Every question we ask an AI application, and every image or analysis it generates, relies in the background on data centers housing vast numbers of servers and processors. As AI use expands and its capabilities grow, it requires greater computing capacity—and therefore more and more powerful data centers.

But building these centers involves more than purchasing advanced chips from companies such as Nvidia. Processors require large amounts of electricity to operate, and they generate substantial heat that must be continuously removed. Therefore, in addition to servers, a data center needs a complete network of electrical and distribution equipment and cooling systems.

This is where a broader economic story—larger than the chips themselves—emerges.

McKinsey estimates that the world will need to spend around $6.7 trillion on data centers through 2030 to keep up with growing demand for computing. This includes building and equipping new data centers with servers and processors, as well as the infrastructure needed to operate them, such as power, cooling, and equipment.

Of this total spending, McKinsey expects approximately $5.2 trillion to be linked to data centers used to run and develop AI applications.

The question is no longer simply: Who makes the most powerful processor? It is also: Where will the electricity needed to run it come from, and how will the equipment in which it operates be cooled?

The Bottleneck Shifts to Electricity

The problem begins at the implementation stage. Building and equipping a data center is not enough if the electricity needed to operate it is unavailable, and obtaining that electricity can take far longer than expected.

According to Reuters, major cloud computing companies may want a new data center within just six months, but connecting the project to the power grid can take two years in some emerging markets and more than eight years in some developed markets.

This delay matters economically because a data center cannot operate its servers or serve customers before power is available, and may consequently be delayed in generating revenue.

As the number of data centers has increased, so has demand for the equipment needed to deliver electricity to them and distribute it inside them. One company benefiting from this demand is HD Hyundai Electric, a Korean company specializing in electrical equipment such as transformers and power distribution systems, which data centers need to operate their servers and equipment. The strength of demand for this equipment is evident in the company’s order book: the value of orders it had received but not yet fulfilled reached $8.5 billion at the end of June 2026, with a substantial portion of its production capacity for major equipment already reserved for the next three years.

This reveals an important shift in the AI race: the challenge is no longer simply supplying processors, but supplying the electricity and equipment needed to operate them on time.

The More Powerful the Processor, the Higher the Heat Bill

The second problem is heat. As power density—the amount of electricity used within a small server footprint—increases, traditional cooling methods become less suitable for AI workloads.

According to Bank of America estimates cited by Reuters, power consumption for a single AI rack may exceed 1.5 megawatts by the end of 2030, nearly 100 times that of a conventional rack. This does not mean all data centers will reach this level, but it illustrates the trend toward systems with higher energy consumption density.

This is where Liquid Cooling becomes increasingly important: liquids are used to transfer heat directly from components instead of relying primarily on air. Bank of America expects this type to account for 70% of new AI data center installations by 2030, compared with around 30% today. Reuters, citing McKinsey, reports that these systems can reduce energy consumption by more than 27% compared with air cooling.

Cooling is therefore no longer merely a supporting building service; it has become part of the economics of computing capacity itself, because it affects the number of servers that can be operated, energy consumption, and operating costs.

Where Are AI’s Trillions Going?

The distribution of expected investment explains why the AI boom should not be reduced to Nvidia and semiconductor companies.

Of the $5.2 trillion that McKinsey estimates will be needed through 2030 for AI-related infrastructure, around 25%, or $1.3 trillion, is expected to go to the “Energizers” category, which includes power generation and transmission, cooling, and electrical equipment. By contrast, 60% will go to computing technologies and 15% to site construction and physical infrastructure.

This means that infrastructure is not a side effect of the AI boom, but one of its largest destinations for capital.

Pressure on this infrastructure may also persist as daily AI use expands. McKinsey expects inference—running a trained model to produce answers and analyses—to become AI’s dominant workload by 2030.

In Saudi Arabia, the Story Begins Before the Server

These developments are directly important to the Kingdom.

In a plan announced by the Ministry of Communications and Information Technology in 2021 to develop hyperscale data centers, the Kingdom targeted more than 1,300 megawatts of capacity before 2030, with expected investments of over $18 billion in data centers and renewable energy projects linked to the sector.

It is important to note that these figures are targets from a plan announced in 2021, not a measurement of current operational capacity in 2026. They nevertheless show that linking data centers with energy was part of Saudi Arabia’s vision from the outset.

This means that the success of the Kingdom’s data center expansion depends not only on attracting cloud companies or purchasing advanced processors, but also on an integrated chain that begins with available electricity, runs through transformers and distribution equipment, and ends with cooling systems capable of handling high-density computing workloads.

At the same time, this chain creates a broader economic opportunity. Every expansion in data centers generates demand for design, engineering, construction, electrical equipment, thermal management, maintenance, operations, and grid connections—all parts of the value chain that will absorb global AI-related spending.

The challenge, however, will be turning announced megawatts into operational computing capacity. Allocating land or announcing a data center does not make it a productive asset; power, cooling, connectivity, and equipment must all be in place before it can begin serving customers.

Strong Demand Does Not Guarantee Strong Profits

Even so, not every supplier of transformers or cooling systems is an automatic winner from the AI boom.

Reuters notes that some power and cooling suppliers face high valuations, competition, and component shortages. The president of Delta Electronics also explained that rising revenue does not necessarily mean margins will continue to rise, due to factors such as deployment delays, component shortages, and changes in product platforms.

This highlights the difference between market growth and the investment appeal of a particular company. Demand may be strong while a supplier’s profits come under pressure from costs, competition, or the difficulty of scaling production.

McKinsey also warns of uncertainty surrounding future demand for computing capacity: overinvestment could leave assets generating inadequate returns, while underinvestment could result in missed growth opportunities.

The point, therefore, is not that power and cooling are the “next Nvidia,” but that the economic value of AI has become distributed across a system far broader than the chip industry.

What Should Be Monitored?

Over the coming years, one of the most important indicators may be less exciting than semiconductor stock prices: How much of the announced capacity has actually become operational?

Globally, grid connection timelines, lead times for electrical equipment, the adoption of liquid cooling, and suppliers’ ability to increase production without squeezing margins will be important indicators of how quickly massive investments are becoming actual computing capacity.

In Saudi Arabia, the question is: How will plans to expand data centers become an integrated system of electrical capacity, cooling, and digital infrastructure?

Nvidia and other chip companies may remain the most visible face of the AI revolution, but behind every advanced chip is electricity that must be generated and transmitted, heat that must be removed, and a project that must be connected to the grid before it produces its first unit of computing. As the global spending required on data centers approaches $7 trillion through 2030, the most important investment question may not be only: Who makes the best processor? It may also be: Who can provide the power, cooling, and infrastructure that allow that processor to operate in the first place?