[h14h]

Fair points all around. Ultimately it all comes down to execution.

In theory, Apple SHOULD have an advantage given they have everything they need in house and can all pull in a unified direction. In practice, it's not always the case that all the teams in a large corporation are all that much better at pulling in the same direction than multiple different corporations in a partnership. And all this will be moot if Local LLMs never catch up to cloud LLMs in terms of quality.

Regardless, it'll be very interesting to see how Nvidia's partnerships with Microsoft & hardware OEMs play out. If the AI inference compute share shifts appreciably to local consumer hardware, I'll want to see strong competition.

[bigyabai]

I'd argue that Apple had the upper hand, but they folded super early. They abandoned OpenCL, which was the most promising CUDA competitor with industry-wide buy in from dozens of companies. Then they transitioned to an ecosystem-first mindset prevented Apple from cooperating to take down Nvidia, and their locked-down software stopped the industry's first high-speed ARM servers from reaching their audience. Nvidia capitalized on both opportunities to the tune of trillions in valuation.

Without Khronos involved, I don't think that Apple has the buy-in to create a real industry-scale CUDA alternative. At this point, it might just be most profitable to support CUDA in macOS and give the people what they want.