Just a reminder that we operate our newsletters on a paid model. Paid subscribers will get three newsletters a month including our China Deep Tech notes and audio recordings of our posts. Paid subscribers will also get early access to the newsletter as well other benefits coming soon. Please subscribe and support our work.
Highlights from Our Blog
We spent a lot of time recently going through Arm’s F-1 Prospectus filing for its upcoming IPO. Our first impressions is that we are underwhelmed. Revenue declined last year and despite growing revenue at a 16% CAGR since being acquired by Softbank, on average they still only earn $0.11 per chip shipped (chip, not core). As we dug more, we were intrigued by the company’s discussion of its end markets - 99% share in mobile, 10% in data center. But in the end we were left deeply ambivalent. The company is doing well, albeit a bit sleepy, but good companies do not always make for good stocks. The IPO looks positioned to extract maximum value for Softbank, which will leave a public Arm in a very bad position with the Street.
China blocked Intel’s acquisition of Tower Semi. This deal never made sense to us. Intel seemed to view Tower as a kernel for the future of Intel Foundry Services, despite massive differences in the scale, capabilities and culture of the two companies. Maybe China did Intel a favor.
If you like this content, you should check out our podcast The Circuit
Noteworthy Items
Semis and Hardware
Japanese employment program turned active industry consolidator Renesas acquired cellular modem maker Sequans for $249 million. Sequans had long overstayed its time as a stand-alone company, but a more interesting question is what is Renesas building. It is tempting to write this off as a misplaced strategy, but we think Renesas is making a credible bid to bulk up across IoT or the Edge or whatever name we are using today to describe the vast number of unconnected devices that are slowly getting attached to the Internet. Worth spending more time on Renesas. On a related note, Silicon Labs expanded its offerings of IoT SoCs, adding SDKs for Amazon’s Sidewalk protocol and a host of other features. IoT semis are messy, but both of these companies have coherent strategies for the space.
Macom acquired the RF business out of Wolfspeed. This looks like a good deal for all involved. Wolfspeed needs cash to build out the SiC capabilities on which it has bet its future. Macom gets a good business at a great price. Wolfspeed’s RF business is one more of the low-volume, high margin RF businesses that are prevalent across the whole RF industry. It did not fit with the rest of Wolfspeed, and could do well with a bit more attention from a company closer to this side of the business. There are many more of these opportunities out there if anyone is interested.
The Register reported that half of Arm’s data center installed base comes with AWS’s Graviton processor, with Ampere providing the bulk of the other half. Interpretation of this is left to the reader. Graviton seems to be doing fairly well, and the market is highly concentrated so no reason to expect a long tail of deployments. On the other hand, after roughly ten years of plugging away at this market, many would have like to see Arm show better traction.
Networking and Wireless
The GSMA, an organization of wireless operators and equipment vendors, unveiled a new set of APIs which offer access to a host of data on cell phone networks. Some of these are intended to ease particular user pain points (e.g. SIM swapping), others seem less user friendly. In particular a host of new location data feeds should raise everyone’s privacy hackles. We are not sure the world needs anyone to make it easier to locate specific phone users in real time.
Amazon received approval for a new fiber link across the Pacific. The hyperscalers are steadily taking over the world’s communications infrastructure.
We have been tracking the market for video game networking for many years. It is a big category that has ample room for improvement, but the dozen or so start-ups who have made a run for the opportunity have stumbled on the complexities of the market. Now larger companies seem to be waking up to the opportunity. Case in point, co-lo and interchange provider Equinix is entering the fray. It is somewhat of a natural fit for them, but they will likely face many of the same challenges.
Everything you wanted to know, and then some, about eSIM. There are some good reasons (and some less good ones) for why it has taken so long to move to this much more user friendly option for cellular communications.
Software and the Cloud
A few weeks ago we linked to a few stories about troubles at Red Hat and the relationship between IBM and the open source community. This former Red Hat employee provides a lot of background to the current situation. Multi-part, nuanced review of the history of enterprise Linux distributions, which concludes that the situation is not quite so black and white, but also that Red Hat did not have many good options. `
An academic study on improvements in GPU usage for AI, a deep technical dive into optimization techniques. This is dense but provides a good background for why GPUs are the preferred tool for neural networks. And a related primer showing why AMD will struggle to catch up with Nvidia (TL;DR - AMD’s software is still lacking).’
One more analysis showing the Cloud is not the solution to everyone’s compute needs.
A coherent argument in favor of building your own Large Language Model. We do not agree with all of this, but full points for the Terminator references.
The mathematics of training LLMs. As we said, this is not for everyone.
A handy chart showing market share of web hosting and CDN providers.
Diversions
Everyone agrees that Electric Vehicles (EV) are an important tool reducing humanity’s carbon footprint. Less thought is given to the source of that electricity. This map shows the sources of electricity around the world. It does not have data from EV leader China, so allow us to fill that in - coal.