D2D: The Limits of Homegrown Silicon?
Apple's Vision Pro, Roll-Your-Own semis, and the latest from AMD
Become a Paid Member of the D2D Community
Just a reminder that we are shifting our newsletters to 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.
Highlights from our blog
Apple launched its Vision Pro VR headset. The device packs a lot of incredible technology and a price tag to match. We think Apple can readily build this into a billion dollar business next year, but we still have a lot of questions about how consumers will use this entirely new computing platform.
At the very least, Apple has raised the table stakes for anyone else making VR headsets. Anyone who wants to compete with Apple is going to have to match all that technology, and really the only way to do this is to build their own operating system and probably a few chips. Absent that, we think the market for this can go down two paths. VR headsets could end up looking like the smartphone market, where Apple sucks up all the profits amidst a sea of low-price Android competitors. Or the market will look like the tablet market where Apple gets all the profits AND all the units.
We attended the AMD analyst event. Despite the company announcing a range of solid products the Street was hoping for more. AMD is stuck between Intel which is fighting tooth and nail to hold on and Nvidia which seems unstoppable in AI. So while AMD is doing well, its valuation is likely to be determined by how much of the AI Inference market it can capture.
Chinese handset vendor Oppo largely disbanded its internal silicon design team. It seems that they were not getting the results they had hoped for from the effort, but this also shows the challenges and limits of homegrown silicon.
If you like this newsletter, you should listen to our podcast.
Noteworthy Items
Semis and Deep Tech
AWS is hosting a day-long event on their approach to semiconductor design on June 21. They have blocked a lot of time for this event, and we will be listening intently.
The next important trend in semiconductor architecture is Backside Power Delivery. IEEE takes a look at how Intel is approaching this, in what good be a really important advance if they get it right. And as usual, Doug O’Laughlin has some important and readable thoughts on the subject.
Jim Keller has seen it all from Apple to Tesla and many other companies. He is now a big proponent of RISC V with his company Tenstorrent promising some interesting advances.
A profoundly bearish take on the Vision Pro from Vice. We do not agree with much of this, but amidst all the excitement it is worth keeping other sides of the argument in mind.
CUDA is an incredibly important part of the Nvidia story. But its origins are oddly prosaic and have nothing to do with AI.
We have written a lot about Global Foundries’ future depending on its ability to encourage the large analog companies to forego building their own 300mm wafer fabs and letting GF be their foundry partner instead. So worth pointing out that GF just signed an agreement to do exactly that with ST Micro.
During the pandemic-induced gaming GPU shortage, websites emerged providing real time feeds into retailers’ inventory systems so you could get updates on who had GPUs in stock. We have now reached that stage for cloud AI GPU availability.
Hardware and Automotive
Ford’s CEO spoke on a recent podcast about the challenges legacy automakers encounter when transitioning to Electric Vehicles (EV). (Here’s the money quote on TikTok.) His point is that after years of de-centralizing their supply chains, Ford now needs to get its hundreds of partners to coordinate more closely when it comes to semis. Many will listen to this and be disheartened by the problems, but we were actually reassured that Ford is at least aware of the depth of the problem.
We like sensors, a big and rapidly changing market. As the market evolves, there is going to be a lot of debate about how many sensors things like cars need.
Nvidia seems unstoppable, but when it comes to telecom networking their solution does not seem to fit. This shows that AI does not matter to everyone in the same way.
Other Interesting Items
Using AI to create artistically appealing QR codes.