We thought this would be a slow news week….To start things off, Google launched their new Pixel 6 phone. As part of that launch they provided a deep dive into their new Tensor Chip, the applications processor (AP) that runs the phone. It should surprise no one that this chip features some advanced AI capabilities, but beyond that it still lags Apple’s A-Series. Ars Technica interviewed the Google team behind the chip and this provided a lot insight into how their chip is better, with good insight into the trade-offs in chip design. We wrote about Tensor a few weeks back when it was first announced, and we retain the skepticism expressed in that piece. Google has some very talented engineers, and they are building up some incredible internal expertise for building their own chips, but that does not make them a semis company. They are not capable of selling this chip to other handset vendors, and they likely do not even desire to do so. With Pixel units making up a tiny fraction of smartphone sales, Tensor is just another way for Google to demonstrate its AI chops.
Nonetheless, it is one more point of pressure on Qualcomm in the AP space, a product they invented and once dominated, but which they are now losing to smartphone makers’ internal efforts. Ben Thompson laid out that case well today in a piece titled “The Qualcomm Squeeze”. He is not wrong to highlight the pressure that Qualcomm is under in this key segment.
That being said, there was another big mobile semis story this week. Qualcomm unveiled its new RF chips. We have written a lot about Qualcomm’s dramatic change of fortunes in the RF space (and here and here). Put simply, Qualcomm has gone from almost a joke in RF to a powerhouse. They have had the capabilities to build RF products for over a decade, but it took them a long time to get the organization to the right place, and now that they have, they are firing on all cylinders.
In particular, they launched a new line of filters. RF modules on phones are a collection of parts: filters; power amplifiers (PA); and a bunch of other stuff (these things are pretty complicated). Since the launch of 4G, which brought a lot of new spectrum bands into play, we have been living in a Golden Age of RF content in phones, and nowhere is that more true than with filters. With phones tapping into all sorts of radio frequencies, and with many of those frequencies laying pretty close to each other filter requirements have gone way up. Leading filter makers like Broadcom, Murata and Qorvo (in that order) have done very well. But now, just like everyone else in RF, they have to contend with Qualcomm looking to eat their lunch.
In all fairness, we still do not know a lot about the new Qualcomm chips. Filters are very difficult to manufacture profitably, the manufacturing process is very challenging with very tight specs. So we do not yet know how well these product will do commercially, but our sense is that they would not have launched them if they had not already gotten these to a fairly mature stage. Note they are already sampling these to customers with volume production expected later next year.
All of this is bad news for the other filter makers, but it is also likely good news for companies like Rubicon who we suspect are providing the specialized SOI wafers for Qualcomm’s RF line and Global Foundries who has built a nice SOI foundry business.
But as we were reading this, we realized there are broader implications. For years, Qualcomm has had to use its core modem and AP business lines to “encourage” customers to buy its lagging RF products. But now the tables may be finally reversing, whereby the RF products can help pull customers to other Qualcomm parts. (We say finally because this was the original promise of Qualcomm entering the market ten plus years ago.) Qualcomm’s RF products may be able to not only expand Qualcomm’s Total Addressable Market (TAM), but may also start protecting the core product lines.
Qualcomm has a big advantage in the RF market. Because they make all the chips in the phone related to communication (i.e. everything but the screen and memory), they can tightly couple their RF products to the compute engine in the AP. This gives “intelligence” (or at least digital logic) to the RF chain which can have a big impact on the performance of the phone. Go back to that Ars article about Google’s Tensor, a lot of the same thinking applies here to the way that well designed compute can have an impact on other parts of the phone. In Qualcomm’s case, this can go even further. Using digital logic and a lot of tricky design work, Qualcomm’s integrated RF chain can actually reduce the filter requirement of a phone (among many other enhancements). So in the hypothetical case where Qualcomm’s new RF chips are maybe not quite as good as other companies’ filters, they can likely make up for the shortfall with this integration. This fully integrated RF Front End (RFFE) has long been the Holy Grail of RF companies, and Qualcomm is likely on the cusp of delivering that. They have already demonstrated one for fixed wireless access points, and if we had to guess we would say that Qualcomm is probably going to announce a fully mobile version at next year’s Mobile World Congress (if the show happens).
The importance of this extends beyond the RF space. If Qualcomm can really deliver on this (and by no means is it certain) then they will once again have a differentiated product, something that no one else has. This will be trouble for their arch-nemesis Mediatek whose market share this has overtaken Qualcomm for the first time. Mediatek has some RF assets, but lags considerably. But this is also important for the smartphone makers. All the leading smartphone makers have now been forced to build their own APs, with Apple going so far as to build its own modems, but none of them can make RF products. If Qualcomm’s RF products are as good as they could be, their customers will have to rethink their AP plans. They will turn to Qualcomm’s AP because it will be able to deliver far superior RF performance. In theory, Apple with a few hundred billion dollars in cash on its books could build its own RF chips, but it would take a long time to get there, certainly beyond the 2024 timeline when everyone thinks they will have their own modem. If they really wanted to avoid Qualcomm, Apple would have to either acquire or fully embrace one of the incumbents – financing construction of a whole new line. Possible, but not easy or guaranteed to succeed.
And so while we remain very worried about Qualcomm’s lagging AP line, it is very possible that their RF products could reverse their fortunes. As we noted above, we still do not have much technical detail about a lot of Qualcomm’s RF line, so nothing is set in stone yet, but their progress here paints the company in a much improved light.