The data centers that make the development of artificial intelligence possible have become a critical new infrastructure for innovation, economic growth and national sovereignty. However, the demand for computing capacity is advancing at a much higher rate than the available supply, which is generating a deficit that will be difficult to resolve without coordinated planning and an unprecedented investment effort.
This is one of the main conclusions of Bain & Company’s 2026 Global Technology Report, which estimates that it will be necessary to invest between 5 and 6.5 trillion dollars to build nearly 150 GW of new capacity by 2030, which would mean practically tripling global capacity in just five years.
The size of the projects is also changing. Just five years ago, a 50 MW data center was considered a large facility. Today, campuses of 5 GW or more are planned, with investments that can be between 150,000 and 200,000 million dollars per installation. In the United States, these amounts already compete with investments destined for railway infrastructure, telecommunications and electrical networks.
This race to expand computing capacity also raises a fundamental economic question: how to make profitable the enormous investment effort required by the new digital infrastructure. Bain estimates that annual spending on AI infrastructure could reach $1.5 trillion by 2031. To sustain that level of investment, the artificial intelligence market would need to generate nearly $6 trillion annually in revenue by that same year.
Of that amount, currently identifiable consumer and enterprise applications would contribute between $1.2 and $1.8 trillion. The remaining 4.2 trillion would have to come from new sources of value, applications and innovations capable of expanding the market.
For Alberto Requena, partner at Bain & Company, the challenge also directly affects Spain. «In Spain, the debate about data centers should not be limited to how much capacity we want to attract, but how we anticipate the system that makes it possible: energy, network, permits, technology and talent. When building a single data center can take more than four years, the advantage will be in coordinating decisions today that will determine available capacity several years from now. “We must go from solving each project in isolation to planning infrastructure, generation and social acceptance together.”
A bottleneck that won’t resolve itself
Realizing this growth potential will require overcoming significant structural obstacles. The availability of power, chips and specialized professionals, along with the complexity of administrative processes to obtain permits and licenses, threatens to slow the deployment of the infrastructure necessary to sustain the expansion of AI.
The magnitude of the challenge requires greater collaboration between companies and administrations, as well as planning that allows capacity needs to be anticipated several years in advance. The expansion of power grids can take more than four years, while essential components such as GPUs, memory and optical networks must be contracted well in advance. Added to this is the shortage of specialized talent and the increasing difficulty in obtaining the necessary authorizations.
Energy pressure will be one of the most determining factors. According to Bain estimates, global data center electricity demand could rise from 388 TWh in 2023 to 1,228 TWh in 2030. In addition, building a 1 GW data center typically requires up to four years, even when power supply is already available. This situation makes energy planning a fundamental piece to guarantee that the new infrastructure can come into operation within the planned deadlines.
Social and regulatory conditions are also gaining prominence. During the first quarter of 2026 alone, opposition from local communities blocked or delayed at least 75 projects valued at $130 billion. Water consumption, the price of electricity, noise and environmental impact are turning social acceptance into a decisive factor for the development of these facilities.
For this reason, companies are paying increasing attention to community engagement initiatives and obtaining the so-called Social License to Operate (SLO), an element that can be as decisive as the availability of land, energy or connectivity.
Comprehensive solutions for a scale challenge
Faced with this scenario, Bain believes that the response cannot be limited to individually resolving the needs of each project. The development of AI infrastructure requires an overview that allows for the coordination of energy investments, distribution networks and the construction of new facilities.
Among the alternatives identified by the consulting firm are the shared purchase of energy, joint investments in network infrastructure and the search for new generation sources. In the short term, renewable energy, storage systems and other conventional sources, such as natural gas, can help cover part of the demand.
In the case of gas, its combination with technologies for capturing, storing and reusing carbon dioxide would allow it to reduce its environmental impact. However, these solutions have yet to demonstrate their viability on a large scale. In the longer term, nuclear power and small modular reactors (SMR) could also be part of the answer to the growing energy needs of data centers.
Orbital data centers: an alternative for the future
The report also considers more disruptive options, such as orbital data centers. These facilities could alleviate some of the limitations faced by land infrastructure, especially in terms of land availability, cooling and energy supply. However, Bain considers that this is still a bet for the future, whose costs, deadlines and technological feasibility are pending demonstration.
The race to build the infrastructure that will sustain the next generation of AI services is a global challenge
In this context, governments could assume an increasingly relevant role as coordinators and financiers of the different initiatives, facilitating collaboration between public and private actors and favoring long-term planning.
For Bain, access to computing power must be managed as a strategic asset. Developers, energy suppliers, administrations and investors will have to coordinate their decisions years in advance to ensure that infrastructure grows at the pace required by artificial intelligence.
Although the United States currently concentrates a good part of the investment drive, forecasts point to double-digit growth in all regions. The race to build the infrastructure that will sustain the next generation of AI services will therefore be a global challenge in which no market can afford to be left behind.
