The hot topic in semiconductor circles recently has all centered on the topic of slowdowns. There is a debate raging, especially among investors, as to how long the current demand trends can persist. We have argued that semis are cyclical and that there will eventually be a downturn, but the timing of that downturn remains highly unclear. There are clear signs of rising inventories across the supply chain, but many still struggle to source parts. This debate tends to focus on PCs – CPUs and GPUs, mobile – where all the vendors are showing increased inventories, and autos. Overlooked in all of this is the bedrock of enterprise electronics – the data center.

The data centers of the world are a massive consumer of electronics. Intel alone generates about 30% of its revenue from sales here, and depending on how you size it, this segment is 20%-30% of total semiconductor revenues, and likely a larger share of industry profitability. This category has been growing essentially non-stop for 20 years.

To be clear, there are no signs of that growth slowing any time soon, but we think it is worth considering all the things we tend to assume as givens so much that we forget about them. We found ourselves confronting this topic this week as we went through Amazon’s earnings and listened to Ben Thompson’s analysis of those numbers. Ben’s point is that Amazon, like everyone else, has been adding capacity during the pandemic on the assumption that 2020/21 growth rates would continue. Instead, we are seeing trends revert to 2019 levels. On their call, Amazon indicated that their data center spend for AWS would remain strong, while they had overspent on other areas of capex (i.e. warehouses).

Amazon spent close to $15 billion on capex in the latest quarter, about 40% of which goes to “Infrastructure” which is mostly AWS. They indicated that spending here would grow year on year, so no problems for data center spend… That being said, we re-read their CFO’s comments on capex several times and could not escape a tingling sense in the back of our brain. He did not say by how much it might grow, which left us wondering how much it would grow.

In 2021 AWS spent $61 billion on capex, 40% of which is about $24 billion. In the first quarter of 2022, they spent about $6 billion on AWS (40% of $15 billion). Which sounds fine, they just have to maintain Q1 levels to come in just above last year’s numbers. The trouble is last year’s capex increased strongly each quarter, with Q4 much higher than Q1, not an unusual pattern. But if you work out all these numbers it implies that capex will decline fairly sharply in the second half. For AWS capex to grow every quarter this year would require total capex 18% over last year’s figure. To hit that hypothetical 18% figure would mean a total increase of almost $4.5 billion over last year, and nothing on their last call gave the impression that any spending would increase at levels like those.

So there is a good chance that AWS capex will decline on an annual basis later in the year.

This is not unprecedented, nor is catastrophic for the semis industry, but is likely going to cause some pain for someone. Likely the worst hit will be Intel, who is already struggling for all the reasons we are already familiar with. And we will likely see trouble on other fronts as well – Facebook for one is probably not in a mood to blow out capex this year. If the other hyperscalers are in a similar position there may be some actual pain in this corner of the industry. Ali, Tencent, Baidu and JD.com are probably moderating data center growth (for a whole other set of reasons), the situation with Google, Microsoft and Apple are less clear, probably not as constrained. Offsetting all this gloom is the fact are some very strong secular trends – we are all spending more time interacting with Clouds, and there are no signs that any of these companies are curtailing data center expansion plans.

Summing this all up, we think there are reasons to be moderately concerned about data center purchasing trends later this year. No reason to be excessively concerned, but worth keeping in mind that we will not be able to take data center growth for granted forever.