Comment by ares623

3 months ago

Selling the _entire_ stake sounds really aggressive though? Is that normal?

Even if you're all-in on OpenAI, does it not make financial sense to have _some_ stake in Nvidia considering they are the only ones with an actual moat?

Unless there are CUDA alternative breakthroughs we will hear about in the next few days.

18 comments

ares623

Reply
vintermann  3 months ago

For researchers and academics, cuda is painful to avoid, but I'm not sure that it is for large companies, once the time comes to train and deploy large models.

creer  3 months ago

> Selling the _entire_ stake sounds really aggressive though? Is that normal?

They did not. They sold MOST of their stake in Nvidia long ago before anyone paid attention to Nvidia vs AI. It's one decision as part of a complex cloud of decisions.

Also this "stake" was around one thousandth of Nvidia. Minor. Most companies have lots of owners with much larger stakes.

greycol  3 months ago

If you believe the AI bubble will pop in the same way as the dotcom bubble and feel you can pick winners it makes sense. It would be like selling cisco stock for google. You can still invest in cisco later if you believe they'll grow but you might expect a stagnant period and definitely expect a drop as they don't have as much demand. NVidia is a safe bet to continue exisitng and being profitable but by being so central to the market you could almost guarantee a large drop in stock price while other companies are a gamble.

l5870uoo9y  3 months ago

Isn't it merely a matter of time before China has an alternative?

  • chemmail  3 months ago

    All you have to look at is ASIC miners. Once they had them, they were 10x faster than GPUs easily and made GPUs useless for those algos. Something very similar can happen soon.

    • adastra22  3 months ago

      The fundamentals are different. Bitcoin mining is not intrinsically suited to acceleration on a GPU. It is a not-very-wide serial integer operation.

      AI inference on the other hand is basically just very large floating point tensor matrix multiplication. What does an ASIC for matmul look like? A GPU.

      8 replies →