Could Deepseek v3’s Training Cost Under $6 Million Mean a Surge in Privately Developed AI Models in 2025?

Seeking Advice
1

OpenAI spent billions to train GPT-4, and Meta spent hundreds of millions on LLaMA. Now, Deepseek has open-sourced its comparable v3 AI, which was trained for less than $6 million, without using expensive H100 chips. The entire process took only several weeks to months.

This cost and time frame are accessible to many private individuals. Does this mark the shift from AI models being developed by big corporations to a new era where powerful AIs are rapidly created by private individuals all over the world?

2

Welcome to the forum!

Some Guidelines for Posting

Here are a few things to keep in mind for your posts:

  • Posts should be over 100 characters - the more detail, the better.
  • Make sure your question hasn’t already been answered. Use the search bar if you’re not getting responses.
    • AI replacing jobs? Yeah, it’s been discussed a lot.
  • It’s fine to discuss both the good and bad of AI, just be respectful.
  • Include links or sources to support your point.
  • No dumb questions unless it’s about AI ending the world. It’s not.
Thanks! Let the mods know if you need anything.
3

I think the ‘no moat’ memo still holds true. While big players are throwing huge amounts of money at AI, others are focusing on smarter, more cost-efficient models.

Models will keep getting smaller and smarter, which means we’ll see a huge boost in AI capabilities soon.

When someone figures out a breakthrough, it won’t be long before others catch up. It’s great for those of us watching the developments.

4

What does this mean for open-source, distributed, decentralized, and crowdfunded AI?

5

Did you mean SOTA or SOA?

6

Jesse said:
Did you mean SOTA or SOA?

Yeah, thanks for catching that! It’s the middle of the night for me, lol. I’ve updated it in the text, but I can’t change the title, lol.

7

Hopefully, we’ll see more open-source options with fewer restrictions.

8

Jin said:
Hopefully, we’ll see more open-source options with fewer restrictions.

I think it might lead to AIs with very specific purposes, almost like independent agents instead of just being tools for bigger projects.