The real challenge for AI will not be money or chips, but the lack of energy in the world

Artificial intelligence continues to grow, and many believe that this progress will be conditioned by funding and chip development. However, behind these technological barriers lies an even greater challenge: energy capacity and the pace of its development.

The value of generative AI is already evident in the reduction of operating costs for sectors such as customer service, information technology, and marketing. New companies have achieved multi-million dollar valuations with minimal teams thanks to the efficiency provided by this technology.

Nevertheless, these achievements fail to distract from an undeniable material reality. The true bottleneck for AI is neither a lack of investment nor a shortage of users or talent. The central problem lies in the insufficient energy supply and physical infrastructure to sustain the pace of advancement.

The Disparate Pace of Energy Growth vs. AI Growth

Computing power is increasing at a faster rate than the efficiency improvements in chips. Conversely, the world's power grids have shown much slower growth in capacity over the past few decades compared to the increase in technological demand.

According to recent estimates published by Expansión, data centers in the United States could require approximately 106 gigawatts (GW) of power by 2035. Globally, the projected demand for data centers reaches 219 GW by 2030.

Without a corresponding structural expansion, the construction of new data centers simply cannot keep pace with the current and projected demand for artificial intelligence.

The increase in energy consumption driven by AI is enormous. Advanced models like GPT-4 can consume up to 463,269 megawatt-hours per year, an amount of energy greater than that used annually by more than 35,000 American households.

The demand for electricity to power large language models, cloud infrastructure, and intelligent services is growing exponentially in data centers.

Projections from the consulting firm Rystad Energy show that global electricity use by data centers will double by 2030 and could reach 1,800 terawatt-hours by 2040, enough to power 150 million U.S. homes for an entire year.

This situation highlights that the pace of AI development no longer depends solely on obtaining the most advanced chips. Today, the price and availability of energy are the determining factors for the consolidation and advancement of the technology.

Why money won't be the ultimate barrier to AI progress

The AI bubble narrative doesn't stem so much from the traditional logic of financial speculation as from the physical and material limits of the planet. In the past, during the dot-com bubble at the beginning of the century, excess infrastructure exceeded actual demand.

Currently, the phenomenon is the reverse: technological demand, driven by artificial intelligence, far exceeds global energy and infrastructure capacity.

The world is heading towards a scenario where the development of artificial intelligence will be hampered by material constraints. The scarcity of energy, chips, and data centers is creating an insurmountable barrier that even the largest budgets will be unable to overcome if the necessary resources are not available.

Faced with the limitations of the power grid, companies like Meta, Google, Microsoft, and Amazon have sought alternatives by contracting their own power plants or signing agreements to secure clean energy.

In the first half of 2025, these companies contracted 9.6 GW of renewable energy, enough to power 7.2 million homes. The outlook suggests that, despite political disagreements and regulatory obstacles, the market will continue to drive the demand for clean energy due to the imperative needs of the sector.

The energy needed to support artificial intelligence is increasing so rapidly that it has already delayed the retirement of several coal-fired power plants in the U.S., and more delays are expected.