Artificial intelligence (AI) is engendering all kinds of breathless headlines, from being able to play Go to spotting rare cancer tumors. But how will AI impact the economy in broad terms? The answer hinges on both on what AI can be used for and the dynamics of a competitive race to adopt AI that’s set to unfold between firms.
New research from the McKinsey Global Institute simulates the potential global macroeconomic impact of five powerful technologies (computer vision, natural language, virtual assistants, robotic process automation, and advanced machine learning). It finds that AI could (in aggregate and netting out competition effects and transition costs) deliver an additional $13 trillion to global GDP by 2030, averaging about 1.2% GDP growth a year across the period. This would compare well with the impact of steam during the 1800s, robots in manufacturing in the 1900s, and IT during the 2000s.
The average effect on GDP depends on multiple factors. At the industry level they include (a) the extent of AI diffusion in economies; (b) the build-up of corporate profit; and (c) labor market dynamics.
The modeling and simulation relies on two important features. The first is high-quality data from two corporate surveys conducted by MGI and McKinsey in 2007, one of around 1,600 executives across industries globally on digital technologies and AI to ascertain the causes of economic impact and the likely pace of that impact, and one of more than 3,000 corporations in 14 sectors in ten countries. The second feature of the simulation is micro-estimates of the pace of adoption and absorption of AI technologies.
A faster pace of adoption
We know that technologies often take a long time to diffuse and to deliver benefits. It took more than 30 years for electricity to diffuse and enable industrial plant design that could generate significant productivity growth. It took several decades for steam to drive the rollout of railways services and create a large market of exchanges in the United States. Amazon, born 24 years ago, had captured about 45% of online retail commerce in the United States by 2017, but still stood for just about 5% of total US retail gross merchandise volume in that year.
How does AI diffusion compare with the absorption of the early set of digital technologies such as web, mobile, cloud, and big data? Those technologies started to be used about ten to 25 years ago, and the average level of absorption of these technologies was about 37% in 2017. Our simulation suggests that it may reach 70% by 2035. In comparison, absorption of AI might reach today’s level of digital absorption by 2027—in roughly ten years.
There are two stand-out reasons why AI adoption and absorption could be more rapid this time. One is the breadth of ways in which AI is used, including in areas where digitization is still under-penetrated, such as the automation of services and smart automation of manufacturing processes. Second is that returns for front-runners tend to be large. They will benefit from innovations enabling them to serve (and perhaps create) new markets and, at the same time, gain share from non-AI adopters in existing markets. Perception of cannibalization is high among firms surveyed, in line with their experience of early digitization and the emergence of many new business models.
We simulate that about 70% of companies might adopt some AI technologies by 2030, up from today’s 33%, and about 35% of companies might have fully absorbed AI, compared with only 3% today. The econometrics demonstrate that peer competitive pressure is the largest influencer of the decision to adopt AI and make it work across all enterprise functions. The peer pressure effect on adoption incentive is an order of magnitude larger than the expected profitability impact of AI, or perception of the impact it has had in recent years.
A race between firms
Even if a technology race develops, some companies will adopt rapidly, but others less so—and the benefits of AI will vary accordingly. The pace could be enhanced by sector dynamics and by characteristics of firms such as the size and extent of their globalization, but could also be held back by constraints such as early capabilities in digitization, or by organizational rigidities.
We simulated the economic impact of AI for three groups of companies: “front-runners,” “followers,” and “laggards.” The first group experiences the largest benefits from AI, and the second benefits but only by a fraction of the general AI productivity uplift. Laggards (many of them nonadopters) may witness a shrinking market share, and may have no choice but exit the market in the long term.
Regarding front-runners, our average simulation suggests that about 30% of companies might have absorbed the full set of AI technologies in their operations by 2030. About half of those will do so in half the time, and may more than double their operating cash flows by 2030. This is equivalent to sustaining a long-term growth rate of 6% per year through AI. These companies would typically be growing at the rate of high-growth performing firms. Cash generation is not linear as the impact of AI scales up over time—it might be negative in the early years and only becomes positive and accelerates after a period of five to seven years. In this initial period, front-runners could experience cash outflows as they invest in, and scale up, AI. Over time, however, front-runners will tend to slowly concentrate the profit pool of their industry in a winner-takes-most phenomenon.
Followers are firms that are cautiously starting to adopt and absorb AI technologies, having seen the tangible impact enjoyed by front-runners and having realized the competitive threat of not adopting and absorbing. We simulated that 20% to 30% of firms would be in this group by 2030. For these companies, the pace and degree of change in cash flow are likely to be more moderate, and typically below the average productivity uplift witnessed by their economy. On the one hand, front-runners have already triggered some spillovers that spread some benefits to followers; on the other hand, followers lose market share to front-runners.
Laggards are companies that are not investing in AI seriously, or not at all. Why do laggards not jump into AI? The answer is that they may face short-term constraints and may bet—wrongly—that time is on their side. The cost of investment in and implementation of AI means that the divergence among firms on their stance toward AI adoption may only affect their economics after a few years. This may dissuade them from acting. These companies could lose around 20% of cash flow by 2030 compared with today. Laggards may have major capability issues that prevent them from joining the AI race, and therefore they may need to respond in other ways such as limiting costs and cutting investment. The drop in cash flow arrives last, but it is a major slide when it comes.
A fierce competitive race among companies appears to be in prospect with a widening gap between those investing in AI and those that are not. This divide can facilitate “creative destruction” and competition among firms so that the reallocation of resources toward higher-performing companies improves the vibrancy of overall economies. But there is no doubt that the transition may cause disruption and shock in the economy. These tradeoffs need to be understood and managed appropriately in order to capture the potential of AI for the world economy.
This article appeared first in Harvard Business Review.