Pretty much everyone.
We have been doing a lot of work lately on in-house silicon, most recently with our post on Google’s VCU. And it got us thinking about who else has a special purpose chip they really wish they had, and who might actually have the resources to build it.
As much as everyone (us included) is talking about in-house silicon, the reasons for building one and the ability to do so are not always there. Google building a TPU and then a VCU is impressive, two whole new chip categories from an Internet company, but Google is fairly unique in many ways. That being said, when you start adding up all the people who we know are working on chips and a few we think are as well, the list is fairly long.
In our view, the important thing that Google has done is demonstrate that a software team can design a chip. We have long maintained that silicon is just software made corporeal. Google is really the first company to turn that corner, and the industry and its tools have reached the point where they will not be the last.
So who else is building a chip?
First are the “hyperscalers” and cloud-service providers – Google and Amazon have announced theirs publicly. Amazon probably has a lot more going on, but that is a topic for a future post. Everyone knows Microsoft is working on something, although no one can really agree on what. Similarly, Facebook is designing chips for its Occulus AR/VR systems and a lot of people believe it may be working on something as well for its data centers. (Our guess is that they are letting the chip companies do most of the work, but cannot rule out the possibility that they may do their own chip eventually.) Alibaba is building some form of processor, and we would guess they are working on more. Baidu is working on AI chips for autonomous driving. Tencent does not appear to be building a chip, but they have a lot of interesting work underway around cloud gaming, a project that would likely benefit a lot from something like Google’s VCU. Although smaller in scale, we think it would make sense for some of the larger networking service providers, like Cloudflare to roll their own.
Next are the consumer electronic makers. Apple obviously tops the list. In addition to its massive memory chip business, Samsung has a mobile applications processor, and is probably building a CPU and/or GPU. Some of the other top handset makers, like BBK’s Oppo/Vivo and Xiaomi, are rumored to be working on apps processors as well. The PC makers do not appear to be going down this path, but the benefits for Dell, HP, Asus and the rest are less clear, and they really cannot compete with Apple’s M1 unless they get a lot more support from Microsoft – so maybe someday, but probably not.
On the enterprise electronics side companies like Cisco and Juniper have been making their own chips for years. Nokia and Ericsson largely exited the chip business a decade ago, and while they could come back, as far as we know they are not. The storage equipment makers have been very active – Hitachi, Fujitsu and Seagate, Western Digital are all working on some form of chip (and are big sponsors of the RISC V project).
Then come the industrial companies. The auto OEMs are somewhere on that list or at least they should be. They are all trying to figure out what to do around autonomous vehicles, but are also thinking about more powerful, special-purpose in-car processors that would come into use before autonomy arrives. These companies are actually really well-positioned to differentiate with silicon, but as the latest chip shortage has shown us none of them seem to have a sophisticated understanding of semiconductors. Next up, not surprisingly, are the aerospace companies – Rockwell and Thales are both RISC-V members, and we would imagine others, like Boeing, are at least considering it. Then there are are the industrial systems companies – Bosch and General Electric both have job listings for semiconductor designers. We can probably add Siemens to the list as well, although decoding their job listings is a bit trickier because they own EDA company Mentor Graphics (which in itself says something). There also seems to be work under way at some of the scientific instrument makers. To be clear, many of these companies are on our radar because we found they have job listings for “semiconductor designer”, and many of those roles are likely for simpler, analog chips which they have been designing for a long time, but equally many seem to be exploring something bigger.
Lastly, there are dozens of other Chinese hardware companies that are building their own chips. As with the industrial companies above, most of this is probably for simpler analog chips, or really chips embedded in modules, but China-based members of RISC-V are probably a majority.
Google had 52 people to design their VCU, so call that about $15 million in annual expense, add in another $50 million to actually bring the chip to production, and it should be clear this is something that a lot of companies can afford. This is not a project for every company, but we expect this list to grow soon.