> I'd wager that Apple has been using AMD for something as mundane as offering better pricing, rather than disagreement 12 years ago. (Again: despite all the lawsuits, Apple is still a major Samsung customer.)

This used to be true, as Apple swapped between AMD and Nvidia chips several times in 2000-2015. Then Nvidia and Apple fell out, and Apple has not used Nvidia chips in new designs in 5 years - a timeframe in which Nvidia coincidentally achieved its largest technical advantages over AMD. Apple goes as far as to actively prevent Nvidia's own macOS eGPU driver from working on modern macOS. A simple pricing dispute does not appear to be a good explanation here.

Whatever the reason may be, the fact that Apple used Nvidia GPUs until 2015 at least debunks the endlessly repeated theory that it was because of the broken GPUs of 2008.

What hardware does apple use for all of its machine learning training (e.g. for Siri, etc. ) ?

What Apple is really worried about with NVIDIA is vendor lock-in. They own the walled garden, they absolutely are not going to accept an "external dependency" they can't control.

CUDA is such an "external dependency". It locks you in to something that's not an Apple product.

I never quite got how the Apple story was an indictment of NVIDIA. The first-gen RoHS-compliant solders sucked and would fracture... how exactly is that NVIDIA's fault? In fact, wouldn't Apple have been the ones who chose that particular solder?

It is the same issue that caused the Xbox 360 red-ring-of-death, and caused "baking your graphics card" to become a thing (including AMD cards). It basically affected everyone in the industry at the time, and Apple would not have gotten any different outcome from AMD had they been in the hotseat at the time. They were just throwing a tantrum because they're apple damn it and they can't have failures! Must be the supplier's fault.

That one has always struck me as a "bridezilla" story where Apple thinks they're big enough to push their problems onto their suppliers and NVIDIA said no.

And as far as the Xbox thing... Microsoft was demanding a discount and NVIDIA is free to say no. If they wanted a price break partway through the generation, it probably should have been negotiated in the purchase agreements in the first place. NVIDIA needs to turn a profit too and likely structured their financial expectations of the deal in a particular way based on the deal that was signed.

Those are always the go-to "OMG NVIDIA so terrible!" stories and neither of them really strike me as something where NVIDIA did anything particularly wrong.