Small capex updates were received again in this week’s earnings announcements.
| Old | New | ||
|---|---|---|---|
| Meta | 145 | 145 | |
| Microsoft | 190 | 190 | |
| Amazon | 200 | 220 | |
| 190 | 205 |
The figures are at the upper ends of the capex guidance. Meta only raised the lower end of its guidance. MSFT’s official figure, on the other hand, dropped to 175B, but this is an accounting matter, and in other respects, the indication remained unchanged (related to lease accounting).
So now we are already above the $750B level I guessed earlier this year. It is worth noting that every company practically indicated a significant need to increase capex levels for the coming year as well. Of course, it is a bit concerning for NVDA that the capex increases were more attributed to rising memory prices than to growth in raw compute demand (though these go hand in hand).
This is a valid concern of mine as well, and the reason why selling NVDA is itching quite badly. For example, AMZN and GOOGL provided very positive indications regarding the demand for their own chips (both internal need and external demand). So, competition is tightening.
Of course, it should be pointed out regarding that capex chart that for the current year alone, big tech investment figures are at least $100B too low, and the figures predicted for 2027 will, based on what was said this week, be realized within the next 6 months according to the calculation I presented above—unless something truly strange happens.
As for the demand outside of the hyperscalers, it is probably easiest to look to Huang for ideas. At the March GTC, it was stated that the HCP/other ratio was about 3:2, as I understand it, for the $1300B in sales already realized and in the order books for 2025–2027. In this comment of mine, I thought about the matter at the time. Anyway, this is certainly one of the most central things to watch regarding NVDA in the near future—I think we are at some kind of watershed moment:
PS: Cash flows turned negative for some of the companies, but that didn’t seem to bother the markets much, because at the same time we are getting ever stronger indications that AI investments are generating significant ROI. For example, AMZN’s Jassy said the following:
"On average, it takes a little less than 3 years to break even on that investment. The servers currently have a useful life of at least 5 to 6 years, and most of our AI capacity these days is being contracted for at least five-year terms. That means that we’re driving significant free cash flow on the servers and networking equipment in the two to three years after we break even.
We’ve long believed AWS could become a few hundred billion-dollar revenue business. We now believe it will be at least double that, and very possibly be a $1T annual revenue business for us in time".