AI has swept everything and has the potential to reshape every industry. NVIDIA is also the biggest beneficiary of AI development in recent years, and CEO Huang Renxun is a fanatical AI evangelist. In a recent interview,Huang Renxun also stated that computers in the future will be able to handle problems that are 1 billion times more powerful than today., but this is not the first time he has made this statement. He expressed this meaning at last year’s GTC conference.
This one billion times mainly refers to AI reasoning, which was also a major keyword in AI technology last year. Especially after reasoning introduced the thinking model, its importance has increased. The three laws of scale of thinking, post-training and pre-training are driving the development of AI.
The billion-fold scale of AI inference naturally requires stronger computing power. Whether it is launching Rubin CPX or acquiring Groq technology for US$20 billion, NVIDIA is preparing for super AI inference.
Although I don’t know when the billion-fold improvement mentioned by Huang Renxun will be realized, one thing is certain that with the rapid growth of the AI industry, its status has begun to replace mobile phone digital, at least in terms of TSMC’s advanced production capacity.
Everyone knows that Apple has been TSMC's largest customer over the past many years, because Apple has the largest orders for A-series processors used in iPhones and iPads, and most of TSMC's advanced processes are first launched by Apple every year.
After the advent of the AI era,NVIDIA has replaced Apple as TSMC's largest customer, and Huang Renxun also confirmed this, there are rumors a few days ago that TSMC CEO Wei Zhejia personally went to Apple headquarters to discuss price increases with Apple CEO Cook. His tone was already very tough, demanding that Apple accept the price increase and lose its priority in shipments.
TSMC's 2nm process was actually not first announced by Apple, but by AMD's 2nm process EPYC processor Venice. The next-generation A16 process is considered to be NVIDIA's next-generation GPU architecture Feynman. This is the first time NVIDIA has launched TSMC's latest process in more than 20 years. The last time was the 55nm process era.
At the same time, as Apple's position in TSMC declines, they are also looking for second-generation factories. Intel's 18A and future 14A processes may take on some of Apple's chip orders, especially M-series PC processors.