The AI gold rush: Investments surge, but gains in GDP are only guesstimates

Huge investments are being committed to AI-related infrastructure, but the technology's positive effects on labour market and GDP are only guesstimates

This week, Google committed $15 billion over the next five years to build a data centre in Visakhapatnam. This AI-focused hub will generate nearly 200,000 new jobs. It would have a non-trivial impact on India’s gross domestic product or GDP, which is roughly $4 trillion. Other global multinational corporations (MNCs), such as Amazon and Microsoft, and local giants like Airtel and Reliance, are also investing heavily in data centres. The aggregated data centre buildout therefore will be very significant.

 

Similar stories are playing out elsewhere. Jason Furman, a Harvard economics professor, estimates that AI-related investments accounted for 92 per cent of the United States’ GDP growth in the first six months of 2025. The money went into data centres, including networking equipment, servers, cooling systems, dedicated fibre optic connections, and power capacity.

 

But while the investments are very real, nobody has a clear idea of how much AI will contribute to overall growth in the long term. Indeed, national accounting methods and norms may make it hard to accurately calculate this.

 

The guesstimates are all over the place. The International Monetary Fund (IMF) says AI will affect 40 per cent of global jobs. It will increase global GDP by $7 trillion (about 7 per cent) over the next 10 years, says Goldman Sachs. It could grow the global economy by between $17.1 and $25.6 trillion, according to McKinsey — that’s more than China’s GDP. These are the conservative projections.

 

In September, the NITI Aayog estimated that AI could drive productivity and efficiency and contribute an extra $500 to $600 billion to the Indian economy by 2035, over and above trend growth rates without AI. The largest impacts would be in information technology (IT), finance and finance-related services, but there would be AI-driven growth acceleration across many industries.

 

Also in September, a three-person report from Goldman Sachs estimated AI-related activity had contributed $160 billion to “true GDP” in the US since the advent of ChatGPT in late 2022. The research draws on company reports and government data, and is focused on investments in AI infrastructure to the tune of $400 billion over that period by US firms.

 

After subtracting for imports and adjusting for inflationary spikes, Goldman Sachs reckoned AI contributed $160 billion to US GDP from 2023 to June 2025. But much of that did not show up in official calculations since it had an “intermediate impact” and only final demand is reflected in official data.

 

Official AI-related activity over 2023-25 was calculated to be $45 billion, which is just 0.1 per cent annualised contribution to a $29 trillion economy. Other reports from the US also point to enormous recent investments in AI-infrastructure. Capital expenditure on AI contributed 1.1 per cent to GDP growth between January and June 2025. But the “true” contribution may not be accurately reflected in the GDP statistics due to the intermediate nature of contributions to growth.

 

Going beyond physical and tech-related investments, what is crucial is the positive externalities from deploying AI across economies. All those data centres are expected to drive innovations, productivity and efficiency gains in multiple sectors, and contribute to basic science research.

 

Expectations are very high, which is why Nvidia and OpenAI command the valuations they do, and everybody is prepared to invest large sums in AI. If those expectations are belied, this could turn out to be a bubble. All those data centres might then become the digital equivalent of physical bridges to nowhere — shiny artefacts that serve little real purpose.

 

Nobel Laureate Daron Acemoglu has an interesting and nuanced take on this in a recent paper “The Simple macroeconomics of AI”. He estimates that only about 5 per cent of all current tasks performed in the US labour market can be profitably performed by AI over the next 10 years. While AI may perform many more tasks than this small subset, cost-benefit analysis suggests that it would be more expensive to implement and deploy AI than to continue using existing non-AI modes for most tasks. 

 

His guesstimate is that AI would thus add around 1 per cent per annum to US GDP growth. This is significant but far less earth-shaking than most expectations. If Dr Acemoglu is right, and similar effects hold across the globe, we could see a cooling off in AI investments as the reality becomes obvious. The other possibility is that Dr Acemoglu is wrong and he’s severely underestimated the positive nature of AI’s impact. That is what every investor is betting on.