On July 30, Amazon reported one of the strangest quarters in its history: record profit, record spending, and — running quietly underneath both — a smaller headcount in the exact team building the thing all that spending is for. The numbers don’t contradict each other. They’re the whole thesis of this site, printed on a single earnings slide.
The two record numbers
Second-quarter net income came in at $62.6 billion, or $5.75 a share — more than triple the $18.2 billion, or $1.68 a share, Amazon posted a year earlier. AWS, the cloud engine funding the whole AI buildout, grew almost 37% to $42.23 billion, its fastest expansion since 2021 and well past the $40.54 billion analysts expected.
Then the other record: Amazon told investors it now expects capital expenditures to hit $220 billion this year, up from the roughly $200 billion it guided to in February, blaming rising memory prices. That is not a typo and not an annual-run-rate trick. In a single year Amazon plans to spend more on data centers, chips, and AI infrastructure than the entire annual GDP of most countries on the map.
So the scoreboard reads: profit tripled, cloud accelerating, and a fifth of a trillion dollars going into machines. Now look at where the humans went.
The team building the AI is the team getting cut
Earlier in July, Amazon cut roles inside its Artificial General Intelligence organization — the group behind its Nova foundation models — eliminating model-customization and post-training jobs as it “sharpens its focus” toward enterprise AI products. This wasn’t the retail side trimming seasonal fat. It was the frontier-research bench — the people literally building the intelligence — being thinned while the budget for the intelligence went up by $20 billion.
Zoom out and the shape is clearer. Amazon has removed roughly 30,000 roles since late 2025 across its restructuring. Over the same window, its AI capex line went from “enormous” to “unprecedented.” The company is not spending less on people because it’s spending less overall. It’s spending more than ever — the money is just being pointed somewhere a payroll doesn’t go.
What $220 billion actually buys
Here is the uncomfortable arithmetic. $220 billion at, say, a fully loaded $200,000 per US tech worker would fund over a million jobs for a year. Amazon is choosing instead to spend it on GPUs, memory, and concrete. That is a revealed preference, stated in the bluntest language a corporation has: the capital allocation slide. When a company that can afford anything decides the marginal dollar goes to silicon rather than salary, it is telling you exactly how it expects to create value for the next decade.
And the market loves it. The same investors clapping for 37% AWS growth are clapping for the mechanism that produces it — more compute, more automation, flatter headcount. That applause is the incentive, and the incentive is not going to reverse because a research team in Seattle got smaller.
The LostJobs read
The comforting version of the AI-layoff story is that cuts hit the losers — the companies AI is eating. Amazon breaks that story in half. It is the single biggest winner of the AI trade, posting a tripled profit and its fastest cloud growth in five years, and it cut the very team building its models in the same month it raised AI spending to $220 billion. “The business has never been better” and “your role is eliminated” are, once again, the same sentence.
The only defensible position is the one that doesn’t sit on the wrong side of that capex line: be close enough to a decision, a customer, or a judgment that the $220 billion can’t be pointed at your seat. Amazon just showed, in the plainest numbers it has, where it thinks the value is going. It is not going to the people. It is going to the machines the people are being paid, briefly, to finish building.
