Artificial intelligence (AI) is poised to transform economies, but history suggests its effects will be neither immediate nor uniform. Previous ground-breaking technologies – steam, electricity, and computing – have followed a three-stage arc: early invention with little measurable impact, meaningful productivity gains emerge as adoption widens, then they begin to tail off as the law of diminishing returns kicks in. 

AI’s most obvious potential benefits lie in efficiency gains it could generate: this is known as total factor productivity (TFP).  AI allows tasks to be performed more efficiently, enables people to be more productive in their jobs and frees up time. It also paves the way for better work practices.

Another way of boosting productivity – potentially larger but more uncertain – comes if AI could push innovation. That could go beyond a one-off boost and support a more durable rise in productivity.

Some futurists speculate that next-generation artificial general intelligence – machines matching humans across all cognitive tasks – could produce unprecedented growth. For now, that remains speculative as the next decade will be shaped by today’s more modest technologies and how far and fast they spread across industries and regions.

The extent to which improvements in productivity have a notable macroeconomic impact primarily depends on the degree of technology adoption. The United States is the global leader in artificial intelligence development, yet the US Census Bureau reports only 9.9% of firms using AI to produce goods and services in September 2025.

Ecidence suggests we may be in the early stages of a potential AI-driven capital boom, with US investment in computers and peripherals growing at 42% year-on-year in the second quarter of 2025.

Yet corporate surveys show many projects stall: the share of firms abandoning generative-AI initiatives rose from 17% in 2023 to 42% in 2024, echoing the ‘productivity paradox’ of the 1980s, when computers spread faster than measurable benefits appeared.

The cost of technology and regulatory uncertainty about its future deployment weigh heavily but there are often internal barriers within organisations holding it back. There needs to be investment in complementary infrastructure, like data management, and encryption, while employees must learn new skills. 

In short, many businesses must undergo profound change to exploit the full benefits of AI. McKinsey found that that while 78% of firms use AI in at least one function, only 16% have integrated it across five or more. Barely 1% consider their rollouts mature. 

Studies of AI’s medium-term macro effects show a wide range of estimates. Conservative assumptions tend to downplay productivity gains and international adoption, while more optimistic assumptions point to larger upside risks.

Under conservative assumptions – limited task exposure and small per-task labour savings – US total factor productivity (TFP) might rise just 0.7% over the next decade, producing a 1%-1.15% GDP increase if capital grows in line with total factor productivity. TFP is a measure of economic efficiency that represents how much output is generated from a given set of inputs, such as labor and capital.

More ambitious projections that envisage broader task coverage, higher adoption and the creation of new jobs, suggest TFP gains of 6.8% in ten years. This would include labour cost savings of 40%, even though new tasks would be required.

Some forecasts place medium-term productivity increases as high as 10-15% by 2040.  The divergence in forecast outcomes reflects differing views on the share of tasks AI could perform. Other factors include the proportion of jobs it is profitable to automate and the creation of new work as labour and capital shift to emerging activities.

Historical experience may provide some guide to the future of AI. The computer revolution saw fewer than 10% of businesses using computers two decades after the microprocessor appeared, with productivity gains emerging only in the 1990s.

Views on AI’s future range widely, from those who think that we may soon reach the point that AI matches human performance across cognitive tasks to those who view AI as a “normal” technology with significant but undramatic upside. There are others who argue that the impact of recent advances may be smaller than past technological revolutions.

We tend to think that Amara’s law, attributed to prominent futurist Ray Amara, will hold for AI where: we overestimate the short-run impacts and underestimate the longer-term ones.

AI adoption is proceeding quickly, but there are still many barriers to be overcome before the technology as we know it could unleash its potential. Outcomes will depend on the policies adopted – and how they align with demographic shifts, climate change, the transformation of globalisation, geopolitics and other secular trends shaping the world.