Aswath Damodaran
Big Tech Has No Idea How AI Pays Off (August 2026)
Q2 revenue growth for big tech was great. The marginal returns relative to the incremental investments show a shift in business model. These are very different companies than 5 years ago.
A lot of these companies have negative cash flow, which would have been hard to believe a few years ago. This is new territory for the Mag 7, with the exception of Amazon which has a history of building back from a negative cash flow position.
Apple isn't really investing any more capital than they did in the past. The other Mag 7 companies are becoming like manufacturing companies.
Measures like return on invested capital, which didn't used to be important for these companies are becoming useful now.
Facebook should have learned from the metaverse that spending money is easy, but building the business model is hard.
Apple has made a distinct decision to not increase their CAPEX. It will be really interesting to see who is making the right decision.
Apple doesn't like jumping in with two feet when they don't know the future. A more ambitious Apple CEO would have charted a different path.
He thinks these companies should make a separate AI division, just like there is a separate cloud compute division. This would give markets a segmented financial statement.
There are some estimates that OpenAI and Anthropic make up 73% of AWS AI revenue. Similar numbers for Google AI revenue.
The Mag 7 are middle-age companies. They are amazing, making a lot of money, and healthy, but middle-aged nonetheless. They all want to be in the cool-AI crowd. It's deeply embedded in the current narrative.
NVIDIA is financing a lot of their customers. Intra-company investing makes valuing companies harder.
Look at the enterprise value to EBITDA ratio. We net cash out cause the income from cash is not part of EBITDA. We want the numerator and denominator to have the same components. This gets complicated when you have cross-holdings. Let's say you have a 30% stake in a frontier AI lab. That raises your enterprise value (the market factors your 30% stake in your stock price), but the revenue from the underlying holding isn't part of EBITDA. So you need to net out the 30% stake from enterprise value or else the company will seem more expensive than reality.
For 20 years, he has been arguing that the P/E ratio is the most dangerous multiple to use.
Net income at companies like Alphabet is a mess. Look at the operating margin and interest expense separately.
It's possible that the market will bifurcate and there will be a high-margin AI and another low-margin, high-scale market.
We still haven't seen a major CAPEX writeoff from a company yet. That would be an admission that a bad bet was made.
Aswath's blog is Musings on Markets.