NVIDIA is no longer a gaming graphics card company: NVIDIA is a ubiquitous company. That means it is the baby at the baptism, the bride at the wedding and the cement of the manufacturing industry. artificial intelligence. Your hardware is in the most powerful data centers on the planethis software controls everything and your money invest in any company that has something to say in AI. Big Tech (and everyone) is blindly trusting NVIDIA and has been given the keys to the house, but something is changing.
And now they want the keys back to regain control.
All the spotlights. Microsoft, Amazon, Google and Meta have bought hundreds of thousands of NVIDIA GPUs to shape your AI aspirations. At some point many began to develop their own hardware, but in the end NVIDIA’s was everywhere and was the one that gave the most guarantees, so they “gave up.” Apple, curiously, opted for Amazon.
And not just the big ones. OpenAI, Anthropic, Mistral or xAI are purely AI companies that They bet very heavily on NVIDIA from the beginning. Its hardware is the one that leads the way, the one that Western and Chinese companies want and the one that has such a brutal demand that it has elevated the company as the best client of TSMC and Samsung.
amd. But no one likes to have all their eggs in one basket, and those same names are moving. From a position of absolute dominance, in a short time we can move to another in which the hardware market is much more diversified. AMD is NVIDIA’s great historical rival in the PC gaming segment (and in consoles), but although they were out of the conversation for a few years, they have returned with force.
They have the hardware and are moving to get the same memory that NVIDIA has (and Samsung wins more than anyone else) and contracts as juicy as the one they achieved recently with Meta. The big rival also has deep pockets and is committed to taking a piece of the AI pie.
The Chinese threat. On the other side of the world we have China. We have said on numerous occasions that China is on to other things when we talk about AI. If the West pursues the AGI (with questionable claims like it’s already here), to China doesn’t care exactly. They want fast chips that allow them to create accessible and monetizable models in the short term.
But they also have Huawei, the company that has become the spearhead of the Chinese technology industry thanks to its collaboration with foundries such as SMIC is allowing, in an unthinkable way due to vetoes, can develop advanced chips. The development of cutting-edge chips still needs to be achieved, but Huawei already has more powerful inference chips than NVIDIA’s H20, according to them, and a supercluster for training.
Taking back control. Because in that term, “inference”, is where the current key is. AI training is important because it is what allows the model to then have the data and have a wardrobe to pull from, but inference is the final layer, which processes the user’s request to provide a response. There is not so much raw power needed, and that is what almost all the companies mentioned above are taking advantage of.
Amazon, Google or Meta have programs in which they are actively researching or developing chips proper for inference. OpenAI has signed an agreement with Broadcom to supply chips and xAI along with other companies Musk also has its own chips and they plan to open factories. And in China things are no different with Cambricon wanting to be a local alternative to NVIDIA and giants like Alibaba either ByteDance getting into chip design.
Groq. Given this, do you think NVIDIA is standing still? Among their hardware proposals, they have Groq, an inference accelerator that is designed for, next to Vera Rubinprocess a large amount of data at enormous speed. Groq was an unknown in the world of AI – until NVIDIA licensed it – and specialized from the beginning in that: chips with minimal latency for inference.
The key is in the architecture of its chips and it was a piece that was missing from the NVIDIA catalog and shows that, although the rest want the keys back, the one that already had them may have made a backup copy to continue being the reference. Because they may all be preparing their chips, but while they arrive, NVIDIA is already there and, in fact, with Groq it seeks to sneak into theThe $50 billion pie: China.
Problem for NVIDIA. But of course, that’s part of the story. The other is that NVIDIA also has all its eggs in one basket: that of AI. In the middle of last year we already mentioned that six customers represent 85% of all NVIDIA revenue in the previous quarter. It is an absolute nonsense that shows that, if there is a shift in technology, a puncture of the bubble or a new player that arrives strongly, the situation for NVIDIA may not be so favorable.
The question is whether a regime change can come and everything will be allowed to collapse like a house of cards. The uncomfortable thing is that an absurd amount of money is being invested and it’s not something that can escalate forever.
In Xataka | Jensen Huang believes we have reached the “coming of the AI wolf.” It is perfect for feeding a Tamagotchi
GIPHY App Key not set. Please check settings