By hmimcu 
Calculate everything.
“I thought that the monthly salary of 140000 yuan for ByteDance had reached the top. I didn’t know what a part-time worker’s ceiling was until I saw the tax records of Nvidia employees!”
No wonder some netizens are so impressed. A screenshot of a tax record recently released shows that a Nvidia (Shanghai) employee’s total income in 2021 reached nearly 11.22 million yuan, and the declared tax amount also reached 4.56 million yuan (not representing personal tax payment). Compared to the current job market, it’s hard not to envy.
WeChat image_ 20230420191817.jpg
Of course, behind the huge income of NVIDIA employees, it is essentially due to NVIDIA’s impressive financial results back then. According to NVIDIA’s 2022 fiscal year report (as of January 30, 2022), the company’s annual revenue reached a historical record of $26.91 billion, an increase of 61% year-on-year, and it also paid $399 million in cash dividends.
Let’s take a look at the global technology company’s rush to purchase Nvidia GPUs after the ChatGPT fire. Microsoft, Google, Alibaba, Tencent, and Baidu are all purchasing a large number of Nvidia’s high-end computing cards to cope with the exploding computing power demand. Amazon, which recently announced the AI model war, immediately purchased tens of thousands of Zhang Yingwei’s H100 GPUs. Musk recently stated that Tesla will continue to purchase a large number of NVIDIA GPUs, and last month he only bought about 10000, even jokingly saying:
It seems that everyone and their dogs are buying GPUs
h100-og.jpg
Nvidia H100, image/Nvidia
Although Nvidia’s latest quarterly financial report will not be released until May, the outside world is already speculating on how much Nvidia can win this time. In the past few years, in addition to continuing to dominate in gaming graphics, NVIDIA has taken on another trend: cryptocurrency and mining, smart cars, metaverse, large (language) models, and generative AI.
Why does Nvidia always grasp the wealth code and always win?
From mining to generative AI, the eternal king of computing power
For the general public, NVIDIA is most well-known for its gaming chip business, but in the past few years, NVIDIA has been increasingly recognized in a different guise.
In 2017, the prices of cryptocurrencies were constantly skyrocketing, with speculators constantly entering, and there was also a large army of “miners” holding GPUs. In order to accelerate the speed of mining, miners purchased a large amount of computing power and graphics cards, continuously driving up the popularity and prices of the graphics card market, and even causing collective dissatisfaction among game players.
But no matter what, Nvidia has made a lot of money. Huang Renxun said, “Our GPU supports the world’s largest distributed supercomputing, which is why it is highly popular in the cryptocurrency field.” In addition, NVIDIA has launched GTX 1060 3GB and P106 and P104 professional mining cards specifically designed for “mining” customization.
Benefiting from the “mining boom”, Nvidia’s revenue for the fiscal year 2018 reached a new high of $9.7 billion, with game chip revenue including mining cards reaching $5.5 billion, accounting for over 50%. Nvidia’s market value is also soaring, from $14 billion in 2016 to a high of $175 billion in 2018, more than tenfold in two years.
However, with the merger of Ethereum in 2022 as a symbol, the mining era has finally come to an end. When ChatGPT caught fire, NVIDIA CTO Michael Kagan also began to conclude that cryptocurrencies are useless for society, and GPUs and computing power should be used to develop AI that is more beneficial to society, such as ChatGPT.
For NVIDIA, computing power is everything.
A consensus in the automotive industry is that the next stage of electric vehicles will inevitably be intelligent vehicles. No matter Tesla, Huawei, Weilai, Xiaopeng and Ideals regard automatic driving as the core of intelligent cars. Among the three keys of autonomous vehicle – sensors, chips and algorithms, chips need to provide extremely high computing power.
This is also where NVIDIA is best at. As early as 2013, Nvidia released the Geforce GTX Titan Titan, which immediately became the computing foundation of global autonomous vehicle and advanced driving assistance systems. Last September, NVIDIA launched a new generation of autonomous driving chip “Thor” at its GTC developer conference, with a computing power of 2000 TOPS.
Snipaste_ 2023-04-20_ 19-28-45.png
Image/Baidu
Among global autonomous driving companies and car companies, except for a few manufacturers such as Tesla who choose to turn to self research, as long as they are interested in targeting L4 level autonomous driving, they are all NVIDIA customers, such as Baidu Apollo.
Although the commercialization of autonomous driving has not yet materialized, the chip market is far from exploding, and NVIDIA’s automotive business revenue accounts for less than 5%, it still maintains a year-on-year growth of 86%.
More importantly, as car companies and autonomous driving jointly promote the upgrade of L2 to L4, once successfully implemented on a large scale, the autonomous driving chip market will inevitably experience explosive growth, and NVIDIA’s computing power advantage will also be transformed into huge scale and profits. This is why NVIDIA has independently calculated it in its recent financial reports.
In contrast, the craze for the metaverse has faded and has entered a dormant period, but Nvidia’s ambitions will clearly not stop there. In fact, prior to the 2021 Metaverse Fire, NVIDIA launched the Engineer Metaverse Omniverse at the GTC 2021 conference in April, leveraging years of experience in the graphics field.
Unlike other metaverse concepts, Huang Renxun sees Omniverse as a platform to connect the 3D world to a shared virtual world. One example is that with the help of its massive GPU computing power, German Railways has built a “digital twin” of its operating railways on Omniverse, including 5700 stations and a total length of 33000 kilometers of track, where AI models can be trained, various unexpected problems can be simulated, and various operational schemes can be verified.
The most shocking thing is that the documentary released later showed that Huang Renxun, who had previously spoken for an hour and a half at GTC 2021, was actually a “digital version of Huang Renxun”, including the kitchen, which is also a “digital kitchen” based on Omniverse and digital twin technology.
Indeed, the metaverse still has a long way to go, from infrastructure to terminal devices to content applications, all in their early stages. However, NVIDIA has created a complete set of technology platforms for the metaverse, ranking alongside HPC (High Performance Computing) and AI as the three major platforms, indicating that it views the metaverse as a long-term war.
AI is even more so.
Huang Renxun’s Faith and Vision
Many people do not know that the reason why Nvidia is called the biggest winner behind the scenes of AI big models is not from the emergence of big models, let alone in 2022.
In March 2016, AlphaGO just defeated Li Shishi, giving humanity a small shock and sparking a new round of discussions and industry fever about AI. A month later, Huang Renxun said at the GTC China conference that NVIDIA is no longer a semiconductor company, but an AI Computing Company. In August of the same year, Huang Renxun also donated NVIDIA’s first AI supercomputer, the DGX-1, to the newly established OpenAI and wrote the following sentence:
For the sake of computing and the future of humanity, I donated the world’s first DGX-1
15OpenAI_ inside_ signed_ web.jpg
The world’s first DGX-1, illustrated by Nvidia
But if we delve deeper, Nvidia’s future dominance of the AI industry will start even earlier. David Kirk, the former chief scientist of NVIDIA, had a dream early on – to “generalize” GPU computing power that mainly serves “games” and only performs 3D drawing rendering, transforming it into a universal computing center.
So under the leadership of David Kirk and Huang Renxun, NVIDIA launched the revolutionary GPU unified computing platform CUDA in 2007, unleashing massive computing power from the game.
Looking back at the history of NVIDIA, especially after the launch of CUDA, NVIDIA and Huang Renxun regarded “computing” and “computing power” as the core of everything. Whether it’s AI, autonomous driving, metaverse, robotics, or cryptocurrency, Nvidia is looking for new opportunities with a huge and even some excess computing power.
Of course, Huang Renxun has a higher vision and a stronger belief in computational power. In August 2021, the Semiconductor Industry Association (SIA) announced that NVIDIA CEO Huang Renxun would receive the highest honor in the chip industry – the Robert Noyce Award. SIA President and CEO John Neuffer stated in a statement:
Huang (Renxun) has foresight and strong execution ability, promoting the development of the chip industry, overturning computing, and promoting artificial intelligence. From gaming to scientific computing, to autonomous driving, Huang Renxun’s achievements are closely related to countless innovations, and he has changed the industry and the world.
In the short term, the development ecology and technological advantages accumulated over the years make it almost impossible for Nvidia to break its rule over the computing world. Although cloud computing platforms such as Google and Microsoft are developing specialized AI chips for large models with high computing power, it is difficult to shake NVIDIA’s market position due to cost and versatility; Intel’s current development focus is still on CPU and manufacturing; AMD chose to bypass the head-on confrontation in computing power and focus on APU, but there is still a significant ecological gap that needs to be bridged.
However, the existence of the “dragon” itself is a threat, and new “warriors” will always be encountered. Moreover, NVIDIA, which was only established in 1993, is not much larger than many internet companies.