Markets expect that AI will accelerate GDP growth by 0.35% per year, forever
Virtually everyone agrees that artificial intelligence is going to have a dramatic impact on the economy. Some people are focused on the benefits this will bring, finally restarting productivity growth after a lengthy lull, while others are paying more attention to the likely disruption to many categories of jobs. These two effects are, of course, the same change viewed from two different angles, and both the optimists and the pessimists are mostly right. There will be productivity gains, and for many individuals, those gains will be painful and could involve significant personal loss, which as a society, we need to mitigate as best we can.
Much of the debate about improvements and losses has been anecdotal. We can all think of ways that AI will make our lives better and drive more growth, and we can all think of people who may be negatively affected. There have been a few attempts to quantify the impact of AI on long-term economic growth; many of these seem to be no more than guesses, while others look at individual job categories and estimate how much of the work of each could be replaced by AI.
There’s another way of estimating how AI will reshape the economy, one which has been driven by the collective problem-solving of thousands of very smart people working extremely hard to get to a precise answer. Those people are professional investors, who have to decide what value to place on the large companies powering the AI boom, value that will ultimately be created by taking a share of the economic benefits AI will bring. By making some reasonable (if bold) assumptions, we can understand the value that these investors are placing on the future economic growth that will be driven by AI.
How do we do this? The simplest approach is to view this as an equation:
Where:
- AIValIncrease is the total increase in the market capitalisation of the major AI-centric companies
- AIGDPIncrease is the annual incremental global GDP to be created through AI
- AICosts is the investment required to deliver this growth (which will be the revenue to AI providers)
- AICostShare is the share of AI provider revenue captured by the major AI-centric companies we are analyzing (as opposed to smaller providers, new startups, etc.)
- AIProfitMargin is the profitability of that incremental revenue at those companies
We’ll come back to the discount rate, which we’ll need to in a different way.
For you corporate-finance types out there, you’ll notice that this is a significant oversimplification in many ways (weighted vs. unweighted averages, for example). As you’ll see as we go through this, however, our estimates on some of these will be so rough that a more precise approach won’t make the results significantly more accurate.
We can rearrange this equation to isolate the variable we’re most interested in, of course, which is total incremental global GDP. Note that
Let’s start by looking at some of the largest players in AI and looking at how much their value has increased in the last year and a half. ChatGPT, which kicked off the AI frenzy, was released on November 30, 2022, so I’ll look at changes in market capitalisation between November 2022 and March 2024. A quick glance at Nvidia’s stock-price chart confirms that November 2022 did indeed mark the beginning of a meaningful change in market enthusiasm:
Here’s a table with the market-cap changes of some of the major players in AI over that time:
This is obviously crude. I haven’t accounted for dividends, buybacks, or secondaries, but far more importantly, AI alone hasn’t driven all of these changes. Amazon, which hasn’t done all that much in AI, has almost doubled in value over the same period. And of course these companies are far from the only ones who have benefitted from AI–OpenAI, for example, has come out of obscurity to command a valuation close to $100 billion over this same time period, along with Anthropic, a host of other startups, and much larger companies that have done something with AI but to a lesser degree than the four on this list.
To all of this, my response is: yup.
So in our equation above, let’s set AIValIncrease to $4.85 trillion. Actually, let’s round it up to $5 trillion, to avoid false precision. Now we just have to set values for our other variables. Leaving aside the discount rate for now, I’ve spent hundreds of hours on this, and have come up with the following highly reliable estimates, cross-checked through a wide range of methodologies by teams of leading experts.
AICosts: 25%
AICostShare: 50%
AIProfitMargin: 50%
Are these accurate? I mean, of course not. But are they plausible? Sure. If you have a different view, you can change these numbers and get to a different figure below. I don’t mind.
When we do the math, using our equation from before, this means that the net present value of the future GDP boost from AI works out to an estimated $80 trillion.
Before you stop and think about that for too long, let’s deal with the discounting question. The total value, today, of all that future growth is $80 trillion. But to make this a useful number, we have to work out what that means on an annual basis. For technical reasons that are both complicated and not interesting enough to explain, the correct discount rate to use here is the one used by the market to value the four companies above (not because that’s an important number, but because that’s why our $4.85 trillion market increase is what it is, and not $2 trillion, or $10 trillion). I’m going to use an after-tax nominal discount rate of 10%. A lot of people would consider that low; under other circumstances, I would go higher, but using a lower discount rate potentially helps mitigate what I think many of us would concede could be some frothiness in the stock prices of those four companies.
I’m going to assume AI’s economic impact arrives in the form of a boost to ongoing global GDP growth. To work out what that boost looks like, we forecast global GDP at its current consensus growth rate and discount it back to present, and then solve for the increase in that growth rate that creates an incremental $80 trillion of net present value. The IMF currently estimates global GDP at about $110 trillion, and estimates of long-term GDP growth vary, but cluster around 3%.
Now we’re ready for our answer, given all the assumptions we’ve made up above. When we do the math, it suggests that the adoption of AI will deliver a permanent 0.35% boost to annual GDP growth, increasing the long-term growth rate from 3% to 3.35%
As per our equation above, the way to think about this is:
- The NPV of global GDP, using a current value of $110 trillion, a 3% growth rate, and an equity-market discount rate of 10%, is about $1.57 quadrillion. Note that in any other context, this is a completely meaningless number; you shouldn’t use it for anything else.
- If you increase the 3% growth rate to 3.35% as a result of widespread AI adoption, then the NPV of future global GDP is about $1.65 quadrillion.
- Therefore, the future value of all AI-powered incremental growth is $80 trillion.
- The world economy will need to spend $20 trillion in NPV terms to deliver this growth (AICosts = 25%), which will be the revenue delivered to AI-enabling companies.
- Half of that revenue (AICostShare = 50%), or $10 trillion in NPV, will go to the four companies listed above–Microsoft, Nvidia, Google, and Meta.
- Half of that revenue, or $5 trillion, will be incremental profit (AIProfitMargin = 50%).
- This $5 trillion increase matches the increase in market cap for those four companies since November 2022.
Of course this number is enormously imprecise, and highly sensitive to our assumptions. If you use a 12% discount rate, the implied increase in the long-run GDP growth rate is 0.55%. If you reduce the expected incremental profit margin, or if you assume that other companies take more than half of the AI-linked revenue available in the market, the growth rate also increases.
The 0.35% increase in the growth rate would probably therefore be conservative were it not for the fact that our underlying methodology relies on the obviously false premise that equity-market investors are bidding up the value of the four companies above based on an informed point of view about long-run future profit increases driven by AI. The markets are of course not nearly that clinical, and it’s entirely possible we’ll see volatility in share prices going forward that would change that 0.35% number further still.
However, the financial markets have one redeeming feature in these situations, which is that they are putting real money on the line. While investors have shown time and time again that they cannot forecast the future especially well, they have very clear incentives to do so, and that lends a level of impartiality to this kind of analysis. In the absence of superior alternatives, the estimate that AI will increase long-term GDP growth by 0.35% is a reasonable working assumption.
This sounds like a small number, but it’s not. The NPV figures show that clearly — $80 trillion works out to almost $10,000 per person alive today. That value will arrive in a trickle over time, a small but persistent tailwind, but it will be enough to change lives. If this boost holds, it will be a significant improvement in the lives of billions of people, now and in the future.
