Hacker News

points by gonzalohm 17 hours ago | 0 comments

I really don't understand the strategy of these companies. I don't care much about AI, but at some point, the current AI models will be able to run in old hardware. Hopefully that would leave room for other manufacturers to get into the market. I understand that right now it's really hard to compete because it requires top notch manufacturing processes. But that's not going to be the case forever.

By being hostile to personal consumers, they may make more profit now, but what's the long term plan?

I still run my 2060GTX and it's fine

[–] aeve890 12 hours ago

> By being hostile to personal consumers

What would be the problem? Do you think the day they come back to the consumer they would be given the cold shoulder? The consumer just consumes. They don't have memory or principles or loyalties.

[–] locknitpicker 15 hours ago

> I really don't understand the strategy of these companies.

I believe I do, and I think it's the most rational thing an established company could do. There are companies burning through billion of dollars a year on hardware to the point they are syphoning all production. If you are a business, what do you do? Do you redirect all your short-term output to these dominant consumers or do you ignore them altogether?

If you read the article, right at the top they state that AMD's data center business is expected to grow 50% in the next 3 months (?!) and double that growth, and is expected to double that as it enters 2027. Thanks to that strategy, AMD doubled it's profits this year, and 63% of that was thanks to data centers. This is free money to them. What would compell anyone to say no to that?

In a gold rush, the ones making money are the pickaxe salesmen. AMD is one of the biggest pickaxe makers in the AI gold rush. Only a fool would think if steering away from this market.

[–] EA-3167 15 hours ago

The people making the profit want to extend that period as long as they can, because it's coming to an end and because the money is terrific. When it comes to an end people at the top will parachute away and rebuilding the mess is someone else's problem.

[–] spottedmarley 17 hours ago

My local models are already quite good enough for 98% of what I would use an LLM for day to day. I feel like I already don't necessarily need a frontier model unless its for raw speed. This is still early days and, if I'm already feeling this way, in just a few years frontier AI will just be another big conglomerate that I avoid like the rest of them.