AI has already cost $1.5 trillion. But how much money has actually come back?
Try to imagine betting huge sums of money on a single project for three years, and at the end you still wouldn't really know how much you got back. That's exactly what almost the whole of Silicon Valley is doing with artificial intelligence right now. Billions are flowing into data centers, chips, and models at a pace the tech world hasn't seen, and yet the question "is anyone actually making money from it?" mostly remains without a concrete answer.
There's a website that decided to answer this question bluntly, number by number.
It's called isaiprofitable.com, it tracks the cumulative spending and revenue of the biggest AI players and updates continuously, complete with a counter of how much money is "burned" every second you spend reading the page. The author is one independent developer, drawing from SEC filings, earnings calls, and estimates from Bloomberg, WSJ, and Epoch AI. So it's not an audit, but it's the most comprehensive publicly available overview that currently exists on the topic.
Who's spending and who's earning?
According to the latest numbers from isaiprofitable.com (as of July 2026), AI has collectively spent roughly $1.5 trillion and generated approximately $769 billion in revenue.
The gap of over $700 billion sounds scary, but the real insight is in how unevenly those losses and gains are distributed.
Amazon $AMZN leads in spending with $331.1 billion in expenses against $22 billion in revenue, for a net position of -$309.1 billion. Right behind it is Alphabet $GOOG with expenses of $302.8 billion and revenue of $25 billion, net position -$277.8 billion. Meta $META spent $238 billion and earned only $3 billion, net position -$235 billion, which, given the tiny revenue, is the worst ratio among the big four. Microsoft $MSFT put $256 billion into AI and got back $31 billion, net position -$225 billion.
Oracle $ORCL is a smaller player, expenses $92 billion, revenue $45 billion, net position -$47 billion. OpenAI spent $62 billion and earned $26.7 billion, net position -$35.3 billion. xAI has expenses of $20 billion against revenue of $7.4 billion, net position -$12.6 billion. Anthropic, among the pure AI labs, is closest to breaking even, expenses $33 billion, revenue $22 billion, net position -$11 billion.
Then comes the turning point. AMD $AMD invested $27 billion and reaped $36 billion from it, net position +$9 billion. Micron $MU put $26 billion into AI and got back $34 billion, net position +$8 billion. And at the very top is Nvidia $NVDA, expenses $228 billion, but revenue a full $516 billion, net position +$288 billion, by far the highest profit in the entire table.
One name stands out at both ends of the table. Nvidia is the only company massively and unequivocally profiting from the whole boom, because it's selling shovels to everyone who's digging. While the large consumers of AI infrastructure sit deep in the red, Nvidia has earned over $500 billion in revenue since 2023 on an investment of less than $230 billion. That's the purest illustration of the old investor truth: in every gold rush, the one who sells the picks and shovels ends up making the most, not the one who digs.
On the other side are the big four US giants, Amazon, Alphabet, Meta, and Microsoft, who together spent over $1.1 trillion and have gotten back less than $80 billion so far. But one important clarification is needed here, which the website itself openly admits. These numbers don't mean these companies as a whole are losing money, because Amazon and Google are still huge profitable businesses. The figure only measures whether their specific AI investment has paid off, and the expenses also include the entire capital expenditure on data centers in the year it's spent, without spreading it over future years. At the same time, the site itself admits that it doesn't include indirect benefits in revenue, like how much AI boosts Google search performance or Microsoft Office sales with Copilot, because that simply can't be reliably isolated. So the reality is probably a bit more favorable than the raw table shows, but not enough to fundamentally change the picture.
The second thing worth keeping in mind is the interconnectedness of the whole system. Google invests in Anthropic, Anthropic runs on Google Cloud, Amazon gives money to Anthropic, and Microsoft is linked with OpenAI. A portion of those trillions thus effectively circulates back and forth among the same handful of companies, which means the aggregate numbers for the whole industry slightly overstate how much genuinely new money is flowing in from outside.
How do I see this whole thing?
Amazon mám v portfoliu jako jednu z největších pozic, takže tahle tabulka se mě osobně týká víc, než by se mi líbilo přiznat. Vidět -309 miliard u firmy, kterou držím, není příjemný pohled. Ale nekoupil jsem Amazon kvůli AI. Koupil jsem ho kvůli AWS, e-commerce marži a schopnosti generovat cash flow i v letech, kdy zrovna nikdo nemluví o umělé inteligenci. AI výdaje u Amazonu financuje především právě zisková AWS, ne nový dluh na hraně přežití. Dokud tahle dojná kráva funguje, dávám firmě čas, aby AI sázka dozrála, přesně tak, jak to fungovalo s cloudem samotným, který taky roky spaloval peníze, než se stal nejziskovější částí byznysu.
At the same time, that's precisely why I parted ways with Apple $AAPL. There I didn't see any comparable cash cow to finance a quiet AI bet, and most of all I didn't see any real bet. Apple doesn't appear in this AI table at all, which in itself says something. The company with the highest margins in the sector has yet to deliver a convincing AI product or a comparable capital commitment to its peers. In an environment where it's being decided who will supply the infrastructure for the rest of the economy in 2030, the absence at the starting line feels to me like a bigger risk than the temporary negative cash flow of the competition.
But what worries me more than individual names is the pace. Depreciation on AI hardware runs at about twenty percent a year, meaning chips bought today will be obsolete before they're fully depreciated, and yet purchases keep accelerating. That's behavior I'd normally label as addictive, not rational. Companies aren't spending because they've calculated a return. They're spending because they're afraid that if they stop, the competition will leave them far behind in three years. And history shows that fear as an investment decision-making process usually ends painfully for those who finance it, and advantageously for those who survive and buy the wreckage cheaply.
Do you think the current corporate spending on AI and infrastructure purchases will show up in margins and earnings?
