Apple didn’t start their chip design journey with M1. They didn’t even start organically. Jobs bought two chip design firms and those people became the core of Apple’s chip design expertise.
Apple also didn’t design its server chips. It designs chips for their end products.
Tesla’s FSD chip in question is a server chip for training. It’s not an inference chip on Tesla cars. Also, Tesla’s R&D budget is actually very small compared to other car firms. Its 2022 R&D expense was 3B, vs 10B for GM and 8B for Ford.
Most of these AI startups will fail. The question is how much used equipment will flood the market? That’s what killed hardware companies during the dotcom bust. The demand for new hardware tanked. I wonder how many are buying chips vs renting them from public cloud providers.
Most bullish count is count $502.66 as wave 1.v.(1). If valid, should be in wave (3) which would be confirmed once above $502.66. NVDA is a hold (till AI megatrend reaches maturity).
To forecast Nvidia’s success, it’s therefore helpful to think about companies like Microsoft Corp. MSFT and Alphabet Inc.’s GOOGLGOOG Google that are buying up Nvidia’s AI gear and paying to access it through cloud services. Will these companies successfully develop next-generation software and services that will generate revenue gains for themselves, and more importantly, when will that happen?
What about the cloud service offer by NVDA for smaller (than megacap) companies who can’t afford the H100?
In a new era of technology, it is always hardware sales that lead, and other companies eventually develop software and services that take advantage of that hardware.
I’m not sold on their cloud service. Maybe it’s good for training models.
Models don’t run in silo. They run in combination with other services. It’s probably best to run models in AWS, GCP, or Azure where the other services run.
No one is asking about when peak training will occur. Training is driving the current demand. Will we hit a point the models are so accurate that there’s not much further training.required? If the training doesn’t turn into services that can be monetized, then there will be a huge drop in training.
I’m sure there’s more runway and growth potential. It’ll probably end up with an epic crash too.