From Moore's Law to Huang's Law, with a New Narrative of Nvidia’s History
Jensen Huang wants to tell the world that Nvidia's success has nothing to do with luck
Nvidia flips Apple as its stock hits the $3 Trillion market cap on Wednesday. It is the biggest winner in today's generative AI wave and Jensen Huang wants to tell the world that this has nothing to do with luck:
it is Nvidia's foresight and strength that created everything we see in the generative AI wave today.
What’s going on here
Nvidia has overtaken Apple as the world's second most valuable company with a market cap of over $3 trillion. And this happened after the keynote Jensen Huang gave at COMPUTEX ---- this is a keynote that can explain today's fever around Nvidia.
On the evening of June 2nd, Nvidia's founder and CEO Jensen Huang delivered the latest keynote speech at COMPUTEX. In addition to systematically and comprehensively reviewing and showcasing Nvidia's latest achievements in accelerated computing and generative AI, Huang also revealed the next-generation GPU architecture.
Yes, just three months after announcing the "most powerful ever" Blackwell, which hasn't even started mass production and shipping, the next generation is already on the way.
Huang said that in 2025, Nvidia will release Blackwell Ultra, followed by the next architecture called Rubin in 2026, and then Rubin Ultra in 2027. A major update every year - a pace more often seen in the smart smartphone industry. Moreover, these major updates will continue to exponentially reduce inference costs. This roadmap reveal is like Huang's law flexing its muscles. Everything is under control. You might as well just buy Nvidia's cards.
During the speech, when recounting Nvidia's history from GPUs to CUDA to the latest NIM, and then to robotics and digital twin platforms, Huang clearly changed his narrative style.
Huang described how Nvidia foresaw the bottleneck of CPUs early on, which is why it took the path of GPUs and accelerated computing. In the past, when mentioning the birth of Nvidia's GPU product line, he would always be humble, and the industry would often portray GPUs as chips that happened to be suitable for AI computing needs.
But today, Huang presented a different causality.
"CUDA was Nvidia's first intimate encounter with AI. After that, we deeply understood the essence of deep learning and consciously reinvented everything. We changed the GPU architecture, added Tensor Cores, invented NVLink, launched cuDNN, TensorRT, and Nickel, acquired Mellanox to launch Triton, and so on."
This is Nvidia's new narrative.
Among these, CUDA received high praise from Huang. "Without the carefully crafted domain-specific libraries we've built, global deep learning scientists would not be able to fully realize the potential. CUDA is like OpenGL for computer graphics and SQL for data analysis," he said. “We spent 20 years doing one domain library and one acceleration library at a time, and now we have 5 million developers." Huang no longer needs to be humble. "CUDA has reached a critical point and is beginning to realize a virtuous cycle."
In his speech, Huang didn't hide it anymore - he defined Nvidia as the source of today's generative AI wave in the world.
"The rise of AI was only possible because we believed that if we made powerful computing increasingly cheaper, someone would find a huge use for it," he said. Before Nvidia did this, "no one anticipated it, no one made this demand, and no one even understood the full potential."
Huang said that today we can generate tokens for anything of value, just like Tesla invented the alternating current motor, giving us a constant stream of electrons. Nvidia invented the AI generator, constantly producing tokens.
Everything is about tokens, and Nvidia created and mastered it all.
Key takeaways:
New architecture roadmap: 2025 is Blackwell Ultra, 2026 will have the new Rubin architecture, and 2027 will be Rubin Ultra.
We are not in an AI era, but a generative AI era. Almost everything in the world can be converted into Tokens (word elements).
The $3 trillion IT industry will become an AI factory, making AI products for every industry. We need AI based on and understanding physical laws.
Every PC with an RTX GPU is an AI PC.
CUDA has not only reached an important stage of maturity but has also entered a self-reinforcing virtuous cycle, continuously enhancing its performance and application value.
As CPU performance growth slows, using technologies like CUDA to accelerate computing tasks is a key strategy to address the exponential growth of computing demands. In the future, all compute-intensive applications and data centers will adopt this strategy to maintain efficiency and cost-effectiveness.
In the past 60 years, we have only witnessed two or three major technological revolutions. Today, generative AI allows us to witness another technological revolution.