of the fastest technological shifts in history as it met its inflection point during the pandemic. It solved problems in a fraction of the time; it allowed people to illustrate what is on their mind, and it enabled those who were able to imagine but were not readily able to execute. I mean this in a profoundly good way. The potential for systemic development is staggering. The World Economic Forum estimates that AI can contribute up to 14 percent of the global gross domestic product in 2030. This is about USD15.7 trillion. Generative systems specifically are expected to boost labor productivity across numerous sectors. In the realm of development, embedded technology is undeniably a game changer. It holds the potential to optimize complex supply chains, streamline logistics, and eventually solve the interoperability nightmares that plague our systems.

However, as we become more locked in with this new system, problems that we once turned a blind eye to have finally caught up with us. The digital cloud is heavily tethered to the ground, and its physical footprint is expanding at an alarming rate. We routinely generate photos, draft content, and create almost anything our heads think of, often not considering whether it is truly necessary or worth the resources consumed behind the scenes. The environmental impact of training these large language models is significant, stemming primarily from the substantial energy consumption required to power high-performance computing.

As a transport person who has observed infrastructure development from local projects to international frameworks, I see striking parallels. In transport planning, we often deal with induced demand, a phenomenon where building more roads simply attracts more vehicles, eventually leading to the same congestion we tried to solve. The current trajectory of AI infrastructure seems to follow a similar paradox. As computing power becomes more accessible and advanced, our demand for generation only increases, compounding the strain on our physical resources.

The energy requirements to sustain this daily usage are staggering. The International Energy Agency previously estimated that global data centers consumed over 400 terawatt-hours (TWh) of electricity in 2024, representing about 1.5 percent of total global electricity demand. Future projections show a doubling in 2030 to 945 TWh. To illustrate better, this small growth is four times faster than the growth of total electricity consumption. Beyond the power grid, there is a silent thirst associated with our digital innovation. These massive data centers carry a significant water footprint for cooling servers and energy production, about 560 billion liters per year that could reach 1.2 trillion by 2030.

The environmental impacts of these infrastructures are not evenly distributed. While the economic and productivity benefits are enjoyed globally, the severe costs are heavily concentrated in the specific communities hosting these data centers. Adding to this local burden, the rapid hardware turnover required to sustain these networks creates massive physical waste.

Despite these sobering realities, there is an optimistic path forward with precise and committed action. We do not need to abandon the convenience and remarkable innovation that these digital systems provide. Instead, we must apply rigorous, systemic thinking to how we deploy them. We must enthusiastically advocate for energy efficient algorithms, prioritize renewable energy for data centers, and cultivate a culture of mindful use.

Do we really need to use AI to generate things that merely entertain or become our captions on social media? Accelerating the use of artificial intelligence is essential for our continued development, but it must be thoughtfully balanced with the physical limits of our environment and our individual responsible use. If we do so, we can ensure that the systems designed to make our lives easier do not ultimately cost us the very world we are trying to improve.