It seems very likely that most AI Training will be run on Nvidia GPUs. The software environments for training are too numerous and changing rapidly, which favors the soft, warm familiarity of CUDA over speculative gains offered by new entrants.
As we keep saying, IoT is not one thing. So maybe we should find a better way to segment the market. We think companies can analyst the opportunities available based on the complexity and engineering required to deliver their product to market.
Arm is now a real force in the data center. With all of the “Super 7” running Arm CPUs, and the software industry increasingly recognizing the benefit of porting to Arm, we are at or approaching a significant tipping point in this massive market.
AWS cannot control the software it runs, but it can still benefit from having its own CPU by reducing power consumption and thus increasing capacity of its data centers. And maybe that also introduces a new form of customer lock-in.
Optical stocks are volatile even in the best market conditions. Someday they will find their value, but it is going to be hard to determine when that will be.
The optical industry is plagued by manufacturing and packaging obstacles, but may finally be closing in the Holy Grail of Silicon Photonics. This time is totally different….
A recap of our emerging thesis on the changing nature of compute and what that means for semis companies large and small.
There probably will not be thousands of companies designing their own chips – the upfront costs are just too prohibitive – but there will be enough to mark a major shift for the industry.
The shift from general purpose silicon to semi-custom and tailored SoCs presents a big opportunity for chip start-ups. If they follow a few simple rules.
The large, incumbent chip companies are all choosing to embrace the trend of Roll-Your-Own chips by offering support services to non-chip companies’ efforts. Done well this may end up driving those customers to buy more catalog parts. Hopefully.