Capital Expenditure (capex) by Sector The US economy typically has $1.28 T capital expenditure. The table below shows all the sectors of the US economy and their capex. US capex is about 1% of world GDP. What is interesting with AI is that the big hyperscaler companies (big cloud providers) have not only aggressively ramped up their own capex to circa $350bn but have almost entirely reshored it in USA. This scaling and reshoring of capex from the most cash rich companies in the world is having an outsize effect on the US economy. What the AI race is doing, via the Mag7, is sucking up trillions of dollars of worldwide consumer sales and converting that into US infrastructure (one of the benefits of building the world’s dominant tech sector). If they weren’t focusing all of this spending in USA it wouldn’t feel nearly as outsized as it does. This capex cycle is not debt fuelled (not yet), it is not over financialised, not much of the expenditure is even investor capital! It is still mostly coming from cashflow (many big tech companies still have enormous cash reserves that they haven’t yet touched). This is a huge river of wordwide consumer cashflow being funnelled into a very narrow new industrial space by the owners of that consumer cashflow themselves. None of these hyperscalers are going bankrupt, even if AI disappeared tomorrow. They are already cash rich businesses with healthy cash rich balance sheets. A lot of them are trading at P/E ratios in the 20’s whilst they have earnings growth of 15-20%. These are fairly normal financial ratios. Some of them even look cheap by financial ratio analysis. The reason this huge capex tsunami is washing over America is because a bunch of companies really do have the money to spend. There are also some new emergent companies like OpenAI and xAI who are equity financed, but these are not on the public market. These emergent companies are much more financially risky and could pop, as they lack the cashflow of big tech. They are still startup snowballs with burn rate and runway. But as an example and for some context, the oil and gas industry spends around $800 bn a year on capex of which $180 bn is spent in US. If the oil and gas industry increased its focus on USA from $180bn to eg $400bn that would feel like a bubble on the ground in USA, but in oil boardrooms not much would be different, and the enterprise risk would be mostly unchanged. The high geographic focus of AI capital expenditure is most of the bubble feeling. Meta isn’t going to disappear even if it misallocated $250bn on datacenters. The US-centric geographic focus is mostly for the early high training/inference spending phase, and will probably shift to more diversified geographies once inference spending really heats up. This is what big tech capex cycles are going to look like now that big tech has built these huge global cashflow pipelines. Big tech is finally eclipsing oil and gas as the world’s dominant capex sector. Oil and gas spends $500 - 900 billion a year worldwide on capex. But I wouldn’t be surprised to see big tech spending over $1 T each year soon, as earnings are still growing double digit in Mag7. TLDR: AI capex numbers don’t look crazy/scary from the perspective of an energy sector analyst, and behind the capex big tech has the money. I wouldn’t even consider the mag7 super mature, as they still haven’t even had their first earnings winter, not even a pandemic dented their growth. Having said all that… AI definitely needs to start generating earnings to justify the capex, but these earning will be small with high exponent growth rates for a while and it’s the second order differentials of growth that matters for now (and a long time to come). So when you have journalists and politicians claiming Elon Musk and Mark Zuckerberg don’t understand business it’s because journalists and politicians are not smart people who don’t understand differential calculus or financial mechanics.