AI Semis – the Have and Have Nots
Qualcomm has a robust AI offering, and the market for AI Inference at the Edge is huge. But Qualcomm and every other company not named Nvidia, will struggle to ride this wave to AI riches.
Qualcomm has a robust AI offering, and the market for AI Inference at the Edge is huge. But Qualcomm and every other company not named Nvidia, will struggle to ride this wave to AI riches.
How do CPU companies market in a newly, highly-fragmented market? There are no easy answers in this very complex, rapidly moving environment, but there is some hope.
There are really three markets for AI semis – training, cloud inference and edge inference. All of them are already fairly crowded. Choose your battles carefully.
We provide a basic look at why GPUs are the chip of choice for AI workloads, with a look at the early days of Nvidia CUDA.
Should we value Nvidia as a software stock? Its CUDA software is a major competitive advantage in AI. Even if we do this, it is hard to get comfortable with the current share price.
Less than a decade ago, CPUs were the dominant form of compute and everyone ‘knew’ the market could only support two vendors. Today, there are over a dozen companies making CPUs.
Every chip company has ambitions to sell software too – but there are very few software models that will work for them.
The data center silicon market is massive, but also challenging for incumbents let alone new entrants. We do some math to back this up.
Nvidia just had a developer event which highlighted all the partners they have, a vision of the verticalized future of semis. In contrast, Qualcomm’s sales structure does not look like that.
We think there will be a few more legs of semis consolidation. So we compiled a list of 5 semis companies who we think will survive:
1. ADI & TI
2. QCOM
3. NVDA
4. Some chip company from China
5. The smoldering remains of Intel, probably owned by others.