There is a metaphor I hear occasionally to describe semiconductors – the idea of a bullwhip. A big part of running a successful chip company is accurately forecasting demand. Semiconductor manufacturing plants, fabs, are expensive to build. And even if you are paying someone else to produce your chip in their fab, you need to have a pretty solid handle of exactly how many chips you need built. Produce too many and you have a big pile of inventory that is depreciating at the same rate as Moore’s Law, i.e. very quickly. Produce too few, then when you run out of stock you are also out of luck because you need to find capacity at the fab, and those things need to be run at such high utilization that it is hard to get back in the queue.
But forecasting accurately is very hard to do. A big part of the problem is that chip companies are very far from end demand. There are about five levels between the chip vendor and the consumer – the end customer, the retailer, the distributor, the OEM and the manufacturer. For makers of smaller, lower-value chips, there are probably another two layers of companies that design circuit boards with many chips and some form of reference design or distributor. Each of those levels has to make their own forecasts, which implies at least five levels of error. By the time it gets to ordering the amount of chips you need built, all those errors compound. As a result semiconductor demand patterns can be very far out of sync of the rest of the product line or industry. Hence the analogy of the bull whip. The end-customer holds the handle, and the chip maker is way out on the tip of the whip oscillating wildly.
I mention this now because almost every chip company seems to be reporting pretty good numbers this quarter. There are some exceptions, but those are mostly companies with specific, well-known issues. As a whole, the chip industry seems to be moving in a positive direction.
However, I think there are some signs that the tech economy is slowing down a bit. I am not referring to a big slowdown like I discussed in my post yesterday, but rather a normal slow down in purchasing for many categories of electronic hardware.
First, there are some signs that all the non-Apple phones are lagging a bit. Some of this is just a factor of immense price pressure and competition for Android handsets. But the iPhone is doing well, and I think the others are seeing lackluster demand, divided among too many vendors. Other consumer devices are sending mixed signals. For instance, Fitbit just reported pretty solid numbers, but GoPro reported very weak numbers. Anecdotally, I hear that several other gadget/wearables companies are struggling a bit. So other than Apple, I think it is reasonable to assume that consumer electronics (CE) could be a bit soft right now.
Another warning sign is coming from the enterprise electronics market. By unit volume, this is much smaller than the CE market, but the chips they buy tend to be much higher priced. A number of notable hardware companies have guided to disappointing numbers for Q4. The two that come to mind are F5 Networks and Dell, among many others. F5 makes high-end networking equipment and their outlook for their year caused a fair amount of concern. Dell, which makes PCs and a whole bunch of other equipment for enterprise data centers does not report results, but when they announced their plan to acquire EMC, we got a glimpse of some of their numbers, and they were pretty modest.
Finally, another important data point is that in most categories of enterprise electronics the webscale companies have come to represent a very large portion of sales. From data center switches to server blades the handful of companies that operate massive data centers, like Google, Facebook, Amazon, Microsoft, etc. – represent somewhere between 30% and 50% of category revenue. This is an important shift in the industry, one which I plan to explore in greater detail soon. Whatever the exact level, these companies have become much more important to the whole industry. I think this has caused an unstable dynamic where one or two companies can meaningfully alter demand level for the industry. Any given quarter, at least one of those ten webscale companies is going to halt purchases to digest what they have already purchased or to iron out the opening of some new data center. And the odds are that every year or so, more than one of them will take a pause at the same time. We could even speculate that all of them have been building at such a breakneck pace for several years, and sooner or later, they will all take a pause on buying more hardware.
I think there is a real possibility that we are on the cusp of such a pause. Do not read too much into this. I am not forecasting the ‘end’ of the ‘Bubble’. But anecdotally, I do see some signs of a temporary slowdown in expansion of some of the smaller public cloud providers (e.g. Rackspace, HP). I am starting to think that the hardware industry is about to enter a normal, cyclical slowdown.
And this brings me back to the bullwhip. For all the chip companies reporting very strong results, I would be concerned that the good times may be coming to a slowdown soon.