AI Needs a $6 Trillion Market. Today’s Use Cases Are Not Enough

The race for artificial intelligence is entering a stage in which access to the most powerful processors is no longer the only challenge. Annual spending on AI infrastructure could reach $1.5 trillion by 2031. At the current relationship between investment and revenue, that would imply the need for a market generating roughly $6 trillion a year. Bain & Company estimates that consumer and enterprise use cases visible today may produce only $1.2–1.8 trillion. The rest would have to come from products, services and entire markets that do not yet exist.

The largest technology companies are spending sums on artificial intelligence that only a few years ago would have been difficult to associate with a single technology. Microsoft, Alphabet, Amazon, Meta and Oracle are increasing investment in data centres, processors, memory, networking equipment and the energy required to power them.

According to Bain & Company, their combined capital expenditure could reach about $780 billion in 2026. That is almost five times more than three years earlier.

The scale of spending, however, is not the central issue. The more important question is whether new AI applications will develop fast enough for the infrastructure now being built at enormous scale to generate corresponding revenues.

$1.5 trillion in annual investment

Bain assumes that annual spending on AI infrastructure could rise to around $1.5 trillion by 2031. The figure includes not only new data centres but also replacement of processors, memory and networking equipment in existing facilities.

This is an important difference from traditional infrastructure. A data-centre building may operate for many years, but the AI servers inside it become technologically obsolete much faster. A new generation of accelerators can perform the same calculations more quickly and with lower energy consumption, giving operators an economic incentive for continuous upgrades.

The result is not only an enormous construction bill, but also a lasting cost of maintaining competitive computing capacity.

$1.5 trillion of investment implies a $6 trillion market

If capital expenditure represented about one quarter of industry revenue, $1.5 trillion of annual CAPEX would require around $6 trillion in revenue. This is not a forecast that the AI market will definitely reach that value.

It is rather an answer to the question of how large the market would have to become to justify maintaining investment at the assumed level.

The problem is that the revenue sources visible today are much smaller. Bain estimates that consumer AI applications, including paid subscriptions and advertising, could generate $200–400 billion in 2031. Enterprise applications driven mainly by productivity gains could add another $1–1.4 trillion.

Together that gives $1.2–1.8 trillion. Relative to the level implied by the current investment trajectory, there is still a gap of roughly $4.2 trillion. That gap is the most important number in the report.

Productivity alone will not be enough

The debate about the economics of generative AI has so far focused mainly on time savings. A programmer can prepare code faster, an analyst can produce a report more quickly, a marketer can draft campaign content and a consultant can answer a client sooner.

That has measurable value for an enterprise. It does not automatically create a new market.

If a company uses AI to perform the same work with a smaller team, or serves more customers without adding staff, productivity improves. Part of the value is captured by the technology provider, but a large share remains with the company using the tool.

With investment measured in trillions of dollars, more is needed. AI has to create products and services that could not exist without the technology, or would otherwise be too expensive to offer at scale.

Advertising could add another $100–200 billion

The first natural direction is the online advertising market. If part of information search shifts from traditional search engines to AI assistants, advertising budgets may move with users. Bain estimates the additional potential of this segment at around $100–200 billion.

That is a large sum by the standards of almost any technology market. Against a gap of more than $4 trillion, however, it looks much smaller. A new way of searching for information will not be enough on its own.

AI has to move beyond the screen

Much larger opportunities emerge when artificial intelligence begins controlling devices operating in the physical world.

Bain points to autonomous cars and trucks, robotaxis, drones and industrial automation. The potential of this group of applications by 2031 is estimated at around $400 billion.

The economics are different from those of a chatbot. An autonomous truck can change transport costs, a robot can take over tasks previously performed manually, and an autonomous system can increase utilisation of an expensive machine around the clock.

AI then stops being an add-on to software and becomes part of the production process.

Physical AI could become a larger market than consumer generative AI

Bain sees even greater potential in so-called physical AI. This category includes robotics, digital twins, industrial simulations and systems that use AI models to design and optimise real-world processes.

The potential value of this area is estimated at around $900 billion.

In industry, the ability to test thousands of process variants before rebuilding a real factory can reduce cost and investment time. In automotive and aerospace, models can support the design of new structures, while in semiconductor manufacturing they can optimise some of the most complex processes in the world.

This is where value far greater than the price of a monthly AI-assistant subscription may emerge.

We do not yet know the biggest market

Even after advertising, autonomous transport and physical AI are included, the numbers still do not add up. The largest part of the required value must come from applications that cannot yet be priced precisely.

Bain points, among other areas, to drug and materials discovery and to partial automation of scientific research. If AI lowers the cost of developing therapies, drugs could become economically viable for patient groups that are too small for today’s business model. A similar mechanism may work in designing materials for energy, batteries or semiconductors.

The biggest part of the AI bet is therefore not whether people will use chatbots. It is whether the technology creates economic activity that does not yet exist.

Meanwhile, data centres are growing from megawatts to gigawatts

The market is still looking for new revenue sources, but the infrastructure is already being built. A few years ago, a 50 MW data centre was considered a large project. The biggest AI-related developments are now measured in hundreds of megawatts, while future campuses are expected to reach gigawatt scale.

Bain says the largest complexes could approach as much as 9 GW by the end of the decade. That changes the nature of the business. Building a data centre increasingly resembles an energy-and-industrial project with a very large technology component.

Data centres may need about 150 GW of additional capacity by 2030

In a related analysis of data centres, Bain estimates that global installed IT capacity could rise from about 79 GW in 2025 to 228 GW in 2030 in the base-case scenario.

The lower-growth scenario assumes about 188 GW, while the high case reaches 262 GW. In other words, within five years the world would have to build infrastructure equivalent to roughly another 150 GW of IT capacity under the base case.

Energy is becoming part of the technology race

Such rapid infrastructure growth is impossible without sufficient energy. According to Bain’s model, electricity consumption by data centres could rise from 388 TWh in 2023 to 1,228 TWh in 2030.

At the same time, demand is increasing for transformers, grid connections, cooling and network equipment. The time needed to build the data centre itself may be shorter than the time required to secure adequate power.

This is one reason why the largest technology companies are becoming increasingly involved in energy. Options include renewable power contracts, gas-fired plants, energy storage and, over the longer term, new nuclear sources.

Access to electricity is no longer merely an operating cost in the background. It is becoming one of the factors that determines where another large AI installation can be built at all.

The boom extends far beyond GPU manufacturers

Growing demand for computing power is often reduced to Nvidia’s results. The actual investment chain is much longer.

Bain estimates that shipments of GPUs and specialised ASIC chips could rise from 3.9 million units in 2023 to 38.6 million in 2030. The cooling-technology market could increase over the same period from $7.9 billion to $28.4 billion.

On top of that come memory, networking hardware, fibre optics, backup power, transformers and construction infrastructure. Every additional gigawatt of computing power therefore triggers investment in many industries at once.

Capital itself may also become a constraint

The larger data centres become, the smaller the group of companies able to finance them independently. If a single campus requires tens of billions of dollars, the investment no longer resembles a traditional IT project. At the scale of the largest planned facilities, financing begins to look like major infrastructure funding.

This may lead to further market concentration. The strongest position will belong to companies that simultaneously have access to capital, energy, processors and customers willing to use the resulting computing capacity.

Bain points to an analogy with semiconductors. As the cost of each new generation of fabrication plants increased, the number of companies capable of producing the most advanced chips steadily declined. A similar process could begin in AI.

The biggest risk is not technological

A data centre can be built. Processors can be ordered. The power grid, although with delays, can be expanded. What is much harder to plan is the creation of a market worth several trillion dollars.

Technology companies, however, have to make investment decisions earlier. Large data centres take years to build, as does securing enough power. If operators waited for full confirmation of demand, they would not have the infrastructure ready when customers started to need it.

That is why some infrastructure is being built ahead of demand. It is the biggest economic bet of the current AI boom.

The stakes include an extra percentage point of global growth

Bain illustrates the scale of the challenge in another way. To generate value corresponding to the infrastructure being built, new AI applications would have to lift the pace of global economic growth by about 1 percentage point a year.

That means AI would have to extend far beyond the technology sector. Its impact would need to be visible in industry, transport, pharmaceuticals, energy, services, scientific research and many other sectors. Only then could enough new economic value be created.

This is how the most important general-purpose technologies worked. Electricity did not create only the market for power plants, and the internet was not limited to selling network access. The greatest value emerged in companies and services that had not existed before.

The next few years will show whether AI can do the same

In the debate about artificial-intelligence investment, it is easy to focus on ever larger models, new generations of GPUs and the number of data centres being built. Bain’s analysis suggests, however, that the decisive test will be economic.

Use cases known today do not generate value corresponding to the scale of infrastructure planned for the beginning of the next decade. The difference cannot be covered only by more chatbot subscriptions or by savings from faster coding and office-work automation.

New markets are needed. Autonomous transport, robotics, drug design, digital twins and automation of research show where they may emerge. It is not yet clear which of these applications will reach sufficient scale.

The technology industry is not waiting for the answer. Data centres, power plants and chip factories are already being built.

The biggest question is therefore no longer whether the world can build infrastructure for AI. The question is whether it can find applications worth $6 trillion a year.