Chinese company trained GPT-4 rival with just 2,000 GPUs — 01.ai spent $3M compared to OpenAI's $80M to $100M

Nvidia
(Image credit: Nvidia)

As Chinese entities do not have access to tens of thousands of advanced AI GPUs from companies like Nvidia, companies from this country must innovate to train their advanced AI models. Kai-Fu Lee, the founder and head of 01.ai, said this week that his company has trained one of its advanced AI models using 2,000 GPUs with just $3 million.

“The thing that shocks my friends in the Silicon Valley is not just our performance, but that we trained the model with only $3 million and GPT-4 was trained with $80 to $100 million,” said Kai-Fu Lee (via @tsarnick). “GPT-5 is rumored to be trained with about a billion dollars. […] We believe in scaling law, but when you do excellent detailed engineering, it is not the case. […] As a company in China, first, we have limited access to GPUs due to the U.S. regulations [and a valuation disadvantage compared to American AI companies].”

In contrast to competitors like OpenAI, which spent $80-100 million to train GPT-4 and reportedly up to $1 billion for GPT-5, 01.ai trained its high-performing model with just $3 million, according to Kai-Fu Lee. According to a company website chart, 01.ai's Yi-Lightning holds the sixth position in model performance measured by LMSIS at UC Berkeley.

TOPICS
Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

  • Parrhesia
    The problem is that China fakes everything, and even when they have something close to what they describe, anything IT related becomes an instant security risk.
    Reply
  • psykhon-
    And of course TH didn't verify any of this claims, otherwise they will have nothing to write about when finding all of this is nothing but the usual "Chinese" claims.

    Here, write another article about this: "one guy manage to build chatgpt7 using only 24 8 bits microcontrollers"
    Reply
  • zolarV
    Honestly, how is that even a feat if true? They're piggybacking on all the work OpenAI's scientists have done anyways.

    Its kind of like the 6th guy to make fire saying he is sooo much better because he did it faster than the first guy. When the 6th guy was able to watch guys 1-5 do it.
    Reply
  • NinoPino
    After writing "01.ai had to train its Yi-Lightning model using 2,000 undisclosed GPUs", it is a nonsense to write "The numbers, however, do not add up. One Nvidia H100 GPU costs around $30,000, so two thousands of such GPUs would cost $6 million"
    Reply
  • NinoPino
    Parrhesia said:
    ... China fakes everything, ...
    This is obviously a false affirmation.
    Reply
  • NinoPino
    zolarV said:
    Honestly, how is that even a feat if true?
    Because can demostrate a better way to do the same things. Obviously.

    zolarV said:
    They're piggybacking on all the work OpenAI's scientists have done anyways.
    And OpenAI is piggybacking all the research done by others previously. Every technological progress work like this.

    zolarV said:

    Its kind of like the 6th guy to make fire saying he is sooo much better because he did it faster than the first guy. When the 6th guy was able to watch guys 1-5 do it.
    Like all competitive sports...
    Reply
  • zolarV
    NinoPino said:
    Because can demostrate a better way to do the same things. Obviously.


    And OpenAI is piggybacking all the research done by others previously. Every technological progress work like this.


    Like all competitive sports...
    I see my entire point completely went over your head.
    Reply
  • JohnE777
    Kai Fu Lee is pretty reputable having an American PhD in Computer Science and leading teams at Microsoft, Google and Apple. I'd believe him.
    Reply