Economy & Markets
1 minute read
AI is a revolution, the optimists say. AI is a bubble, the skeptics reply.
We think AI’s impact could be truly transformative, as we discuss in our 2024 Mid-Year Outlook. The path ahead is uncertain. But powerful forces could drive AI forward.
In this article, we apply an economist’s perspective to an issue at the heart of the optimist-skeptic debate: How might AI impact the overall economy? Among the questions we tackle:
Along with a macro outlook on the potential for AI, and particularly for generative AI,1 we also consider the prospects for investing in this powerful trend. Although valuations on AI stocks have had quite a run, we don’t see signs of a bubble. And while AI-related companies now account for a relatively high share of the overall U.S. market, we don’t think market concentration is necessarily a cause for concern. In short, you may find a wide range of AI-related investing opportunities to consider now and in the years ahead.
In assessing AI’s economic prospects, productivity is a key metric. Faster productivity growth allows the economy to grow more rapidly, and living standards to rise more rapidly, without generating excess inflationary pressures.
The U.S. economy hasn’t experienced sustained productivity gains since the 1990s.2 A reprise of 1990s-style productivity gains could usher in a new era of economic growth.
To understand how AI might (or might not) transform the economy, we consider historical precedents.
In his latest shareholder letter, our Chairman and CEO Jamie Dimon likened AI to the steam engine, electricity and the personal computer. Looking at those episodes of technological innovation, we see that productivity benefits don’t appear overnight.
It took more than 60 years for the steam engine to deliver any observable economy-wide productivity benefit. With each subsequent technological innovation, productivity gains came more quickly.
If the trend holds, and we think it will, by the end of the 2020s, evidence of AI’s productivity boost could show up in U.S. economic data. Think of it this way: It took 15 years for the personal computer to increase the economy’s productivity. AI could do it in seven.
To quantify the potential impact of AI, we modify an International Monetary Fund (IMF) framework. We conclude that AI’s impact could be much greater than the productivity assumptions baked into projections by government agencies such as the Congressional Budget Office.4
The IMF identifies which jobs could potentially be displaced by AI.5 We assume that half of the vulnerable jobs in the United States will be automated away over the next 20 years. The cumulative productivity gain would be about 17.5% or $7 trillion beyond the current Congressional Budget Office projection for GDP.6
It's important to remember that technological innovation tends to boost overall economic productivity only when it changes the amount of labor and capital needed to create a given service or product in the economy. So, for example, Uber did not change the labor-capital inputs (still one driver, one car). But driverless cars, if they eventually appear on our roads, could.
What jobs could be most at risk of AI displacement? Not surprisingly, white-collar professional service jobs such as budget analysis and technical writing look more vulnerable than child care work or pipelaying.7
We also expect to see an educational divide. A Pew Research Center study concludes that workers with bachelor’s degrees or higher are more than twice as likely to be in jobs exposed to AI than those with only high school diplomas.8 The chart below is illustrative:
Across the global economy, AI will likely impact some economies more than others. The IMF has found that workers in advanced economies are more vulnerable to AI displacement than those in emerging market economies. For example, the IMF estimates that 30% of U.S. jobs could be displaced by AI versus less than 13% for India.
All estimates and projections, including our own, should be taken with a grain of salt. No one can precisely predict AI’s economic trajectory, and estimates of AI’s economic impact vary widely. A specific uncertainty relates to the cost of implementing AI technologies in the workplace. Already we are seeing infrastructure costs related to the buildout of AI computer platforms spiral higher.9 Just because a particular job can be automated using AI technologies doesn’t mean it will be if it’s not cost-effective.
Among the optimistic prognosticators, Goldman Sachs projects a 15% boost in GDP from AI over the next 10 years. Our estimates are a little more tempered; we see an 8% to 9% increase to GDP over the next decade. MIT economics professor Daron Acemoğlu takes a much more circumspect view of AI’s potential macro impact, projecting only a 1%–1.5% boost to GDP over the same time span.10
We should also remember that economies evolve in ways that may be best understood in hindsight. According to a study by economists at MIT, more than 60% of today’s U.S. job occupations didn’t even exist in 1940.11
New technologies explain much of that change. Through each successive technology transition, aggregate demand increased and the economy created jobs that didn’t previously exist. The chart below shows the pattern of that change:
Policymakers will need to respond to the societal challenges AI poses. Public policy focused on job training and transitioning vulnerable workers will likely be needed to minimize the upheaval from a greater pace of job displacement.
Next, we shift our perspective from the broader economy to sectors and companies. AI’s economic impact will depend on whether (and how) CEOs and management teams make AI a critical part of their business strategies and operations. Right now, it’s early days.
While most U.S. companies are considering how they could use AI, so far only about 4% of companies have actually adopted the new technology.12 We think adoption rates need to rise to 50% or higher before AI-driven productivity will start to impact the overall economy.13
Adoption faces many headwinds. These include: concerns around the supply of advanced semiconductors, legal and regulatory issues, potentially limited power and energy resources for data centers, and firms’ ability to optimize potential use cases.
Still, the technology makes possible a wide array of potential use cases across various industry sectors. That’s what gives AI the potential to have broad macro implications for the U.S. economy.
The table below highlights the potential use cases:
Many of these use cases will develop over time. But an immediate need is already evident: AI-related activities require extensive new infrastructure. That infrastructure falls into three categories: semiconductors, data centers and electricity power generation. (In an oft-cited example of AI’s thirst for power, a query on OpenAI’s ChatGPT LLM requires 6–10x more electrical power than a Google search.)14
While most investors focus largely on the need for more semiconductors, demand is likely to increase for other parts of the AI infrastructure supply chain as well. These include: data center real estate, engineering and construction firms, copper wire to transmit electrical signals, nuclear and renewable power to support energy requirements, cooling technologies to offset heat produced by the servers, and the electrical components used to connect it all.
In the short run, AI adoption will likely pressure inflation rates higher. But we think the inflationary effects across the entire economy will be modest. That’s mainly because the AI infrastructure buildout doesn’t look to be that large compared to the broader economy.
By our estimates, the cumulative rise in AI capex spend over the next five years will total about $400 billion, which is just 0.3% of expected cumulative GDP over the same time span.15 That’s a far cry from a recent spending surge that did prove inflationary. Cumulative excess government spending on fiscal transfers to households during the COVID years of 2020 and 2021 totaled over $2 trillion, or more than 4% of GDP.
In the long run, we believe, AI’s impact should be more disinflationary than inflationary, but that benefit isn’t likely to be realized until later in the decade.
Currently, AI-related capital investment (capex) is being funded by highly profitable companies with low levels of debt. This was not the case in past periods of technology-driven capex, such as the internet buildout of the 1990s.
The 20 largest capex spenders in the S&P 500 today report EBITDA (earnings before interest, taxes, depreciation and amortization) margins of 26% and net debt to EBITDA (one way of measuring debt leverage) of below 1x. Compare this to the top 20 capex spenders in 1999: They reported 16% EBITDA margins and a net debt to EBITDA ratio of 2.45x.
What’s more, the hyperscalers (large cloud service providers) that are investing 50% or more of the overall AI capex are sitting on huge piles of cash, and their businesses generate high levels of free cash flow. These companies are not issuing excessive amounts of debt to fund their capex. They don’t need to. Given hyperscalers’ ability to self-fund their AI initiatives, it is less likely that today’s high interest rates will curtail their investments.
The absence of excess leverage is important because it means that AI-related capex is not, in our view, creating a worrisome debt imbalance in the economy. This makes for a notable contrast with the dot.com bubble of the late 1990s. That bubble expanded in part through excess leverage—and ended in a bust.
A discussion of leverage circles back to inflation. While today’s AI capex spend may result in pockets of sector-specific inflation (e.g., in the power sector, where we are already seeing a fierce bidding competition by hyperscalers for long-term contracts16), we don’t anticipate that AI will produce meaningful inflationary pressures at the macro level.
How might you think about AI investments in your own portfolio? AI has already catalyzed a wave of excitement, investment and earnings growth. We want to invest for the long term across the value chain.
For now, we suggest clients focus on the AI infrastructure buildout: semiconductors, data centers and new power generation—including the many smaller and less well-known companies. Facing increased demand for their products and services, companies in these sectors have been already meaningfully boosting their sales outlook.
The below chart shows 12 months forward sales expectations for companies in the three infrastructure categories, indexed to when ChatGPT was released in November 2022. Only one category, data center sales, lags the overall market, while the other two have seen substantial upward revisions.
Fair enough, the skeptics say. But isn’t all that promise already priced in? Some of our clients worry that AI stocks are too expensive, and the market too concentrated, to justify investing new money in the trend. AI, they worry, is currently in a bubble.
Certainly, the biggest mega-cap stocks, many AI-related, have surged since the S&P 500 hit its cycle low in October 2022. Yet, when comparing the market of today to the 1990s bubble, we see significant differences. During the last five years, the Nasdaq 100 rose just over 200%. That is nowhere near the >1,000% appreciation that preceded the 2000 market peak. Further, while valuations of the AI leaders today are not undervalued, they trade at nearly half the PE multiple of the leaders in 2000.
AI skeptics often point out that the market is considerably more concentrated than the historical norm. The top 10 stocks currently make up just under 35% of the total market capitalization of the S&P 500 versus an average of about 20% during the 2005–2019 pre-COVID period. Importantly, though, these 10 companies also account for a greater portion of the index’s earnings, now at 26%, than at any time over that period. In 2000, not long before the dot.com bubble burst, the top 10 stocks in the S&P 500 made up 27% of the index’s market cap, but only 17% of its earnings.
We think today’s high valuations for AI-related stocks are justified, given the potential productivity benefits—and the resulting corporate earnings growth—that AI could deliver.
In other words, AI is not a speculative bubble, in our view. It is indeed a revolution, and it has only just begun.
Your J.P. Morgan team can help you consider what kinds of AI-related investing make the most sense for you and your family.
1Generative AI” refers to large language models (LLMs). LLMs are computer programs that learn and generate text, images, audio, software code or other media using an architecture trained on large pools of data. Examples of LLMs include Open AI’s ChatGPT, Google’s Bard, Microsoft’s Copilot, and many others.
2The natural trend labor productivity growth rate of the U.S. economy is about 1.5% annually. That rate picked up to nearly 3% in the 1990s and early 2000s. This allowed GDP per capita to accelerate even as inflation was falling. That combination—rising productivity, falling inflation—occurred in the U.S. economy only once on a sustained basis in the past 50 years, which underscores how important and rare AI’s impact could be.
3We estimate that the full productivity benefits from AI will take effect in 20 years. The comparable timeline for the prior technological breakthroughs varied, from 43 years for the steam engine to 20 years for electricity, to 17 years for PCs and the internet.
4This is not a knock on the CBO projections, which are conservative in nature and serve a valuable purpose in policy planning. However, our framework suggests that private sector economists might entertain further upside GDP potential from AI. When we looked at CBO projections in the early 1990s, we saw they completely missed the macro productivity boom from the internet.
5Mauro Cazzaniga, Florence Jaumotte, Longji Li, Giovanni Melina, Augustus J. Panton, Carlo Pizzinelli, Emma Rockall, and Marina M. Tavares, “Gen-AI: Artificial Intelligence and the Future of Work,” International Monetary Fund, January 2024.
6More specifically: We assume AI labor market displacement follows a classic “S” curve peaking at 15% of U.S. employment, and we make the crucial assumption that aggregate demand follows potential growth through the entire 20-year period, and that displaced workers can easily find new jobs throughout the adoption process. In reality, AI will likely cause some increase in structural unemployment. As a result, our estimates are probably on the optimistic end of the spectrum. To calculate our estimate of cumulative real GDP gains ($7 trillion), we apply our productivity results from AI labor displacement to the CBO’s baseline path of nominal GDP and then net out the cumulative expected (PCE) inflation. We don’t think the AI capex buildout will have a meaningful impact on GDP or inflation, as we explain later in the article.
7That is, at least when it comes to the initial impact of the technology. Much further into the future, AI may interact with robotics to displace manual labor or blue collar professions.
8Rakesh Kochar, “Which U.S. Workers Are More Exposed to AI on Their Jobs?” Pew Research Center, 2023. https://www.pewresearch.org/social-trends/wp-content/uploads/sites/3/2023/07/st_2023.07.26_ai-and-jobs.pdf
9https://www.itpro.com/technology/artificial-intelligence/the-costs-of-building-generative-ai-platforms-are-racking-up
10 https://www.project-syndicate.org/commentary/ai-productivity-boom-forecasts-countered-by-theory-and-data-by-daron-acemoglu-2024-05
11https://news.mit.edu/news-clip/fast-company-202
12Based on data from the U.S. Census Bureau’s Business Trends and Outlook Survey.
13AI adoption rates are higher for specific sectors. For example, a recent survey found a nearly 40% adoption rate of AI technologies in the contact center industry, where productivity gains from AI are already manifesting (https://www.nojitter.com/ai-automation/generative-ai-already-embedded-contact-centers?secureweb=AcroRd32). However, contact center employment in the U.S. represents only about 0.3% of total jobs. Ultimately, the economy-wide adoption rate will matter most for gauging overall macro productivity benefits from AI.
14https://www.epri.com/research/products/3002028905
15The AI projected capex spending numbers are taken from J.P. Morgan Asset Management estimates for the so-called “hyperscalers” (Microsoft, Meta, Amazon, Oracle and Alphabet). We scaled up these estimates to represent the total U.S. economy, and considered the capex ramp-up in excess of the pre-2023 trend.
16Just recently one hyperscaler purchased a 960MW data center from a power provider along with a 10-year commitment to provide a minimum of 480 MWs of power and an additional payment for carbon-free energy sales. https://www.ans.org/news/article-5842/amazon-buys-nuclearpowered-data-center-from-talen/
We can help you navigate a complex financial landscape. Reach out today to learn how.
Contact usLEARN MORE About Our Firm and Investment Professionals Through FINRA Brokercheck
To learn more about J.P. Morgan’s investment business, including our accounts, products and services, as well as our relationship with you, please review our J.P. Morgan Securities LLC Form CRS and Guide to Investment Services and Brokerage Products.
JPMorgan Chase Bank, N.A. and its affiliates (collectively "JPMCB") offer investment products, which may include bank-managed accounts and custody, as part of its trust and fiduciary services. Other investment products and services, such as brokerage and advisory accounts, are offered through J.P. Morgan Securities LLC ("JPMS"), a member of FINRA and SIPC. Insurance products are made available through Chase Insurance Agency, Inc. (CIA), a licensed insurance agency, doing business as Chase Insurance Agency Services, Inc. in Florida. JPMCB, JPMS and CIA are affiliated companies under the common control of JPMorgan Chase & Co. Products not available in all states.
Please read the Legal Disclaimer for key important J.P. Morgan Private Bank information in conjunction with these pages.
Bank deposit products, such as checking, savings and bank lending and related services are offered by JPMorgan Chase Bank, N.A. Member FDIC.
Not a commitment to lend. All extensions of credit are subject to credit approval.