The world is entering the most expensive technology race of our time. Companies are building gigantic data centres, buying up processors, laying power grids and reserving computing capacity for years to come, in the hope that artificial intelligence will become the new backbone of the global economy. PwC estimates that the investment required in AI infrastructure by 2050 will amount to $31.6 trillion, whilst Reuters highlights the key question behind this boom: will revenue from AI grow quickly enough to justify the scale of the expenditure?.
- PwC estimates global investment in AI infrastructure up to 2050 at $31.6 trillion.
- Annual capital expenditure on data centres could rise from around $800 billion in 2026 to $1.8 trillion in 2050.
- Anthropic has committed to spending at least $518 billion on AI infrastructure over the next decade.
- AI is already having an impact on the labour market and the energy sector.
$31.6 trillion for infrastructure that does not yet exist
The scale of the current AI boom is best gauged not by the number of new chatbots, but by the volume of physical infrastructure the world is preparing to build to support them. According to PwC’s baseline forecast, global capital expenditure on data centres could reach by 2050 $31.6 trillion. Under a scenario involving faster AI adoption, the figure could potentially reach as high as $50 trillion.
In 2026 alone, the annual volume of such investments is estimated at approximately $800 billion. By 2030, it could rise to $1.1 trillion, and by the middle of the century — to $1.8 trillion annually. The reason is that AI infrastructure requires constant upgrading: servers, accelerators and other equipment will need to be replaced on a massive scale approximately every four to six years.
This fundamentally distinguishes the current wave from the construction of railways or fibre-optic networks. A data centre cannot simply be built once and used for decades without significant new investment. Computing equipment becomes obsolete quickly, and new generations of models require ever-increasing resources.
Anthropic has already committed to $518 billion
One of the most striking examples is — Anthropic, developer Claude. The company’s documents revealed plans to spend at least $518 billion over ten years for cloud and computing resources. Approximately 80% of these commitments are binding, regardless of how actively the company actually utilises the reserved capacity.
Among the largest contracts, Reuters cites approximately $111 billion with Google, $110 billion with Amazon and over $31 billion with Microsoft. Certain long-term commitments relate to Broadcom and other infrastructure suppliers.
For AI companies, the logic is clear: if they do not secure processors and data centres now, there may simply not be enough of them in a few years’ time. But this is precisely where the financial risk lies. Companies are, in effect, placing a huge bet on future demand that has yet to materialise.
Anthropic’s infrastructure commitments are already on a scale comparable to the Stargate project, which involves hundreds of billions of dollars of investment in AI infrastructure.
Money is being spent faster than AI can earn it
It is precisely the gap between capital expenditure and revenue that is gradually becoming the central theme of the AI boom.
According to Bain’s estimates, companies building AI infrastructure may need more than $4.2 trillion in new revenue, in order to justify the current level of investment from an economic point of view.
Goldman Sachs expects that US hyperscalers alone — the largest owners of cloud platforms and data centres — could bring total investment to approximately $1.1 trillion in 2027. The markets are currently viewing this optimistically: AI remains one of the main drivers of capital inflows into technology assets.
The problem is that simply having popular chatbots is not enough to justify such costs for a business. AI must start generating huge, stable revenues — through corporate subscriptions, software development, automation, advertising, robotics, medicine, finance and dozens of other sectors.
In other words, the question is increasingly being phrased not as «Does AI work?», and how «Is it possible to turn its potential into trillions of dollars» worth of profitable economic activity quickly enough?’
Electricity is becoming just as much a resource as processors
Another constraint on the AI boom is electricity. According to the International Energy Agency’s baseline forecast, global electricity consumption by data centres could more than double by 2030, reaching around 945 TWh. That is slightly more than Japan’s current annual electricity consumption.
AI is the main driver of this growth. In the US, data centres could account for around half of the total increase in electricity demand by the end of the decade.
This is already reshaping the geography of the technology industry. Previously, the main considerations for a data centre were access to the network, fibre-optic cables and land. Now, access to large volumes of stable electricity is becoming critical.
This is precisely why technology companies are taking an increasingly active interest in nuclear power, gas-fired power stations, renewable energy sources and new energy storage systems. PwC explicitly identifies the availability of electricity as one of the key factors that will determine in which countries future AI centres will be built.
AI is already changing the labour market — but not in the way we expected
One of the main arguments in favour of the current trillion-scale investments is the expectation of a sharp rise in productivity. If AI enables a single person to carry out work that previously required several employees, the enormous computational costs may be economically justified.
But so far, the results have been mixed.
Researchers at the Stanford Digital Economy Lab, using data on the salaries of millions of American workers, found no evidence of widespread job losses across the economy as a whole due to AI. At the same time, amongst young workers aged 22–25 in occupations most susceptible to automation, employment is approximately 19% lower than the level that might be expected based on trends in occupations less dependent on AI.
The study suggests that the changes are mainly due to fewer young professionals being recruited, rather than through mass redundancies of experienced staff. Where AI As it complements human labour rather than directly replacing human tasks, the impact on employment is considerably less pronounced.
A separate Stanford study, which analysed 1.25 billion job vacancies and 154 million employment records across 41 countries, also revealed changes in the ratio of junior to senior staff within companies that are actively implementing generative AI.
Is this a new dot-com bubble?
Parallels with the late 1990s are obvious. Back then, investors correctly realised that the internet would change the world, but at the same time overestimated the speed at which new business models would start to turn a profit. As a result, the technology proved to be revolutionary, but many of the companies built around it disappeared.
Something similar could happen with AI. The mere fact that artificial intelligence will have a huge economic impact does not mean that every data centre, every AI company or every equipment manufacturer will be able to recoup the current estimates and investments.
Historical technological revolutions have often been characterised by periods of over-investment: initially, society built more infrastructure than it could use effectively, and the real economic benefits only became apparent later.
Therefore, the concept of «AI bubble» This does not necessarily mean that AI itself is overvalued as a technology. It may be that individual company valuations, payback periods and assumptions about the rate of future revenue growth are the ones that turn out to be a bubble.
Even if the bubble bursts, the data centres will remain
This is precisely where the main difference lies between short-term financial risk and long-term technological transformation.
Even if some of the current investment in AI turns out to be excessive, the data centres, electricity grids, power generation facilities, fibre-optic infrastructure and computing capacity that have been built will not simply disappear. They may pass to new owners, become cheaper or be utilised by new companies — just as the infrastructure from previous technology booms continued to operate after financial crises.
PwC believes that AI creates a multi-year investment cycle which differs from previous infrastructure waves precisely because of the need to regularly upgrade computing equipment.
Therefore, the main question for the coming years is no longer whether artificial intelligence will remain. Virtually all the major tech players are behaving as though the answer to this question is obvious to them.
A much more complex issue is — Who exactly will gain more from this revolution than they lose?.
And if revenue from AI does not start to grow as rapidly as data centres, processors and energy capacity are growing today, the world may end up with a paradoxical outcome: one of the most important technologies of the 21st century — and, at the same time, one of the most expensive investment bubbles in history.
