Comment by Festro
9 hours ago
There is an AI bubble narrative, people are curious when it will burst. This is an indicator that people will want to analyse, discuss, and think about (intellectually).
9 hours ago
There is an AI bubble narrative, people are curious when it will burst. This is an indicator that people will want to analyse, discuss, and think about (intellectually).
I see. I guess I simply think that Softbank isn't a good indicator for that. To me, they don't seem better than any other investment company that puts their money into technology stocks with a growth narrative. For example, they invested in WeWork and FTX.
They don't know the future, just like the rest of us.
If we're talking indicators, if 5 Softbanks would do it in rapid succession one after another it just mentions to me that the "smart money" is showing signs of moving out.
In terms of whether AI will or will not fuel growth, I think it will fuel growth. Self-driving cars seems to be a solved problem for cities at least fairly soon (e.g. Waymo, anti-example: Tesla, camera's is not the way).
It's a question whether LLMs state of the art models will grow more, but what hasn't been done that well yet is integrating it into current software. I know, because in part, that's my job. There's still a huge productivity unlock there, also in ways that people can't fully imagine.
Right now, LLMs seem to be an enabler for software engineers, especially software engineers on smaller projects (I can't find the research at the moment, it was a while ago that I read it). It seems to be an enabler for many people, but they do need to put time into prompting it in a way that works for them.
Fixing the context window issues and others I think will be really hard tasks, because I suspect we then need to know what goes on inside the black box.
If an LLM could continuously learn, so somehow continuously keep updating its weights such that it learns better, that would be a breakthrough.
I agree to be fair. I see Softbanks move as just wise strategy. Nvidia's growth won't be uncapped.
I wouldn't be surprised if they reinvest some of that money into AMD or similar, as those companies play catch up. AMD is bringing ROCM up to speed with AI models, and their consumer hardware is having pretty good press despite some PR fumbles.
Selling Nvidia stock isn't a sign of lack of faith, it's just an exit after strong results. Softbank think this is a peak or near enough, good for them.
As you say, until there's some breakthrough, there's little point keeping money in the dominant company.
Also, Nvidia appear to have delayed their 60-series GPUs, and this will cascade to their entreprise models, due to the 3GB module memory shortage. The next gen will be VRAM heavy, but you can't do that during a shortage, you need the market to correct tor for supply to increase. So any 'breakthrough' powered by new hardware is now about a year away.
This won't stop people secondguessing on the bubble front though. Personally, I think bubble popping is going to come from a lack of faith from investors in the downstream companies like OpenAI who are struggling to make money from their resource intensive products. Nvidia are already profitable with the hardware, MS/Google/Amazon will always make money with the servers. And if AI bursts, other sectors can soften the blow for those companies. It's the massively inflated AI model makers that need to worry about who makes it out of this profitably.