Why Should Nvidia Start to Worry?

It's Only a Matter of Time Before It Loses the Title of World's Most Valuable Company

If we asked who the rising star of technology companies has been in the last 10 years, I think most answers would be Nvidia. OpenAI would also get many votes, but let’s not forget that OpenAI reached today’s success thanks to Nvidia chips. This rise gave Nvidia the title of the world’s most valuable company. But can it keep this position? It will be hard if CEO Jensen Huang can’t pull a rabbit out of a hat. Why?

Breaking Points

Before we get to the reason, let’s briefly look at how Nvidia got here. You’ve probably read the historical information in many places. Nvidia started as a company that made graphics cards for computers and became very successful at it. But we can talk about two big turning points in the company’s history. The first was when people discovered that Nvidia graphics cards were very efficient for crypto mining. The second important development was artificial intelligence and training large language models, which also need a lot of computing power.

During the pandemic, global supply chain problems made it very hard to get Nvidia graphics cards. People who really wanted these cards for gaming couldn’t find them because crypto miners were buying them all. This point can be seen as the peak for Nvidia in crypto mining. I think this was a situation that developed without much planning. People who rushed to buy these cards to mine cryptocurrency found more efficient ways to do it at this peak point. When people face difficulties, they look for new alternatives. Because of this, ASIC circuits were created for crypto verification, and Nvidia’s dominance in this area moved to ASIC circuits.

An Application-Specific Integrated Circuit (ASIC) is an integrated circuit (IC) specially designed for a specific task or application. Unlike FPGA cards that can be programmed after production to meet different conditions, ASIC designs are adapted to special needs in the early stages of the design process.

While Nvidia was hiding from its investors that most of its GPU income came from crypto mining (they went to court over this and paid a large fine to the SEC), it was also producing chips for growing AI companies. When OpenAI announced ChatGPT in November 2022, all the computing power behind it was Nvidia chips.

With ChatGPT’s huge success, all AI companies started knocking on Nvidia’s door to train large language models. Orders that no one could predict made this graphics card company the world’s most valuable company in a very short time. Nvidia still prefers to hide its Achilles heel. Looking at its last quarter results announced last week, 61% of its total revenue comes from four big customers whose names are not revealed. These are probably Amazon, Microsoft, Meta, and Google (Alphabet). Maybe Oracle could be in place of one of these. This concentration naturally creates a systemic risk.

Warning Signal

Google, probably one of Nvidia’s biggest customers, announced its new large language model Gemini 3 and related tools on November 20. The success of Gemini 3, which the whole world could access on the first day, was really amazing. Unlike OpenAI’s relatively less successful new versions recently, Gemini 3 really showed serious progress compared to its previous version.

The real bomb is hidden in the details. Google used its own designed TPUs to train this model. TPU, which stands for Tensor Processing Unit, is a processor type that Google and Broadcom designed together. Actually, this can also be called an ASIC. With these circuits that are designed only for model training by Google, brought to production by Broadcom, and of course manufactured by TSMC, Google is becoming able to offer a “full-stack” solution from beginning to end (from chip design to model development and cloud integration).

Also, looking at the shared data, they completed this training process almost 5 times cheaper than OpenAI, which uses Nvidia GPUs. This is a very important step in terms of efficiency.

Google has been investing in this area for almost 10 years. With Tensor processing units, it seems to be getting the return on this 10-year effort. Actually, like Google, other big AI companies are also working to produce their own chips. This includes OpenAI too. Nobody wants to depend on Nvidia forever. And as we understand from this first warning signal, they will be successful on this path.

After all, there are very powerful chip designer companies besides Nvidia. Like Broadcom and Qualcomm. At the end of the day, they all go and have these chips manufactured by companies like TSMC anyway. So Nvidia is also a fabless manufacturer like the others, meaning it has no factory. We know AMD also has serious work to have a say in this field. The quiet giant Intel, even though it might seem like an unimportant competitor, is still a manufacturer that should not be ignored with its production capacity.

Nvidia’s Dark Future

For Nvidia, it’s time to make a new plan. When we look at the distribution of Nvidia’s revenue items in the graph above, we see a company completely leaning on data center revenues. Also, the fact that four big customers whose names are not revealed make up 61% of its revenues is a very big risk. Even one or several of its customers switching to their own chips will cause a huge drop in revenues.

What’s Waiting for Us

Unfortunately, it’s not possible to see the future from today. But as Ilya Sutskever, one of OpenAI’s founders, also mentioned, we have reached the end of language models trained with more processing power and bigger datasets. The part we will focus on from now on is efficiency. That is, models that can be trained cheaper and process cheaper. Google seriously shook Nvidia’s throne with its new Tensor processing units. I expect other companies to announce their models trained with their own chips one after another. All of this could start an AI era where we can access it cheaper with competitive advantage.

(Since Nvidia, the subject of this article, is a publicly traded company, I want to state that what I write here is not investment advice.)

Which company do you think will be next to produce its own chip?

See you in the next article.