Hacker News Reader

Comment by cs702 | Hacker News Reader

Key quote:

> The hundreds of billions of dollars companies are investing in AI now account for an astonishing 40 per cent share of US GDP growth this year.

In other words, investment in AI is a massive private-sector stimulus program, powering economic growth.

Questions for which no one knows the answer include:

* Will these AI investments earn a positive rate of return in the aggregate?

* Without these investments on AI and their second-order economic effects, would the US economy be growing?

* Are the higher-order effects of AI investments temporary, or can they become self-sustaining (for example, via creation of new jobs and industries)?

View on HN

> Without these investments on AI and their second-order economic effects, would the US economy be growing?

It's only 40% of the growth so.. yes.

That 40% is only the investment in companies making AI products. It does not include "second order" effects.

Not necessarily. It depends on the second-order economic effects of these investments in AI.

For example, companies that build new giant data centers for AI pay salaries to thousands of temporary construction workers who then spend some of it on retail goods and services. That additional spending by the workers is a second-order effect, and it shows up on GDP growth. There are higher-order effects too, but they are more diffuse.

Given how much of the investment goes into the stuff inside of the data center and then the power to run it both of which don't really go out to the US economy so much as the builders' wages I feel pretty safe in saying it's not the remaining 60% of GDP growth.

The incremental revenues to software and hardware vendors and energy providers is also used to hire new people, build new infrastructure (manufacturing plants, energy production systems, etc.), with second-order economic effects too. There are higher-order effects too, also more diffuse.

Again I'm not denying second order effects exist. Just saying it's not 150% of the direct effects.

Ah, got it. I'd say it's possible but no one knows for sure. There's too much related spending that's hard to measure, including, for example, organizations spending to try out newfangled AI services. I don't think that kind of spending is counted in the investment figure, because it's not capex.

You're right, but I find your take too clinical and ignoring the human factors, such as overall market confidence.

Another way to look at this is asking if AI investments get wiped out, would they take down GDP growth with them? I think there would be a pull back across the board. I will speculate that without AI investment juicing growth, we would have seen broad investment pullbacks in the 2nd quarter, and a shrinking GDP.

I wouldn't be so sure, especially since we know a lot of it is actually vendor financing/circular financing (Nvidia give money to OpenAI who buy dedicated datacenter from oracle who buy GPUs from Nvidia), so a part of the growth might have to be counted twice, or more.

Wouldn't that just mean it's less of the actual GDP growth? Either way if it's only 40% of the calculated figure double counted or not I have a hard time seeing how the secondary effects would account for the remaining 60% too, which would be 150% of the original first order effects.

Market confidence in growth is a second-order effect. Non-AI investments would decrease markedly if growth is stalling.

The 40% investment could presumably go into jobs and infrastructure that might be more durable.

That 40% investment used to be sitting in giant cash slushfunds on company balance sheets. This isn't government investment. It would have never been spent on 'infrastructure'.

On the upside, even if AI is a bubble and implodes spectacularly we're going to have metric tons of compute available for climate models, drug discovery and anything else that requires a super computer to model.

Will we? The hardware you want for AI and the hardware you want for super computing seem to have different priorities, e.g. concerning floating point precision.

compute is compute, yes. Maybe some older models might have to be rewritten to take advantage of a gpu farm's parallel processing capabilities, but humanity will ultimately benefit if the current AI boom fizzles

I think you would find a large political block that would be more than supportive of economic stimulus taking the form of a public jobs and training program to improve infrastructure, landscaping, and other public works.

I personally would implement it as a "civilian military" to make it a cool high-status line of work that people might choose for a few years out of civic duty. Uniforms, ranks, bases, parades, you get your food/housing provided, you get 'deployed' on projects. Get your college paid for if you stay on long enough.

And there is a precedent for that the Civilian Service Corp, which was created as part of the new deal. If you go to a national/state park there is a good chance they built some of that infrastructure.

LLMs are infrastructure.

is this real? by first look USA only spent like 4% of GDP on space stuff at the peak of space race. 40% is insane.

40% of growth, not 40% total

I think its 40% of growth, not gdp?

Whoops, my bad.

40% of *growth*, not 40% of total GDP.

Reply on news.ycombinator.com