South Korea's Severe Stock Market Crash 🇰🇷💱📉

SK29

Well-known member
  • Apr 14, 2017
    6,999
    8,381
    113

    hesian99

    Well-known member
  • Jul 17, 2020
    3,410
    4,287
    113
    Gampaha
    Yes ... HyperScalers will win this AI game. Infrastructure and the application layers are where the real value is ( not the models itself ).


    NVIDIA has a chip monopoly. Thats why it has such a humongous Valuation. Even CHINESE Vendors can't replicate the chip magic NVIDIA does. Soon They will catch up + Elon's TerraFab ; Google new TPUs etc coming soon.
    Tpus are a different game betting future on tpus would be a huge mistake for google cuz the frontier has been going away from just hyper speed matrix multiplications

    nvidia does have a huge monopoly, just not on chips, now the chinese use their own chips for some of inference, they are not as optimized as nvidia but for inference it's not gonna be a huge issue, at least for them, nvidia has a big monopoly also on cuda and the whole ecosystem around their gpus which is far more interesting. they were saying for a lot of time china actually can replace asml's euv tech but they have repeatedly failed to do it, i dont think they will have a huge development in there anytime soon, even if they do cuda is irreplaceable for some more years

    but this is WRONG
    "Infrastructure and the application layers are where the real value is (not the models itself)"
    if that is true china wouldnt be much of a threat to us than they are now, with all dario tryna ban open weight and all those scenarios.

    scaling is working and it will continue to work in the future but scaling alone wont get us anywhere,
    breeding faster and faster horses doesnt mean you will get a locomotive
     

    Yeezus

    Well-known member
  • Nov 20, 2024
    4,690
    1
    6,532
    113
    if that is true china wouldnt be much of a threat to us than they are now, with all dario tryna ban open weight and all those scenarios.
    "Infrastructure and the application layers are where the real value is (not the models itself)"

    That's the exact reason Dario/Altman trying to get a ban to OpenWeight models use inside USA, Cuz their product is MODELS.

    Why KIMI K3 stopped registering new users recently. Cuz CHINA can't handle the compute it needs to serve such a demand that US Data centers do.

    Why NVIDIA; META ; GOOGLE are supporting OpenWeights - cuz they are hyoerdcalers. Their business models are infrastructure & application layers. They don't give a flying fuck about the models itself.

    Their stance is like - yeah we are all in for OPENWEIGHTS but you need cloud to host em. And we are here to serve you 👀 Thas INFRASTRUCTURE LAYER which I've mentioned above.

    Facebook is like - We don't care about delivering the latest model - We already have billions of users and our business model is DATA ; ADVERTISING and we will implement AI to streamline it. It's APPLICATION LAYER.

    China is catching upto compute through building new data centers where GOOGLE ; META ; MICROSOFT had already built for decades.

    Whatever the latest god level thinking model will US fronter releases - CHINA will distill it within days and either open source it or sell for a fraction of the cost ; where frontier charges and delivers 99% of its capabilities.

    The exact reason why ELONs trying to bring compute into SPACE - Spacial Computing cuz he knows the future is Infrastructure.

    That's why Ive said earlier
    HyperScalers will win this GAME ; not model vendors like Anthropic OpenAI.

    But both Closed Source and Open Source Models should exist. Did you see a recent news that an open source model blocked a cyber attack Claude and GPT failed to ask their "guidelines and cyber activity limits". That's why we need to both open and closed weights co-exist peacefully.
     
    Last edited:
    • Wow
    Reactions: Inigo Montoya

    Bitter Truth

    Well-known member
  • Dec 19, 2015
    23,932
    46,192
    113
    Village People Dancing GIF by de chinezen
     

    hesian99

    Well-known member
  • Jul 17, 2020
    3,410
    4,287
    113
    Gampaha
    That's the exact reason Dario/Altman trying to get a ban to OpenWeight models use inside USA, Cuz their product is MODELS.

    Why KIMI K3 stopped registering new users recently. Cuz CHINA can't handle the compute it needs to serve such a demand that US Data centers do.

    Why NVIDIA; META ; GOOGLE are supporting OpenWeights - cuz they are hyoerdcalers. Their business models are infrastructure & application layers. They don't give a flying fuck about the models itself.

    Their stance is like - yeah we are all in for OPENWEIGHTS but you need cloud to host em. And we are here to serve you 👀 Thas INFRASTRUCTURE LAYER which I've mentioned above.

    Facebook is like - We don't care about delivering the latest model - We already have billions of users and our business model is DATA ; ADVERTISING and we will implement AI to streamline it. It's APPLICATION LAYER.

    China is catching upto compute through building new data centers where GOOGLE ; META ; MICROSOFT had already built for decades.

    Whatever the latest god level thinking model will US fronter releases - CHINA will distill it within days and either open source it or sell for a fraction of the cost ; where frontier charges and delivers 99% of its capabilities.

    The exact reason why ELONs trying to bring compute into SPACE - Spacial Computing cuz he knows the future is Infrastructure.

    That's why Ive said earlier
    HyperScalers will win this GAME ; not model vendors like Anthropic OpenAI.

    But both Closed Source and Open Source Models should exist. Did you see a recent news that an open source model blocked a cyber attack Claude and GPT failed to ask their "guidelines and cyber activity limits". That's why we need to both open and closed weights co-exist peacefully.
    i dont actually think sam tried much to ban open weights but dario might, he is playing a very efficient game, he financed a bigger model recently, he is on tight budgets unlike openai, i think anthropic actually has a very little userbase compared to openai, revenue might be similar but yea

    on the nvidia thing, yes it might look like selling shovels in a gold rush might make the shovel guy pretty big, but the problem is this is intelligence what we are talking about, if tomorrow a company can make a model that is a few years ahead of the competition they will own the throne (id say the current gap is maybe a month cuz kimi caught up within a month after fable), better training better inference will always have its own advantages, these companies cant profit simply because of the competition, not anyone owns a monopoly now, everyone is fighting to their deaths, but there will be a winner probably and it will be bigger than anything

    this is also very wrong,
    "Whatever the latest god level thinking model the US frontier releases - CHINA will distill it within days and either open source it or sell for a fraction of the cost; where frontier charges and delivers 99% of its capabilities."

    first no one makes their models open source, no one, id be very happy if they did, open weight means dropping the model weights somewhere, its like getting the files of a software that we needed to run, not codebases, training data, rl pipelines or anything, so its just anyone could host it

    if that hyperscaler logic actually works then google should be on top of everything, they have literally everything, massive power plants, tpus, models, application layer to run it, but what has happened, they cant even release their dense model with the competition, but they are the ones who actually found most of the things llms are built on top of, if they hadnt done the attention paper we wouldnt even have gotten transformers, but what happened to them, i think they are ethically conflicted

    and nvidia itself is actually reinvesting a lot of money inside labs like openai, this money is going in loops cuz thats how it works
     

    hu KANNA

    Junior member
  • Dec 8, 2025
    66
    37
    18
    The exact reason why ELONs trying to bring compute into SPACE - Spacial Computing cuz he knows the future is Infrastructure.
    😂 ira pennala pol gahata gal gahanwa wage wadak oka. same as saying, sending a man to mars is the ultimate goal, to get funding for his profitable businesses. SPACE isnt COOL, it doesnt have enough molecules to contain heat, basically a vacuum, so theres not enough molecules to transfer heat. i saw a paper from singaporean scientists that you will need a 2Km x 2Km radiator just to cool down a 1GW data center. you cant jus send data centers to orbit and expect them to work.
     
    • Like
    Reactions: hesian99 and Yeezus

    hesian99

    Well-known member
  • Jul 17, 2020
    3,410
    4,287
    113
    Gampaha
    😂 ira pennala pol gahata gal gahanwa wage wadak oka. same as saying, sending a man to mars is the ultimate goal, to get funding for his profitable businesses. SPACE isnt COOL, it doesnt have enough molecules to contain heat, basically a vacuum, so theres not enough molecules to transfer heat. i saw a paper from singaporean scientists that you will need a 2Km x 2Km radiator just to cool down a 1GW data center. you cant jus send data centers to orbit and expect them to work.
    true i think machan his whole argument is like in space the solar is very efficient and has no peak time surge unlike on earth becux day and night changes or something but lot of things elon says take very long time to happen he sometimes even changes his plams
     
    Last edited:
    • Like
    Reactions: Naughtykolla48

    Yeezus

    Well-known member
  • Nov 20, 2024
    4,690
    1
    6,532
    113
    he is on tight budgets unlike openai, i think anthropic actually has a very little userbase compared to openai, revenue might be similar but yea
    Anthropic ARR is around $70B
    OpenAI ARR is around $40B

    Some speculates Anthropic IPO will be more than $3T valuation in the end of this year - It's the fastest growing tech company history have ever recorded. Every time a Chinese Models being released on par with Anthropic models - it's valuation drops & that's the exact reason Dario fear mongering government to get a band for them

    not anyone owns a monopoly now,
    There is ... Nearly 80% of this AI chip market is owned by NVIDIA. I call it a monopoly 👀
    first no one makes their models open source, no one, id be very happy if they did, open weight means dropping the model weights somewhere, its like getting the files of a software that we needed to run, not codebases, training data, rl pipelines or anything, so its just anyone could host it
    Exactly... It's easier to say open-source than weights that's why I referred that term.
    if that hyperscaler logic actually works then google should be on top of everything, they have literally everything, massive power plants, tpus, models, application layer to run it, but what has happened, they cant even release their dense model with the competition, but they are the ones who actually found most of the things llms are built on top of, if they hadnt done the attention paper we wouldnt even have gotten transformers, but what happened to them, i think they are ethically conflicted


    They are behind the frontier; Anthropic openAI were the early birds but they are catching up fast dude, Gemini 4 or 3.5 pro will be on par with frontier models ( hopefully ) They are winning the long term game. That plunge in Free Cash Flow is exact indicator of that - cuz their CapEx is doubling down.
    and nvidia itself is actually reinvesting a lot of money inside labs like openai, this money is going in loops cuz thats how it works
    Thread '🚀🤖 AI Bubble එකක් ද ? 🤖🚀' https://elakiri.com/threads/🚀🤖-ai-bubble-එකක්-ද-🤖🚀.2257837/
     

    hu KANNA

    Junior member
  • Dec 8, 2025
    66
    37
    18
    true i think machan his whole argument is like in space the solar is very efficient and has no peak time surge unlike on earth becux day and night changes or something but lot of things elon says take very long time to happen he sometimes even changes his plams
    "in space the solar is very efficient" thats just pointing out an advantage and ignoring all the foundational requirements to get to that point. main issue is that we cant dissipate heat. if humans magically synthesis a room temperature super conductor within next 50-100 years, sure space data centers will be a POSSIBILITY ignoring cosmic rays, space debris and funding.