The new Artificial Intelligence Gold Rush risks colliding with a limit as invisible as it is crucial: the availability of computing power. To advance AI, in fact, enormous data centers are needed that are more than energy-hungry: they devour energy as much as entire cities. The alarm on this issue is being raised by Wall Street Journal, according to which the growing demand for advanced AI systems is rapidly draining global computing resources, generating tensions along the entire technological and energy supply chain.
“Tokens are worth more than oil”
In recent months, as we know, Artificial Intelligence has seen an unprecedented acceleration, already presenting the risks of a speculative bubble, with investments of hundreds of billions of dollars and a future yet to be clarified. In particular, the rise of "agent" AI, i.e., systems capable of autonomously performing complex tasks, from writing code to operational planning, has multiplied the computing requirement. These are no longer simple chatbots, observes the Wall Street Journal: these tools orchestrate sequences of actions, requiring continuous and intensive use of digital infrastructuresAs engineer and investor Ben Pouladian noted, “the key resource is no longer oil, but tokens,” the units that measure the computational consumption of each request.
Microchips have become a scarce and expensive resource
This Exploding demand is putting pressure on an already fragile systemGPUs, the microchips essential for training and running models, have become a scarce and expensive resource. Hourly rental prices have increased significantly since the fall, in some cases doubling or tripling compared to the previous year. Production, concentrated in a few global companies, is unable to keep pace with market demands. The consequences are already visible: according to the Wall Street Journal, some companies have been forced to withdraw or scale back products due to lack of computational capacity. Others are introducing usage limits, especially during rush hour.
The case of Anthropic and the chatbot Claude
It is the case of anthropic, chatbot developer Claude, which has faced frequent service interruptions and has begun restricting access to resources, sparking protests from its most intensive users. The problem, however, doesn't just affect individual companies, but, as mentioned, the entire technology ecosystem. Data centers, the beating heart of artificial intelligence, require enormous investments and long construction times. Furthermore, their operation depends on an increasingly critical factor: energyTraining advanced models can consume amounts of electricity comparable to thousands of homes, while large-scale daily use further amplifies the demand.
To give some parameters, a large data center of around 100 megawatts can consume over 2.400 MWh of electricity every day, an amount comparable to the needs of 60.000-120.000 homes. Alongside energy, another critical resource emerges: water.Data centers use enormous amounts of water to cool servers and prevent overheating: a single large facility can consume up to 2 million liters per day, equivalent to the water consumption of approximately 6.500 households. In some extreme cases, particularly large facilities use up to 5 million gallons per day, equivalent to a city of between 10.000 and 50.000 inhabitants. In North America alone, total consumption has reached approximately 1.000 billion liters in 2025.
The exponential demand for new technologies has outstripped infrastructure capacity
This dynamic has been seen at other times in industrial history. From the construction of the railways in the 19th century to the internet boom of the early 2000s, demand for new technologies has often exceeded available infrastructure capacity. In these cases, rising prices have been a natural response to scarcity. However, in the current context, such a strategy could prove risky: AI companies are engaged in fierce competition to acquire users and market share, and rising costs could slow adoption.
The result is an unstable equilibrium. On the one hand, millions of users and businesses are starting to rely on Artificial Intelligence to increase productivity and automate complex processes. On the other hand, the technical foundations of this revolution are showing signs of fatigue. The risk, highlighted by the Wall Street Journal, is that the lack of computing power becomes a real bottleneck, limiting access to the most advanced tools precisely at the moment of maximum expansion.
How the industry is addressing the challenge
To address this challenge, the industry is exploring several strategies: from model optimization to designing more efficient chips, up to the expansion of energy and digital infrastructure. However, these solutions require time and significant capital. Ultimately, the growth of Artificial Intelligence will depend not only on software innovation, but on the ability to physically support this transformation. In an increasingly data-driven economy, the truly scarce resource may not be information, but the energy and power needed to process it.
