D2D Highlights, March 27 2023

Qualcomm, Nvidia, Android and More

Mar 27, 2023

Highlights from Our Blog

We have been writing a lot about Qualcomm lately. This came about because of a series of investors conversations we have been having starting in Barcelona. These were all fairly negative, probably overly negative, but it still stood out for us.

The core thesis holds that Qualcomm has no path to easy growth. Many we spoke with hoped that the company would engage in some form of ‘transformative’ M&A. We do not think this is a viable path.

Instead, we think Qualcomm needs to rethink how it goes about sales. This is not a well developed muscle at the company. For instance, every product line tends to engage with customers on their own. This is sup-optimal, especially when probably the only way to win back Apple would be to provide a complete solution, bundling the modem, transceivers and RF. This could be meaningfully better than what Apple is working on, but Qualcomm struggles to sell products this way.

Qualcomm’s approach stands in sharp contrast to Nvidia, who held their GTC developer event this week. Nvidia’s entire presentation can be summed up with the word Partnership. Almost every one of their 37 press releases mentioned specific customers or a whole army of partners. Nvidia’s salesforce is not perfect, but with the industry moving towards custom compute, it certainly looks like the future.

Android is a mess, beset with fragmentation, security holes and privacy problems. As a result, it is losing share in almost every market to iOS. The best path to saving it is to make it truly open source, let the whole ecosystem contribute. Google has many reasons for not going down this path, but the arguments in its favor counter to mount.

The source of competitive advantage for many semis companies is shifting from having the hottest product to having the best strategy for cost accounting. Cost accounting can be the most boring topic in accounting, but it is importance for a world of semi-custom chips is immense.

Other News We Found Interesting

Wireless and Networking

Amazon unveiled the hardware for its Kuiper satellite network. Interesting devices, but also interesting to think about a world in which there are two massive satellite constellations operating largely beyond the control of any governments.

Ian Cutress wrote about how Intel has (finally) completely exited the cellular modem business. Those playing at home may have thought Intel exited this business when they sold most of the unit to Apple, but they still had a team making modems for laptops, who are all now victims of the company’s latest cost-cutting measures. Intel probably spent $20 billion (or more) over the years trying to find someway into this business.

There are 142,100 cell towers in the US, and 412,000 small cells. Interesting to note how this is a much smaller number than China, despite covering a larger geographic footprint. China’s cell coverage has to contend with much denser, urban deployments.

Wired waxes poetic about TSMC.

Semis and Hardware

The Financial Times wrote about Arm’s attempt to shift its licensing model to cover the cost of the device, not just a price per core. This appears to be an attempt to gin up Arm’s financial model ahead of its IPO, which is something we wrote a whole book warning against. Arm has needed to raise its prices for a long time, but this seems like the most challenging way to accomplish that. Ben Bajarin and I hash it out on our podcast.

Nvidia unveiled their CuLitho suite, software for improving the design of photomasks for semiconductor manufacturing. Of course it is powered by “AI”. An interesting look at how AI is moving into everything, but we could also just call this better software tools. It is also probably worth watching the Nvidia keynote section on this which does a good job of explaining just how complex lithography is and how the software is so hard to do conventionally.

Another look at infrastructure software for RISC V. This time a virtual machine, an important tool. RISC V software momentum continues to build.

Electronics conglomerate Toshiba has sold itself to private equity. Japan’s corporate model for electronics stopped working 20 years ago. All of these companies have some attractive businesses stranded within unworkable conglomerates. Toshiba’s problems were particularly acute, but maybe this will be the start of a process to unlock that value.

Cloud and Software

Qualcomm open sourced a set of audio encoders used for Bluetooth. Esoteric, but noteworthy because open source is not something Qualcomm has done much.