Last year, multinational company Nvidia's market value became $5 trillion, which is bigger than the annual production size of most major global economies at this point. At about the same time, the world's biggest tech companies announced investment plans that may total more than a trillion dollars over the next few years, much of which will be focused on building infrastructure for AI, data centres, and advanced computing systems. AI-led investments are guiding a new trend in the world stock market trends and across major economies - while sucking out capital from emerging market economies and other sectors.
This is proof of a new revolution in technology-led capital investment that has caused the emergence of what is an AI-anchored world order. There are issues with this trend, of course. Commitment to AI-led investment is growing, but there is limited evidence of widespread productivity gains in/across these firms - which we explained earlier in terms of the formation of a tech-fueled mania or bubble. Goldman Sachs estimates that cumulative capital expenditure by the world's largest hyperscalers could exceed $1.1 trillion over the coming years, while Morgan Stanley projects nearly $2.9 trillion in global data-centre spending through 2028. Broader AI-related investment is expected to surpass $1.6 trillion over the same period, extending far beyond software into semiconductors, cloud infrastructure, electricity networks, and specialised computing systems.
While all major companies are blindly betting on big investments in AI, (stock) markets have been signaling other companies to follow trend and governments - some of which hardly understand AI at all, are rethinking their industrial policies and growth-strategies around it. This is troubling-and a trend too well seen in economic history before where the history of economic change through investment pledges made are rarely as apparent as the financial optimism often seen around it.
There is of course a few countries benefiting from this pivot to an AI-led world order. Nations with their stack of industries promoting AI benefit while others lose out. US and China with their foundational AI models; South Korea, Taiwan with their chip-manufacturing industrial strength, and, Israel and Japan with their skilled capital beneficial for AI-investments are obvious gainers.
For others, like Europe, India, Brazil, Philippines, South Africa etc. see muted benefits. The potential for commercial returns from AI is not yet clear in these countries or in other EMEs which are seeing a greater amount of foreign capital move out as well.
In AI-companies too, the productivity gain verdict isn't fully out. OpenAI's own forecasts reportedly predict that it will suffer cumulative losses of more than $140 billion from 2024 to 2029. The price of advancing the technological frontier is continuing to increase rapidly throughout the industry.
According to industry estimates, the incremental improvements in language models will be increasingly expensive in terms of computing power, energy usage and capital expenditure. The economics of AI is starting to look more like massive infrastructure initiatives than software.
Meanwhile, there is less dramatic evidence from the workplace than the market valuations may indicate. A recent study by the National Bureau of Economic Research showed that more than 90 per cent of the firms surveyed said that AI has had little or no impact on their workplaces. Business leaders are still hopeful for future productivity gains, but they are still ahead of what is being seen.
This disconnect between expectation and economic reality is not new. There have been many occasions when significant technological changes have happened in a similar way. The railways were drawing in huge speculative capital before they had had any impact on national markets.
Electricity produced unprecedented investment prior to the productivity impact that it had in industries. The dot-com era saw a lot of excitement about new technologies that have since changed the world, but not always in the time investors had envisioned.
The uniqueness of the present moment is the extent of the financial concentration around a small number of companies and technologies. Nvidia now makes up over 7 per cent of the S&P 500. Capital is also going to the semiconductor manufacturers, cloud infrastructure providers, energy suppliers, and data centre operators. The entire ecosystem has turned into a global investment magnet.Some observers have started to doubt that parts of this ecosystem are becoming more self-referential. The Bank for International Settlements (BIS) and various market analysts have noted the appearance of financing models where investment funds flow within the AI industry itself. Chipmakers invest in start-ups, start-ups buy chips, valuations go up, and more money comes in. These can produce a strong impression of strong demand where it is not actually there. The worry is not so much about fraud, but about the risk that the financial momentum starts to feed back in.
The consequences aren't limited to the stock market alone.
The impact of artificial intelligence on labour markets is becoming more nuanced. The widespread notion of mass technological unemployment has not yet come true. What is coming in the future is a widening gap between those whose jobs can be assisted by AI and those whose jobs can be automated by AI. New data from PwC's AI Jobs Barometer suggests a growing polarisation of the labour market. There are jobs that are growing faster in wages and demand that are using AI to complement specialized skills.There are jobs that are growing in wages and demand that are using AI to complement specialized skills. Routine cognitive tasks are under increasing pressure. Some segments have already experienced significant decreases in translation, localisation and basic content production. Customer-support operations are becoming more and more automated, with companies reporting substantial savings on the number of routine interactions they are having with human representatives.
There's another limitation that's hardly talked about.
Physical infrastructure is becoming a key component of the future of AI. This year, global data centres will use approximately 565 terawatt-hours of electricity, and AI optimised servers are expected to be a growing portion of that consumption. Reliable power is one of the major constraints to expansion in several areas.
This has introduced a curious paradox. The most popular digital technology of our times is increasingly relying on very traditional issues, such as electricity generation, transmission systems, land acquisition, and industrial supply chains. In the advanced economies, energy planners are dealing with speculative demand forecasts from developers wanting future power capacity.
More than two-thirds of utility executives are facing requests for data-centre loads that may never come, according to a recent Capgemini survey. The impact of artificial intelligence is starting to influence investments in industries outside the tech space. In the interim, geopolitical rivalry is getting more serious. Chinese open-weight models are closing performance gaps with the top Western systems, and at much lower costs. As skills grow more common, the potential for fast commoditisation hangs over parts of the industry. The challenge now might be to monetise it in a sustainable way, rather than creating a strong model.
That process can speed up innovation, bring in investment and establish new industries. It can also create distortions when there is a disconnect between expectations and outcomes. The difficulty will be not so much about predicting what artificial intelligence can do, but about understanding the implications of acting as if its benefits are already here.
And so, as the AI booster effect continues to help economies sail through one crisis after another, perceived and projected expectations of such investments in terms of growth and its distributive prosperity are yet to be figured out. Developing nations, including India, will require a different orientation to its industrial policy going forward to absorb benefits in terms of external gains from AI-capital investments to some sectors (like advanced manufacturing and services) while thinking differently for the shared prosperity of its population through other sectors where maximum labour-force participate rates are embedded.
[Deepanshu Mohan is Dean and Professor of Economics, O.P. Jindal Global University. He is a Visiting Professor at the London School of Economics (LSE) and a Visiting Research Fellow at the University of Oxford. Ankur Singh studies Economics and is a Research Analyst with Centre for New Economics Studies (CNES)]
Disclaimer: These are the personal opinions of the author
