[natroniks]

One point she makes is that these chips don't last forever and must be replaced every couple years at least. That there is a cycle. Nvidia is, imo, trying to replicate the iphione model, only with chips for enterprise rather than smart phones for consumers. For example, Nvidia went from a 2-year upgrade cycle to a 1 year, just like iPhones started as an every-other year upgrade cycle (with an S transitionary year) to annual upgrades. In fact, the iPhone gets at least a few new models each year, including regular, pro, and pro max. Nvidia is also trying to turn Cuda into the App Store of their platform. There's even the premium market parallel, ie, iPhone and Nvidia GPUs are the most expensive of their ilk. The way I conceive of Nvidia is, it's like the iPhone only on the enterprise side. Nvidia's sales will therefore be tied more to the business cycle than the consumer cycle; and this could mean more of a cyclical business that Apple ever achieved. However, it is also a much differenct business model than Cisco. It's almost like Apple's business model with Cisco's market segment.

[mgh2]

This is a good analogy, but not sure where the innovation and customer loyalty will come from.

[pjmlp]

Tooling, because CUDA has support for C, C++, Fortran, Python JIT, and anything else targeting PTX, graphical GPU debugger, and the accelerated libraries ecosystem.

It just needs to be better than what Intel and AMD have been delivering since OpenCL 1.0.