[smallmancontrov]

No moat.

Yes, CUDA, but CUDA is maaaaaybe a few tens of billion USD deep and a few (more) years wide. When the rest of the industry saw compute as a vanity market, that was sufficient. Now, it's a matter of time before margins go to, uhhh, less than 90%.

Does that make shorting a good idea? I wouldn't count on it. The market can always remain irrational longer than you can remain solvent.

[yen223]

I used to think that CUDA was something that would get commoditised real fast. How hard could building it be?

However, given that the nearest competitor AMD has basically given up on building a CUDA alternative, despite the fact that this could grow the company by literal trillions of dollars, I suspect the CUDA moat is much bigger than I give it credit for.

[tiahura]

And MS and everyone else have plenty of interest in helping AMD commodify CUDA compatibility.

[stefan_]

It's so weird it's taking them so long, because as far as anyone can tell AMD is mostly competent enough to make GPUs within some percentage points of Nvidia, the "breadth of complexity" in what these things do at the end of the day is ... rather underwhelming, the software stack may appear to be changing all the time but is also distinctly JavaScript-frotend-esque... is there an insider that knows what the holdup is? Is AMD just averse to making a ton of money?

At this point AMD investors should be rebelling, it's pissing money out there but they are not getting wet, and management might have doubled the stock price but that's little consolation if "order of magnitude" is what could have been.

[sangnoir]

> At this point AMD investors should be rebelling

Looking at the chart for $AMD over the past 5 years gives plenty od reasons to be happy, and no reason to rebel. A rational AMD investor should not be Jonesing Nvidia's catching lightning in a bottle via crypto + AI. The Transformers paper was published a few months before AMD released Zen 1 chips - they did not have a lot of money for GPU R&D then.

The timing of the LLM-craze was very fortuitous for Nvidia.

[treprinum]

AMD pays very little to its SWEngs (principal engineer in SFBA for ~200k), so they can't attract top end people in SW to implement what they need. Semi companies are used to pay HW engineers peanuts and that doesn't work in SW.

[wkat4242]

It's kinda great for those of us wanting GPUs though. Nvidia might eventually decide it's not worth their time to bother with.

[cma]

They also bought infiniband which has played a big role in being the best at clustering, though Google's TPU reconfigurable topology stuff seems really cool too.

Tesla went after them with Dojo and has still ended up splurging on big H100 clusters.