News DeepSeek might not be as disruptive as claimed, firm reportedly has 50,000 Nvidia GPUs and spent $1.6 billion on buildouts

[Pierce2623]

spam: DeepSeek owns 50,000 Nvidia GPUs, spent $1.6B—may not be as disruptive as claimed.

Yet now they can’t train v2 because they don’t have access to enough GPUs. I don’t think they ever had 50k H100/H800 or they wouldn’t be unable to train their new model.

 

[YSCCC]

I’m aware of their distillation method. They basically just write it to give the same reaction as ChatGPT/Llama etc through the minimal code base and training possible. I’ve got zero issues with that method UNLESS they’re using. whole chunks of open source code but there’s not been any evidence of that

The issue for me is that whenever it's utilizing distillation method, their cost isn't as low as what they initially claimed as ground breaking. they just used a short cut for part of their model, much like how one copies a top student's project, without doing all the initial work by themselves of course saved a ton of money and time needed.

Also their V3 instantly lost steam in even domestic market or utilization, it suffered one of the most AI hallucination out there right now, with thing like asking "where could we obtain coal" in chinese literally got answers of "go kill some skeleton" as in some popular RPG games.. the initial success came from distillation so they can quickly goes from a startup to catch up with the big players like Open AI, then the refinement and development lags and required similar massive AI GPUs to even go further.