{"id":60382,"date":"2025-12-03T14:13:10","date_gmt":"2025-12-03T13:13:10","guid":{"rendered":"https:\/\/mariuszbrandt.pl\/the-ai-chip-market-situation-is-currently-a-silicon-version-of-game-of-thrones-nvidia-sits-on-the-iron-throne-but-more-and-more-houses-want-to-dethrone-it-meaning-technology-influence-global-pol\/"},"modified":"2025-12-03T14:13:10","modified_gmt":"2025-12-03T13:13:10","slug":"the-ai-chip-market-situation-is-currently-a-silicon-version-of-game-of-thrones-nvidia-sits-on-the-iron-throne-but-more-and-more-houses-want-to-dethrone-it-meaning-technology-influence-global-pol","status":"publish","type":"post","link":"https:\/\/mariuszbrandt.pl\/en\/the-ai-chip-market-situation-is-currently-a-silicon-version-of-game-of-thrones-nvidia-sits-on-the-iron-throne-but-more-and-more-houses-want-to-dethrone-it-meaning-technology-influence-global-pol\/","title":{"rendered":"The AI Chip Market Situation Is Currently a Silicon Version of Game of Thrones. Nvidia Sits on the Iron Throne, but More and More Houses Want to Dethrone It. Meaning: Technology, Influence, Global Politics, and Poland’s Place in the World’s Technological Development!"},"content":{"rendered":"

TL;DR:<\/strong> This article discusses the key aspects of the AI chip market situation, which is currently a silicon version of Game of Thrones. Nvidia sits on the iron throne, but more and more houses want to dethrone it. Meaning: technology, influence, global politics, and Poland’s place in the world’s technological development!, emphasizing the importance of authenticity and modern strategies in building market advantage.<\/p>\n

\n

Be sure to read to the end! I promise it’s worth it! \ud83d\ude0e<\/p>\n

I got into a discussion with Gemini on a topic that is not very popular in the media, yet the true heroes of the world’s technological development are among us.<\/p>\n

Does NVIDIA even have competition?<\/strong> Yes, and it is getting fiercer, but for now, the “greens” (Nvidia’s corporate color) are winning not just by raw computing power, but by something called a “moat” in the industry.<\/p>\n

Here is the schedule so you know who is who in this sandbox:<\/p>\n

1. Why is Nvidia “rushing” so much in the first place?<\/strong><\/h3>\n

Before we get to the competition, you need to understand why H100 systems (so-called Hopper) or the latest Blackwell (B200) are in such high demand.<\/p>\n

    \n
  • It’s not just about the chip:<\/strong> It’s not like Nvidia simply has a faster processor. They sell the whole system. Their chips talk to each other incredibly fast thanks to NVLink<\/strong> technology. When you connect 10,000 cards, they act like one giant brain. The competition often chokes at such scale.<\/li>\n
  • CUDA (The Holy Grail):<\/strong> This is Nvidia’s software. AI developers have been writing code for this standard for over a decade. Switching to another company is like moving from Windows to Linux without a graphical interface \u2013 it can be done, but you need a good head on your shoulders and a lot of time. This keeps customers tied to Nvidia.<\/li>\n<\/ul>\n
    \n

    2. Who wants to kill the king? (The Competition)<\/strong><\/h3>\n

    We can divide Nvidia’s enemies into three camps:<\/p>\n

    A. The Eternal Rival: AMD (The Reds)<\/strong><\/h4>\n

    This is the only company making graphics processors (GPUs) that can compete in “muscle” with Nvidia.<\/p>\n

      \n
    • Their weapon:<\/strong> The Instinct MI300X<\/strong> system.<\/li>\n
    • Is it good?<\/strong> Hardware-wise? It is a monster. It has more memory and wider data transfer bandwidth than Nvidia’s flagship (H100). In theory, it is more powerful.<\/li>\n
    • The problem:<\/strong> Software. Their CUDA counterpart (called ROCm) is still less polished and more “annoying” to use. But companies buy AMD because Nvidia is incredibly expensive and perpetually out of stock.<\/li>\n<\/ul>\n

      B. “False Friends”: Google, Amazon, Microsoft, Meta<\/strong><\/h4>\n

      This is the most interesting group. These are Nvidia’s biggest customers who spend billions with them, but quietly… are building their own chips<\/strong>. They are tired of paying the “Nvidia tax” (Nvidia’s margins are astronomical).<\/p>\n

        \n
      • Google:<\/strong> Has its TPU<\/strong> (Tensor Processing Unit). They use them to train their models (like Gemini). You can’t buy them in a store; they are only available in the Google Cloud.<\/li>\n
      • Amazon (AWS):<\/strong> Has Trainium<\/strong> and Inferentia<\/strong> chips. They are cheaper to rent than those from Nvidia.<\/li>\n
      • Microsoft:<\/strong> Created the Maia<\/strong> chip.<\/li>\n
      • Why is this dangerous?<\/strong> Because these giants might say at some point: “Thanks Nvidia, but for 50% of our tasks, we’ll use our own blocks because it turns out 10x cheaper.”<\/li>\n<\/ul>\n

        C. Snipers (Startups): Groq, Cerebras<\/strong><\/h4>\n

        They don’t fight on general power, but attack specific weaknesses.<\/p>\n

          \n
        • Groq:<\/strong> Their chips (LPUs) are not used for training AI, but for its operation<\/em> (inference). They are absurdly fast. When you talk to a chatbot on their chip, the answer appears instantly, without that annoying “character-by-character typing.”<\/li>\n
        • Cerebras:<\/strong> Instead of cutting small processors from a silicon wafer, they make one giant processor the size of a dinner plate<\/strong>. This avoids delays in data transmission between small chips.<\/li>\n<\/ul>\n
          \n

          3. Which way is this going?<\/strong><\/h3>\n

          The war is now moving to two new fronts:<\/p>\n

            \n
          1. Inference, not Training:<\/strong> Until now, what mattered was model training (brain building) \u2013 here Nvidia rules absolutely. Now we are entering the phase of using models (inference). Here, such powerful machines are not as necessary; what matters is price and energy efficiency. Here, Nvidia is more vulnerable to competitor attacks.<\/li>\n
          2. The fight for links (Interconnects):<\/strong> Giants (AMD, Intel, Google, Microsoft) have formed the UALink<\/strong> alliance. They want to create a chip connection standard that is not owned by Nvidia. They want you to be able to connect different processors from different companies, rather than being forced to buy everything from the “Greens.”<\/li>\n<\/ol>\n

            Summary \u2013 Who Wins?<\/strong><\/h3>\n

            Nvidia still holds about 80-90% of the market<\/strong>. They are like Apple \u2013 you have expensive hardware, but everything works perfectly and the ecosystem is great.
            \nHowever, the biggest threat to Nvidia is not AMD, but the Big Three of Cloud (Google\/Amazon\/Microsoft)<\/strong>. If they successfully switch to their own chips, Nvidia will lose its best customers.<\/p>\n

            Processing centers in space!<\/strong><\/p>\n

            Nvidia is not building its own space station or “laboratory”<\/strong> in the sense of a physical building suspended in orbit. This is a media shortcut referring to Nvidia’s cooperation with startups (such as Starcloud<\/strong> or Lumen Orbit<\/strong>) that are actually sending its most powerful chips into space.<\/p>\n

            Here are the facts without marketing:<\/p>\n

            1. What is this project? (Facts vs. Myths)<\/strong><\/h3>\n

            It is not “Nvidia’s Lab,” but Nvidia is providing the “brains” for orbital server rooms<\/strong>.<\/p>\n

              \n
            • Key players:<\/strong> Startup Starcloud<\/strong> (backed by Nvidia) announced plans to put clusters with Nvidia H100<\/strong> chips (these are the most expensive and powerful ones) into orbit.<\/li>\n
            • Lumen Orbit:<\/strong> Another startup from the Nvidia Inception<\/em> program that raised funds to build space data centers.<\/li>\n
            • Goal:<\/strong> Testing so-called Space-Based Data Centers<\/em>. First tests involve sending smaller satellites to verify if the hardware can survive space radiation at all.<\/li>\n<\/ul>\n

              2. Why are they doing this? (Business logic)<\/strong><\/h3>\n

              This is not a whim, just math. Data centers on Earth have two massive problems that space solves “for free”:<\/p>\n

                \n
              • Cooling:<\/strong> On Earth, we spend a fortune and waste water cooling servers. In space, it is cold (in the shade) or heat can be easily dissipated through radiation.<\/li>\n
              • Energy:<\/strong> In orbit, the sun shines 24\/7 (if you choose the right orbit), without night or clouds. You have free, infinite power from solar panels.<\/li>\n<\/ul>\n

                3. How will this affect the competition?<\/strong><\/h3>\n

                If this works out, Nvidia gains a massive advantage in two areas:<\/h2>\n
                  \n
                • Running ahead:<\/strong> While the competition (AMD, Intel) fights for performance “per watt” on Earth, Nvidia positions itself as the sole hardware provider that is Space-Grade<\/em> (certified for operation in space).<\/li>\n
                • Data processing “at the source” (Edge Computing):<\/strong> Spy, weather, or communication satellites generate petabytes of data. Currently, they have to send this data to Earth for processing (which takes time and clogs channels). If Nvidia’s H100s were flying in orbit, the satellite would process photos on the spot and send only finished conclusions to Earth (e.g. “tanks are standing here”). This is a revolution for the military and analytics companies.<\/li>\n<\/ul>\n

                  4. Will there be an offer for retail (for you and me)?<\/strong><\/h3>\n

                  Absolutely not.<\/p>\n

                  You won’t buy a “graphics card from space” for gaming.<\/p>\n

                    \n
                  • This is a purely B2B (for businesses) and B2G (for governments\/military)<\/strong> service.<\/li>\n
                  • For the average Joe, the impact will be indirect: more accurate weather forecasts (thanks to the Earth-2 project, which is a digital twin of Earth simulated on these systems) or faster satellite internet that manages traffic more intelligently.<\/li>\n<\/ul>\n

                    Summary<\/strong><\/h3>\n

                    Nvidia is not building a Death Star. Nvidia is doing what it always does: selling spoons (chips) during a gold rush, only this time the mine is in orbit. It’s a risky move, but if space data centers become a reality, Nvidia will be there first, while the competition remains on Earth.<\/p>\n

                    This video material shows the backstage of the Starcloud startup, which, in cooperation with Nvidia, is implementing this wild plan of placing servers in orbit.<\/p>\n

                    Inside The Startup Launching AI Data Centers Into Space<\/p>\n

                    \n