David vs. Goliath in Semis
We also look at AMD's and Marvell's analyst events and AT&T's very odd deal with Ericsson.
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.
This will be our last note of the year. We will be back in January with a packed CES. We wish everybody a Happy New Year. Thank you for reading.
Highlights from our Blog
We just wrapped up a very busy, end-of-year cycle of analyst briefings. Two events in particular stood out for us.
Six months ago AMD announced their big Iron for AI, the Mi300, but the stock market was disappointed by the lack of a customer announcement and the shortage of performance details provided. Last week, AMD announced two big customers for the Mi300 and some competitive specs. And of course the stock fell on that news too. AMD is just steaming along, executing well. They do not need to beat Nvidia’s GPUs or Intel’s CPUs, they just need to keep rolling out competitive products.
Marvell’s analyst day left us with mixed feelings. On the one hand, they demonstrated the depth and breadth of their technical capabilities across a wide range of fields. On the other hand, it was hard for us to see how all those pieces hang together. Since publishing this piece we have had a lot of conversations with people about the company. We think they have a good story that needs to be told better.
Personal Area Networks (PANs) are like an idea out of science fiction. A constellation of personal devices we carry with us, that are connected to each other and to the Internet. Advances in data transmission and power management may finally allow bolder visions of PANs to become a reality. In particular, we are looking forward to the seeing Ixana’s demos at CES, which we think can open up a lot of doors.
AT&T announced it is shifting all of its Radio Access Network (RAN) gear to Ericsson, going so far as to rip out all the existing Nokia gear they currently operate. This is a dramatic, and frankly bizarre move. Not clear to us how Ericsson can earn a payback from what is likely to be an expensive deal for them. That being said, it puts them very much in the driver’s seat when it comes to “open” RANs, and is likely very bad news for the Open RAN initiative.
Could you start a new chip company today and grow it into a stand alone, public company? It is possible, but we think it is very hard with many obstacles and appealing off-ramps along the way.
Elliott Investments launched an activist campaign against tower owner Crown Castle. We are not big fans of activist investor campaigns, but Elliott does solid work. And we are not big fans of the customer experience at Crown Castle.
If you like this content, you should check out our podcast The Circuit
Noteworthy Items
Semis and Hardware
JD Powers says US automakers need to build smaller, cheaper electric vehicles (EVs). Or they could just lay off more EV employees.
Some of the most important AI Edge inference chips will be built by device makers rather than semis companies themselves. Expedera has some very useful IP for delivering on those capabilities.
We read very widely to learn about the industry. Case in point, this post on graphics drivers for Linux revealed a lot of interesting details about the complexities of building software ecosystems on top of semis. It also revealed that Nvidia has begun shipping RISC V cores in all of its GPUs.
Intel’s CEO say the “Entire industry is motivated to eliminate the CUDA market”. Nvidia has 16 billion arguments this quarter to the contrary.
High performance computing (HPC, aka super computers) are a tempting market for new chip vendors. HPC systems are highly technical buyers, with a much more approachable software stack than the hyperscalers. Which means these customers are much more open to buying solely on the technical merits of the chip in question. These researchers tested out a RISC V CPU (the Sophon SG 2042), but were unimpressed with the results.
Networking and Wireless
Chetan Sharma does some of the most insightful analysis on telecom and data networks. So when he publishes a report on one of the stickiest subjects out there, network opex, it is worth reading.
Telegeography published the latest version of their submarine cable map. Anyone looking for a last-minute gift for the networking heads in their life now has a new option.
This blogger likes to import weird vehicles from other parts of the world that are typically unavailable in the US, what does “street legal” even mean. He recently bought a container of electric construction equipment from China. Interesting in its own right, but also an important reminder that China is building desirable things that are far beyond what anyone else is building.
Canadian researchers have developed a light switch that runs wirelessly, powered by ambient power. This could eliminate 50% of home wiring costs, without the need to change batteries all the time. Some of the most interesting developments in hardware right now have nothing to do with semiconductors or AI.
One of the critical technical developments Qualcomm introduced to cement its rise in wireless system was the elimination of Intermediate Frequency (IF) chips in its architecture. Competitors’ IF components were rendered obsolete when Qualcomm incorporated them into their modems and transceivers. But mobile infrastructure still make us of IF, at least for now.
Some big developments in the use of AI in micro-controllers (MCUs) and the use of software systems like TinyML. This article is from Chinese analysts. Interesting read on its own, but noteworthy that these developments got little attention in the US. We do not seem to care much about MCUs here, while in China they are a big topic.
Software and the Cloud
A visualization of large language models (LLMs). We are not sure what we learned from this, but it is fun to play with.
Google released a set of low-level features for Android which let developers run virtual machines on phones. Important for security, but a clear demonstration of why developing operating systems (OS) is so challenging. They also posted on their efforts to make cellular modems more secure. It’s Google, so we have to wonder if these two teams have ever spoken to each other. Also, our phones are nowhere near secure enough.
Google unveiled its Gemini system, their answer to ChatGPT. This has been widely covered in the press, which pointed out the nebulous release dates and some questionable responses. Most interesting to us is the interplay of the Gemini and Google’s homegrown TPU silicon.
Diversions
The highest paying tech companies for software engineers. We are not saying you should read this list, just that you could.
Last decade there were many stories about CIA assets being uprooted in China and Iran due to a systematic communications failure which allowed those countries to run through the CIA’s networks. Now we know more about how that worked and what systems the CIA was using then.
A nuclear reactor simulator. More fun than Fortnite.
We have been working to raise a venture fund through some of the toughest fundraising environments anyone can remember. A big part of the problem is that many fund investors (LPs) placed big bets on software growth funds in 2020 and 2021. Those turned sour, but the LPs have been stuck. On paper they are over-allocated to venture because those VCs have not marked down their portfolios yet. In reality LPs are probably underweight venture, and they are heavily underweight Deep Tech venture. So when one of the biggest software growth funds (finally) marked down their portfolio, we see a ray of hope. We are also encouraged by an academic study that says physically unattractive fund managers outperform attractive ones.
Photo by Steve Barker on Unsplash
Thank you for reading D2D Newsletter. This post is public so feel free to share it.
There’s a lot here, as always. Commenting just on the Open RAN piece. The rip and replace element of the Ericsson deal with AT&T is a big surprise. As you point out, Nokia rolled up Lucent so maybe there’s some Western Electric branded equipment in there somewhere.
But also wondering if Ericsson provided a better answer to RU (RRH) TCO than other Open RAN suppliers have. That’s been the sticking point so far and as you know was at the heart of what Ericsson proposed to the O-RAN Coalition. And yes, it will ostensibly be Made in America.