Goldman Sachs: Why AI Spending Is Not Boosting GDP 63
Goldman Sachs, in a research note Thursday (the note isn't publicly posted): Annualized revenue for public companies exposed to the build-out of AI infrastructure increased by over $340 billion from 2022 through 2024Q4 (and is projected to increase by almost $580 billion by end-2025). In contrast, annualized real investment in AI-related categories in the US GDP accounts has only risen by $42 billion over the same period. This sharp divergence has prompted questions from investors about why US GDP is not receiving a larger boost from AI.
A large share of the nominal revenue increase reported by public companies reflects cost inflation (particularly for semiconductors) and foreign revenue, neither of which should boost real US GDP. Indeed, we find that margin expansion ($30 billion) and increased revenue from other countries ($130 billion) account for around half of the publicly reported AI spending surge.
That said, the BEA's (Bureau of Economic Analysis) methodology potentially understates the impact of AI-related investment on real GDP by around $100 billion. Manufacturing shipments and net imports imply that US semiconductor supply has increased by over $35 billion since 2022, but the BEA records semiconductor purchases as intermediate inputs rather than investment (since semiconductors have historically been embedded in products that are later resold) and therefore excludes them from GDP. Cloud services used to train and support AI models are similarly mostly recorded as intermediate inputs.
Combined, we find that these explanations can explain most of the AI investment discrepancy, with only $50 billion unexplained. Looking ahead, we see more scope for AI-related investment to provide a moderate boost to real US GDP in 2025 since AI investment should broaden to categories like data centers, servers and networking hardware, and utilities that will likely be captured as real investment. However, we expect the bulk of investment in semiconductors and cloud computing will remain unmeasured barring changes to US national account methodology.
A large share of the nominal revenue increase reported by public companies reflects cost inflation (particularly for semiconductors) and foreign revenue, neither of which should boost real US GDP. Indeed, we find that margin expansion ($30 billion) and increased revenue from other countries ($130 billion) account for around half of the publicly reported AI spending surge.
That said, the BEA's (Bureau of Economic Analysis) methodology potentially understates the impact of AI-related investment on real GDP by around $100 billion. Manufacturing shipments and net imports imply that US semiconductor supply has increased by over $35 billion since 2022, but the BEA records semiconductor purchases as intermediate inputs rather than investment (since semiconductors have historically been embedded in products that are later resold) and therefore excludes them from GDP. Cloud services used to train and support AI models are similarly mostly recorded as intermediate inputs.
Combined, we find that these explanations can explain most of the AI investment discrepancy, with only $50 billion unexplained. Looking ahead, we see more scope for AI-related investment to provide a moderate boost to real US GDP in 2025 since AI investment should broaden to categories like data centers, servers and networking hardware, and utilities that will likely be captured as real investment. However, we expect the bulk of investment in semiconductors and cloud computing will remain unmeasured barring changes to US national account methodology.
Can you understate AI investment? (Score:2)
Because it isn't increasing productivity (Score:5, Insightful)
AI investment isn't magical new money being created just for AI. If AI didn't exist, that money would still be invested, it would just be invested in other things. So investing money in AI won't raise the GDP unless AI itself manages to raise the GDP. Of the three factors of production, AI effects only labor. By either replacing people or making people more productive it would reduce labor cost allowing people to do more work. Thus far, it isn't achieving this on a large scale (there are probably individual cases where it has, but they're minor). Until AI actually makes people do more, it won't increase the GDP. Which may or may not ever occur, but it definitely isn't there yet.
Re: (Score:1)
Indeed, by definition AI takes away human labor, leading to less human economic activity. Unless AI starts earning adn spending money for itself, GDP is on a downward course.
Re: (Score:2)
What about the stock market returns produced by AI which is not counted in GDP but actually buys stuff (like cloud computing which isn't included in final output according to GS)?
Re: (Score:1)
Indeed, by definition AI takes away human labor, leading to less human economic activity.
That is not true, and certainly not "by definition".
Most of what AI does is not displacing humans but assisting humans.
Even where AI does replace humans, those humans are mostly reassigned, not going to the unemployment line.
It is a very commonly believed fallacy that there is a fixed amount of work to do in an economy, so any labor-saving technique necessarily leads to unemployment.
Lump of labor fallacy [wikipedia.org]
Re: (Score:2)
I think the real fear is having to retrain and more then likely end up making less money for a while. It's just how life is but that doesn't make it less frustrating if you are not the type that naturally is drawn towards continued, purposeful learning. Sometimes our hobbies overlap nicely and other times they have absolutely nothing to do with anything remotely to do with our day jobs.
There will always be winners and losers. Good luck everyone.
Re: (Score:3)
Indeed, by definition AI takes away human labor, leading to less human economic activity.
Others have already responded to this, but I want to add my own $0.02 because this is a fallacy.
Every technological invention throughout all of human history, going back to the wheel or the most basic of primitive tools, has been created in the service of reducing the amount of human labour required to perform a given task.
The trap you've fallen into is " leading to less human economic activity." This is, to put it politely, bullshit. When people have to do less manual labour, they can perform more tasks in
Re: (Score:1)
Why ignore that banks expand their balance sheets to create money, and if AI companies give them good pitches, they can in fact magicalky create money just for AI that they mighg not have created otherwise?
In short why ignore finance in your zero-sum story of investment?
Re: (Score:2)
Well, they create money by giving loans ("inside money creation"). That money is destroyed when it's paid back, so I think in terms of GDP it's neutral?
Re: (Score:2)
It would be if there was no interest on the loans.
Re: (Score:2)
Can the AI companies pay off loan interest with stock market gains from the hype they've created around AI?
Re: (Score:2)
Can the AI companies pay off loan interest with stock market gains from the hype they've created around AI?
Sure, but smart CEOs do the opposite: Borrow money to buy back stock and push up the price.
When the stock price goes up, the CEO gets a big bonus.
Re: (Score:1)
Re: (Score:2)
When a private bank expands its balance sheet on both sides, is it creating currency without the prior approval of the Fed? And if the debt associated with that new currency defaults, how often does the central bank end up creating the money to backstop the bank?
Re: (Score:2)
ah ! That's why they use "monies", plural. Thanks for the info.
Re: (Score:2)
How long can the money stay in circulation with new lians created to pay off the interest on the old loans, kicking the can down the road until the Fed bails them out?
Have you heard that the business model of banjs can be summed up as "put off final settlement for another day", indefinitely?
Re: (Score:2)
GDP doesn't care if a human produced something or not. It measures production, not how..
GDP doesn't care if the produced object is valuable. It measures production, not value.
Right now, our economy is spending a lot of energy and money in data centers to produce AI. It doesn't matter if the AI ends up being broken, or over-hyped. They put a lot of effort into it, and produced s
Re: (Score:3)
GDP more or less equates to goods being produced for consumption. But at the end of the day it's humans that buy and consume things. If technology ultimately completely displace all humans in the work force, who do you expect to buy and consume these goods? Other AIs? I'm sure a society of AIs could build some kind of economy among themselves, but at that point it's essentially just a glorified computer simulation.
Marx was right about one thing. An economy is about people.
Re: (Score:3)
Marx was right about one thing.
Marx was right about a lot of things. He wrote a LOT. But his economic theory is outdated (although new at the time), and any new communist theory should be based on modern economics.
Re: (Score:2)
If there are no consumers (they are all unemployed and can't buy anything), of what use are capital goods? Even with awesome, labor-saving technologies, if people aren't working, they can't buy (read: live), and none of that cool technology amounts to anything in the end.
Re: (Score:2)
Re:Because it isn't increasing productivity (Score:4, Insightful)
But the money wouldn't be sitting around doing nothing if there wasn't AI. It would be invested in other ventures and industries. So it wouldn't be buying servers to run AI, but it would be used to build a factory. Or make video games. Or a dozen other industries. For AI to increase GDP over what it would be without AI in the world, AI actually has to provide value. Otherwise its net neutral to GDP (arguably net negative, if those other industries would have increased the capacity of the nation to produce).
Re: (Score:2)
How much money has AI already created through the stock market? If you didn't invest in NVIDIA does that mean your other investments would have made just as much?
Re: (Score:2)
The stock market doesn't create money. It moves it around.
Re: (Score:2)
GDP doesn't care if the produced object is valuable. It measures production, not value.
This is incorrect.
GDP measures the monetary value of goods and services produced in a country over a time period.
Citations:
https://en.wikipedia.org/wiki/... [wikipedia.org]
https://www.imf.org/en/Publica... [imf.org]
Re:Because it isn't increasing productivity (Score:5, Funny)
Re: (Score:2)
AI is actually good in questions that are hard to find from the Internet, but which are easy to understand and answer by large amount of population. For example chemistry. For example "What molecule contains sulfur, oxygen, nitrogen but not carbon". Searching for this with google is pretty hard, because it will very hard try to give you answers where carbon is included, even when you that is exactly the thing you don't want.
AI (Gemini) on the other hand can give you instantly the answer: Sulfamic acid (HNSO
Re: (Score:2)
I want to clarify that it seems the numbers were lost from the formula when I posted the answer, but Gemini had those originally correct in the answer.
Re: (Score:2)
Post-scarcity economy. (Score:4, Insightful)
The ultimate end-game of capitalism: Its own decommission.
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
Have you considered that actually abundance is the natural state and capitalism creates scarcity by enclosure and selling subscriptions to what used to be free (i.e., you used to be able to sleep outside on commons for free, but the Supreme Court has determined that the state can enacg any anti-sleeping laws that the landlords want so we are forced to pay them to do what used to be free)?
Because it's another tech investment bubble (Score:5, Insightful)
Like crypto, metaverse, its all promise and little or no practical value. Scams enabled by Social Media, a great tool for marketing, fakes and lies, creating value where there is little, like a Trump
Where's the end user? (Score:2)
AI is not adding wealth... for now (Score:4, Insightful)
With a few exceptions, AI is just being used to boost or replace human labor.
Human labor is not the bottleneck of the current economic model. A lot of countries have an excess of human labor. Of course, they are always open to CHEAPER human labor. That's good for the business owner no doubt.
And that could be shown as an slightly progress. After all, human labor is like any other cost. If you reduce it, you can allocate more resources in other place.
BUT... that's the thing. PEOPLE is different, from macroeconomics perspective. People is gonna eat, drink, live in a house, etc. He's gonna consume, though salary or through social welfare.
So saving salaries doesn't change that the economy needs to invest on people one way or another.
If human labor were the economy bottleneck, then after reduce the labor through optimizations, the economy would expand and would get richer.
But, what if there is bottlenecks as energy and raw materials?
Then you only have a wealth redistribution. The owner of the business that got an advantage implementing AI will get richer. The people that work in the replaced/optimized job is poorer. Maybe through government intervention, like taxes, another wealth redistribution occurs and everything ending remain more or less the same.
But that's temporary. As AI gets smarter and smarter, they will start to add real wealth in form of advancements that could allows us to break the real economy bottlenecks. Or at least, we can hope that could happen.
If nor, AI won't benefit most of us. It will just destroy the labor market, with no guaranties that a wealth redistribution will at least compensate that.
But if AI can unlock the solutions for the problems, everything can change.
It's just, AI is still not smart enough. We don't need a million of computer monkeys that replace humans we have in excess. We need a thousand of new Einsteins that give us the right answers for our problems.
Re: (Score:3)
> We need a thousand of new Einsteins that give us the right answers for our problems.
You also need someone to formulate the questions.
42.
Re:AI is not adding wealth... for now (Score:5, Insightful)
We have working solutions for our problems. The wealthy don't want to implement them because they would be less wealthy, and The People don't rise up together and demand them because they have been brainwashed to think that regulations are keeping them from becoming billionaires.
Re: (Score:1)
Re: (Score:3)
Or everyone now has the tools to be wealthy
Having "the tools" doesn't necessarily include having the aptitude, the ambition, or the willingness to shit on everyone else in order to be wealthy.
Re: (Score:1)
Re: (Score:2)
If more people contribute more, everyone is richer for it; the pie grows.
The problem is with the velocity of money. It slows down when you hand it to the rich. They like to claim that they are the job creators, but it's people spending money that makes the economy do work. When you hoard it, it doesn't employ people. It's better for working people if there are no billionaires for this exact reason, they slow the money down and that prevents actually eating the pie.
Re: (Score:1)
With AI, it's different. AI empowers people to create value for their money, and it evens the playing field; I don't need a marketing department, a sales department, etc., to start a business, I can use AI agents to do the work. I don't have money to pay employees, but now billions
Re: (Score:1)
Yes, the velocity of money is higher for poor people. That's why government handouts devalue the currency; no value was created for the additional money printed, so it's just more money actively chasing the same number of goods.
HAHHAhaHAHAHAHAHAHAHHA
no.
Printing money is separate from giving money to people. You only have to do that if you don't adequately (and accurately) collect taxes from the wealthy.
At the time of greatest average prosperity in the USA, the tax rate on the wealthy went up to 90%. Now, it doesn't. It's really that simple.
I don't have money to pay employees, but now billions of people will be able to create products and services without needing a boss.
Explain how AI helps with that when people will have to compete with the AI for work.
Re: (Score:1)
Printing money is separate from giving money to people. You only have to do that if you don't adequately (and accurately) collect taxes from the wealthy
Yes, at least for over the last 20+ years [stlouisfed.org], and there's no sign that's changing, especially with how hard the Ds are fighting DOGE. Clinton did a great job at reducing the deficit, I want those Ds back.
Taxes won't get us there. 90% taxes? That was on income, and the rich don't tend to earn much income. And wealth taxes would cause white flight right out of the US.
Explain how AI helps with that when people will have to compete with the AI for work.
When you create plenty, then things don't have to cost as much. Between robots and AI, we could have plenty for everyone, if the divisiveness do
Re: (Score:2)
90% taxes? That was on income, and the rich don't tend to earn much income.
There is of course a simple solution to that. It's to charge corporations tax on income, not profits.
Re: (Score:1)
Re: (Score:2)
Raising corporation taxes makes them charge more, basic supply and demand.
No shit.
Then, when people can't afford what they're selling, these corporations go out of business
And people buy from corporations which price their products at what the market will bear, instead of more than that.
Then the government has some money, even though they've restricted output and made American's lives worse, for doing things like helping people on the other side of the planet, or killing people all over the planet, or imprisoning people, or colluding with companies that sell us poison to eat and as medications.
One of those things is not like the others.
Re: (Score:1)
You're just taxing the poor with extra steps.
Re: (Score:2)
Re: (Score:2)
GDP only cares if something was produced. It doesn't matter if it "added wealth." It only matters that it was produced.
And it also doesn't care if the people who produced it can afford to buy groceries.
Re: (Score:2)
That's all. It's a measurement.
Re: (Score:1)
The owner of the business that got an advantage implementing AI will get richer. The people that work in the replaced/optimized job is poorer. Maybe through government intervention, like taxes, another wealth redistribution occurs and everything ending remain more or less the same.
But that's temporary. As AI gets smarter and smarter, they will start to add real wealth in form of advancements that could allows us to break the real economy bottlenecks.
Who's going to be in control of the AI? Will it be you and me, or will it be the wealthy who own everything now and who are the only ones who can afford to build the AI? What makes you think they're going to use new technology to benefit humanity at their own cost, when all of history is a counterexample?
Re: (Score:2)
So saving salaries doesn't change that the economy needs to invest on people one way or another.
Would that we could convince anyone involved in the business world or our government that this is a universal truth that can not be escaped. There was a time where our government at least recognized that people on the bottom of the economic ladder becoming slightly wealthier automatically leads to the entire economy growing. Now? They seem to be convinced that the only way to keep the economy running is to gut the finances of the lower classes in order to enhance the finances of the upper classes. Somehow,
Short term idiots (Score:1)
That anyone would think effects of AI would be showing up this early in the game shows how out of touch these people are. This is not the industrial revolution where changing from horse driven plows to motorized plows produced sudden, verifiable gains in productivity. AI is incremental. As we've all seen there is still a ton of work which needs done to get AI to produce even basic outcomes (put glue on your pizza to hold your cheese in place).
AI, like the early days of computes, is a long term project.
Re: (Score:3)
Re: (Score:2)
I do. To think that any effects of AI would show up in the GDP at this point is ridiculous. It hasn't been around long enough to make any meaningful contribution.
That is why I mentioned motorized plows. Once they came into existence their effect (i.e. contribution) to GDP was clear and identifiable. Not so with AI. It will take time for their use to show.
Re: (Score:2)
We are extracting enormous value from LLMs (Score:2)
can you build another couple nuclear reactors, please? We need this tech
Re: We are extracting enormous value from LLMs (Score:3)
I have yet to see an LLM that can weld the 30cm thick reactor steel.
Not difficult to calculate (Score:5, Insightful)
Really? How many times has the news been "Investors question GDP?" The technical answer is simple: Investing in something has nothing to do with GDP. GDP is a measure of consumer spending, so when the bosses give the jobs to an AI, there will be fewer people spending money.
It's not difficult to calculate why US GDP/GNP is dropping. When the US's largest employer (counting permanent, short-term and contractor) stops paying wages/welfare, there will be less spending. When the US government stops buying stuff, that counts as less spending too. Which is why they just decided to stop including themselves in the GDP/GNP totals.
Correct. That's how GDP is calculated (Score:2)
What GS is saying is just restating the obvious about how GDP is calculated. The hardware AI datacenters are buying depreciate so quickly that they are not considered fixed assets, and therefore not included in the GDP totals. In theory, as US Company AI revenue increases and particularly if it increases faster than non-asset investment, GDP will reflect that growth.
The statement being made here is not that the investments can't/aren't contributing (either positively or negatively) to GDP, but rather that