D2D: The Other AI-pocalypse
Wrestling with our AI zombies, Qualcomm's results and more
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. Paid subscribers will also get early access to the newsletter as well other benefits coming soon. Please subscribe and support our work.
We will be attending Mobile World Congress March 26 to March 29 in Barcelona. If you would like schedule a meeting in Barcelona click here.
Highlights from our Blog
Like everyone else in semis, we have spent a lot of time lately thinking about what “AI” means for the industry. A few things have stood out for us. Most of the use cases for transformer-based machine learning systems are still to be discovered, which is a polite way of saying there are not many right now. That being said, a lot of the gains will come in the form of boring, mundane, out-of-the-way small improvements in systems we are already using. There are a lot of these and their contribution alone is enough to merit continued exploration of the field.
At the same time, the general press is having a field day with Artificial General Intelligence (AGI), talking robots and Hal-2000. No one actually working in the field thinks AGI is in the cards right now. So instead of worrying about Skynet wiping out humanity, the semis industry should be much more worried about how - if - there is money to be made here, or if all of the gains go to Nvidia and a handful of others.
Qualcomm reported earnings last week - and the results were ok. This will be the year in which their long awaited automotive business will (should) translate into material revenue, and judging from the latest quarter that seems to be playing out. On the other hand, the company’s reliance on the saturated mobile market remains heavy.
Nvidia reports earnings in a week. We took a look at TSMC’s, AMD’s and Intel’s earnings to see if we could estimate guess at how those numbers will look. Our very rough math agrees with general market sentiment that Nvidia is in pretty good shape.
If you like this content, you should check out our podcast The Circuit
Semis, Hardware and Deep Tech
European man who sells equipment to manufacture semiconductors says the EU is not spending enough on equipment to manufacture semiconductors. The CEO of ASML states the obvious - the gap between the EU’s goal of 20% share of semis by 2030 and the amount they are spending to reach that goal is massive. Just because he has a heavy dose of self-interest does not mean he is wrong.
Google changed its accounting policies to extend the life of its servers, which “saved” them $3 billion. These sort of accounting changes are common, especially with companies trying to streamline operations in the face of Street pressure. But this is interesting when viewed in the context of the demand for more AI compute. Further evidence that big companies are scrambling to come up with the capital to invest in AI-capable data centers, likely adding to the size of the semis market.
A call for a US National Technology Assessment Center. We are fully in favor of the US government spending more money on technology analysts. But we argue that the US should go further and direct resources to increasing US technical capabilities and prototyping.
Nvidia’s Grace Hopper CPU system is competitive for building super computers. For years, high performance compute (HPC) systems like these was seen as a bastion for x86, but Arm based solutions like Nvidia’s have come a long way.
Who says innovation has ended in smartphones? HiSense just launched a phone with an e-ink display. This device is not for us (or we suspect many), but full credit to them for trying something new.
Purdue is offeing a free, online intro class on semiconductor fabrication. We have already signed up.
Networking and Wireless
There is a growing trend in the wireless industry to incorporate or at least accommodate satellite links into cellular communications. IEEE Spectrum casts its usual sober analysis on the prospects for this to happen as well as the still-serious remaining technical challenges. The industry has been trying to do this for years. In the 2000’s the rapid decline in the cost of cellular systems dampened enthusiasm for satellites. But now the sharp price declines have come to the satellite world which seems to be opening up the solution frontier. On a related note, Starlink is now handling 42 PB of data a day. Small when compared to global Internet traffic of 8 EB a day, but impressive for a very young system.
The networking industry has been talking about the transition from IPv4 to IPv6 for over a decade, and the drama seems to have finally reached the point when everyone else has to take it a bit more seriously. AWS has started adding a surcharge for the use of the dwindling pool of IPv4 addresses.This will likely be a nice earner for them for a while, but they would probably be just as happy if everyone just moved to IPv6 already. And another periodic reminder that AWS makes a lot of money simply from sending data out of its systems, so much that it makes a lot of people very angry.
An engineer’s detailed primer on networking. We like to bookmark these things because they always come in handy when we need to assess changes in the market from first principles. Networking has gotten fairly sleepy in recent years. Hopefully, someone will invent something new soon and we will have to pull out this link to understand its significance.
Software and the Cloud
An economist at a large investment bank is arguing that the value of software and coding schools is low and falling as evidenced by his own firm’s needs . Having worked at a large financial institution, we would counter that investment banks are not the best at understanding the value of software.
A software engineer re-discovers the perennial problem that software has a hardware dependency. And issues in one can constrain the other.
Diversions
The best story in tech this year - the Vesuvius challenge awarded its grand prize to researchers who read a full scroll page. Incredible technical achievement with more prizes on the way to unlock a lot more data. Fun times to be an AI researcher or a scholar of Classic Rome.
A very detailed look at the market for building 155mm artillery shells. For those watching the war in the Ukraine this is very interesting, but it is also a good case study in the complexities of designing national industrial policies. Getting it right for just this single process is taking years. Hard to see how well US policies can do with semis.
Image by Microsoft Co-Pilot
Thank you for reading D2D Newsletter. This post is public so feel free to share it.
Agree with the idea that small boring improvements is the best short term use of AI that will actually make money. Excited to see what those ideas are!