AI is appearing everywhere, but at heart it is just a highly efficient solution to a specific set of computer science problems – this is sparking a new class of opportunity for chip companies and IP providers to tap into the opportunity.
Google has a new chip – the TiN – for networking. An interesting chip in its own right but we also see it as a sign that Google is creating tools that let any software coder design a chip. And if Google can do it, maybe everyone else will too. Someday.
Software company Red Panda did an in-depth analysis of Graviton’s Arm-based server CPU, and demonstrates that Arm is starting to look very strong in the data center.
The cloud service providers – AWS, GCP and Azure – have to support the software of all their customers, this complicates the case for them building their own chips.
Meta’s struggles with designing their own silicon for their AR glasses shows that rolling your own silicon does not make sense for all companies.
Part of the magic of semis is the ability to integrate multiple chips into a single chip. History shows, that the vendor whose chip sits closest to a system’s critical software wins the strategic high ground. What will that mean in autos?
Nvidia’s Analyst Day demonstrated the company is now the leading force in the data center. They have risen to this crest on the back of some incredible execution, and their rise shows the very powerful wave washing through the market for compute semis。
RISV V’s leading proponent SiFive has some great momentum, but it also faces many obstacles ahead.
We believe, with increasing conviction, that the market for IoT chips is not going to go to Arm. But it is not going to go to x86 or Intel either. It is going to go RISC V.
Google recently published a paper on the history of its TPU chip. There are some valuable nuggets of information in here that can help others think about building chips, and also help outsiders understand many of the changes in the semis industry today.