Ostensibly, this post looks at the market for semis for wireless networking gear, but we readily admit it is a bit of a rant.
A couple weeks back Light Reading published an article about silicon in the mobile radio access network (RAN). The article, “Intel risks losing Arm wrestle as open RAN splits into rival camps,” is a good read but it touches on so many important topics that we wanted to peel apart some of the layers.
First, some basics. The RAN is the system that provides “the last mile” connecting mobile devices to the operators’ network and the Internet. The RAN is the critical battleground for mobile networks because this is what makes phones mobile. Historically, RAN equipment has been dominated by the wireless equipment makers like Nokia, Ericsson and Huawei. They make the base stations, antennas and associated switches which then plug into the operators’ core networks.
Over the past years, other companies have attempted to open this market up by creating the Open RAN initiative (ORAN). The goal of ORAN is to make this layer of equipment more modular. The big vision is to convert all the special equipment in the network into commodity switches and servers. The incumbents, do not like this vision, to put it mildly, and so they are promoting their own vision for ORAN which opens up some of the network interfaces to more module hardware, but in which they still control the key control points (aka the software). There is an analogy to be made here, the incumbents are proposing a “walled garden” or “cathedral” for networking software, while the ORAN maximalists want something closer to an “open bazaar” which is wide open, the perennial dance between bundles and best-in-breed point solutions.
Intel has been working hard for close to a decade to promote ORAN. At heart Intel favors the more open flavor of ORAN, on the assumption that all that commodity gear will run on Intel CPUs. They have invested heavily in building ORAN, in a pattern that is reminiscent of the work they did promoting Linux and supporting that open source ecosystem for years, and upon which they built their dominance in the data center. Mobile networks are dependent on the standards bodies who define how 5G and the RAN operate. A few years back, we wrote about Intel’s heavy activity in the key 3GPP standards group (and a follow up here). But the mobile market is different than the far more open market for data center software. And this gives the incumbents considerable leverage. Intel recognized this and sought to partner with Ericsson among others to deliver a modified x86 CPU solution.
And now the time has finally come for Intel to reap the rewards of all those years of work?
Not quite. ORAN is an important topic this year. A handful of operators have begun implementing some version of it, and there seems to be considerable momentum behind the project. But suddenly, Intel finds it may be left out in the cold. Marvell, Qualcomm and others have entered the market, and they are perfectly happy to build truly custom, purpose built chips for anyone buying.
The article digs into the technical debates around stand-alone systems versus the addition of accelerator cards. We will come back to this debate after Mobile World Congress in February, but there is a bigger theme emerging in the ORAN silicon. This is a multi-sided conflict, with erstwhile partners like Ericsson and Intel agreed on some topcs, but not enough so that Ericsson is not also pursuing other silicon options. On the one end, there are Nokia and Ericsson looking to maintain their key strategic territory around the networking software and interfaces, then there is a separate fight taking place among the chip vendors looking to supply any and all of the companies fighting on the other side.
In all of this, it looks a lot like Intel may have tripped itself up over its attachment to ‘openness’ and general purpose compute. Maybe this will be the winning approach, but Intel has a bit of history here, and it is not pretty. Intel tried for decades to enter the mobile market. They spent close to $20 billion on various attempts to make smartphone chips, only to sell their smoldering ashes to Apple (who still can’t quite get it to work). Intel has long favored an “open” approach, witness their years promoting the WIMAX standard, which failed by overlooking the operators entirely.
The CTO of Intel’s current mobile group, Sachin Katti’s, remarks on the subject demonstrate Intel’s view that general purpose compute (aka x86 CPUs) will eventually come to dominate..
Yet Intel has never sounded so bullish. “General-purpose technology will eventually have so much investment in it that it will outpace custom silicon,” said Katti.
This problematic for a number of reasons, not least Intel’s biggest customers do not agree, and have a heavy vested interest in seeing that the whole RAN remains as custom or proprietary as possible. But Katti does not stop there.
“Every chip is a pretty large investment of nine figures, and the RAN market is not big enough for someone to spend that amount every year keeping up with the process node improvements,” Katti said.
The RAN market is $40 billion a year, so probably largely enough to support ASIC development. But it is really the second half of that statement that boiled our blood. Katti’s position is that Intel’s current chip performance may not match that of its competitors, but Moore’s Law means that the next process node will give them the lead. This is problematic because: 1) every other competitor has access to Moore’s Law; and 2) Intel is the one most limited when it comes to those process node improvements. This whole line of reasoning ignores the reality of Intel’s position so baldly, that it calls into question all of Intel’s plans for the space. We admit to being somewhat jaded by this argument, because for the better part of ten years, we heard Intel make the exact same argument about the performance of their mobile baseband, that was perennially just one node away from dominating the market. The argument did not work then, and it certainly does not work now that Intel’s next node is still so far away.
All in all, we think this is going to be one of the most interesting markets to watch in the New Year. This remains a sizable market, and seems ripe for some major shifts in competitive dynamics (looking at your Marvell). Telecom equipment looks like a prime candidate for homegrown silicon, add in the unfolding drama around the ultimate fate of ORAN, a host of operator drama and we get a the kind of battle which have been missing from mobile for too long.