文 | Age
In the middle of the nineteenth century, there was a vigorous gold rush in the western United States, but it was not the gold diggers who gambled their lives to make money. A businessman named Sam Brennan became the first millionaire in California's history by selling shovels to gold prospectors.
Every time the tuyere appears, there are always companies that want to play the role of "shovel sellers". Especially in the wave of the metaverse and artificial intelligence, some companies directly wrote the identity of "shovel sellers" on the PPT, for fear that the outside world would not understand their positioning.
But until now, it seems that only Nvidia can really be called a "shovel seller".
On June 18, Nvidia's market capitalization reached $3.34 trillion, officially surpassing Microsoft to become the world's most valuable company. Hans Mosesmann, an analyst at Rosenblatt Securities, even boldly predicted that Nvidia will continue to rise in the coming year, and its market capitalization is expected to reach almost $5 trillion.
Looking at the growth process of NVIDIA's market value, this "shovel seller" founded in 1993 did not begin to show a soaring trend until after 2017, and the mainstream technologies that currently support NVIDIA's market value all originated before 2017.
In the 25 years before that, why was Nvidia unremarkable?
01 I chose the right graphic track, but I almost lost to the "wrong route"
Back in early 1993, 30-year-old Huang quit his job at LSI Logi and started his first company, Nvidia, with graphics processors on the track.
At that time, there were already more than 20 companies in the market, including large companies such as IBM, HP, Sony, Fujitsu, and Toshiba, as well as many entrepreneurs like NVIDIA. With their connections and experience in the semiconductor market, Huang and Nvidia still got the ticket.
In May 1995, NVIDIA launched its first product, the NV1. At that time, there was no concept of a graphics card, and this product was called a multimedia accelerator, which technically integrated 2D and 3D graphics processing, and had functions such as video processing, audio wavetable processing, and gaming ports.
In the second month of NV1's release, Nvidia received its first round of funding from Sequoia Capital and Ciria Ventures; Because the NV1 integrates the sound card and controller control unit into the chip, Nvidia has also successfully attracted the attention of game console manufacturer Sega, winning an order for the graphics processing part of Saturn's next-generation console, as well as a $7 million development deposit.
originally thought that Nvidia would make great progress in the game market and open the way to rise in the graphics processor market, but the emergence of Microsoft directly rewrote the trajectory of Nvidia.
At that time, graphics acceleration technology was in the ascendant, and there was no unified standard in the industry, and many companies adopted the specification of "triangle drawing", and NVIDIA chose "quadrilateral drawing". In 1996, Microsoft released the Direct 3D standard, announcing that graphics software would only support "triangle drawing". However, Nvidia did not realize the changes in the external environment at the first time, and still adhered to the route of "quadrilateral drawing".
Faced with the general situation that the industry standard has been set, NVIDIA was forced to pay for its "mistakes" - the NV1, which took two years to develop, was quickly ignored, and the NV2 developed for Sega ended in failure, and Nvidia was on the verge of bankruptcy. And 3Dfx, which conforms to the laws of history, with its first product Voodoo, occupied 80% of the market in one year and pry away Sega's orders from Nvidia.
Huang realized that "if you want to survive, you have to change."
On the personnel front, Huang appointed David Kirk, chief technology officer of console game maker Crystal Dynamics, as Nvidia's "chief scientist"; In terms of R&D, NVIDIA switched to the mainstream market to benchmark 3Dfx, which has set off a 3D technology revolution; At the same time, a six-month internal cycle goal was established, so that even if a product fails, it will not threaten the survival of the company, because the next generation of replacement products is readily available.
In April 1997, NVIDIA launched the third-generation product NV3, which became known as the RIVA 128. Although it is inferior to 3dfx's Voodoo in terms of image quality, the 100M/s pixel fill rate, compatibility with Open GL, and price advantage have won the favor of OEMs, with more than 1 million pieces shipped in less than a year, forcing NVIDIA to pull back from the "brink of death".
In the following year or so, NVIDIA successively launched Riva TNT, Riva TNT2 and other products, and continued to expand its market share through a set of strengthened combinations. 3DFX, which briefly dominated the market, failed to do so, and Nvidia went public on the Nasdaq in 1998 with revenues of more than $150 million.
1999 is an important point in the history of NVIDIA, in addition to the highlight of the listing, but also in August of this year launched the first product named GeForce - GeForce 256, the concept of GPU was first proposed, and then Dell, Gateway, Compaq, NEC, IBM, etc. have announced the pre-installation of Nvidia's GPU. Nvidia, which almost "collapsed in the middle way", has become a "Silicon Valley rookie" in the eyes of the media.
02 CUDA, who became a god in the future, created a third crisis
An interesting episode is that NVIDIA, which was in trouble because of Microsoft's Direct 3D standard, "pitted" Microsoft soon after.
The best-selling of Nvidia's GPUs has accelerated the popularization of the DirectX standard, and Microsoft has also praised the start-up, not only allowing Nvidia to participate in the formulation of the DirectX standard, but also allowing Nvidia to get the largest order since its inception - the display chip of the original Xbox.
In order to let Nvidia fully develop GPUs for Xbox, Microsoft directly paid a deposit of $200 million, and the contract amount reached $500 million, and later handed over Xbox's media delivery processor (MCP) to Nvidia R&D, and the relationship between the two sides can be described as a "honeymoon period".
On November 15, 2001, Xbox was first released in the United States at $299. At the time, the starting price of the NVIDIA GeForce 3 was $329, and the Xbox's X-Chip chip was equivalent to a modified version of the GeForce 3. At the time, some analysts said that Microsoft would lose $125 for every Xbox it sold. Microsoft wants to reduce the cost of Xbox's components, and Nvidia can't get around it.
When Microsoft asked Nvidia to reduce the price, it was met with a rejection from Huang. The conflict between the two parties then erupted quickly and eventually resorted to arbitration. After the news of the "breakup" with Microsoft, Nvidia's market value shrank from $11 billion in early 2002 to $1 billion, and the price was not tragic.
Under Microsoft's fierce fire, Huang had to bow his head and formally settle with Microsoft in February 2003, but failed to redeem Microsoft's "true love." The injured Microsoft threw an olive branch to Nvidia's rival ATI, which will serve as the GPU supplier for the next generation of Xbox 360.
With Microsoft's support, Nvidia and ATI were in a state of struggle for the next many years. Not only is there a competition in products, but there is also a tug-of-war in market share. In the third quarter of 2004, for example, ATI had a 59% market share in the discrete graphics card market, while NVIDIA had only 37%.
It wasn't until July 2006, when AMD acquired ATI for $5.4 billion, that the scales began to tip in favor of Nvidia. To acquire ATI, AMD almost ran out of cash on its books and incurred $2.5 billion in debt. Because of the debt crisis, the merger of the two failed to appear 1+1 greater than 2 "synergy", but let Nvidia pick up the "pie falling from the sky".
Prior to this, Nvidia was almost locked in the GPU track, unable to expand horizontally. After that, Nvidia accelerated the pace of breakthrough, and the representative action was CUDA.
In 2007, NVIDIA held the first CUDA Technology Conference, the full name of which is Compute Unified Device Architecture, a parallel computing platform and programming model that enables developers to write GPU-accelerated applications using programming languages such as C, C++, and Fortran.
Without CUDA, the GPU is just a "graphics processing unit" responsible for drawing images on the screen; With CUDA, GPUs have the ability to solve complex computing problems and help customers program processors for different tasks.
The main use cases of GPUs are no longer limited to gaming.
The story behind it is familiar to many people. Deep learning, which became popular in 2012, and ChatGPT, which exploded in 2022, finally made CUDA a god. GPUs are widely used in scenarios such as image recognition, face recognition, and voice recognition, and their value even exceeds that of CPUs.
But in the context of 2008, Nvidia spent about $500 million a year on research and development costs on CUDA, accounting for one-sixth of its overall revenue, but it barely saw commercial value. Reflected in the financial report, Nvidia's net profit and gross profit after 2008 fell sharply, and there was a loss for a time.
During this time, Huang repeatedly painstakingly explained the value of CUDA, but Wall Street analysts did not buy it, and Nvidia's market capitalization hovered around $10 billion for a long time.
03 Failure to impact the mobile market, deep learning is "not popular"
With $3 trillion in Nvidia under management, Huang's every word can be praised as "visionary," and every act has the potential to become a famous spectacle. But with $10 billion under management for Nvidia, Huang has to find ways to boost its stock price and regain analysts' confidence.
So when CUDA is not favored, Nvidia must chase the wind.
The PC market has shown a downward trend after 2008, and Nvidia, whose main business is GPU, knows that it is warm and cold. It was in this year that Nvidia launched the Tegra series of chipsets for the mobile market, using the ARM architecture, a built-in GeForce-based graphics processor, and with a clear idea: as long as there is demand for mobile games, Tegra can have a place in the mobile market.
Judging from the time alone, Nvidia's entry time is not late.
Qualcomm launched the Snapdragon S1 in 2007, but production only began in the fourth quarter of 2008. Apple only launched its first self-developed chip, the A4, in 2010; MediaTek entered the smartphone market in 2011; Samsung's first Exynos chip was also ...... in 2011 What's more, Nvidia has more than ten years of experience in chip research and development, and its relationship with TSMC is also quite close.
It's just that the results are far from what I expected.
Due to its lack of experience in mobile, when the first Tegra chip was released, Nvidia chose to cooperate with Microsoft and carry it on a product called Zune HD, which was not as well-known as the iPhone. By the time the Tegra 2 series was launched, it was already compatible with the Android system, and therefore won the support of brands such as Motorola and LG.
The Tegra 3 series has made a name for itself and is used in flagship products such as the HTC One X and Nexus 7, but the problems have also been exposed. The self-developed mobile GPU provides powerful graphics computing performance, but at the cost of high power consumption and heat, which directly drags down the reputation of related products.
So there is a familiar scene for domestic netizens: at the Xiaomi 3 press conference in 2013, Huang Jenxun came to the scene as a special guest, "staying up late" to learn Chinese, and chanting "Xiaomi mighty". At that time, Xiaomi was just starting out, but it was one of the few end customers of Tegra 3.
This time, Nvidia is no longer "favored by God", Tegra, who gambled on the wrong technical route, did not wait for the opportunity to "come back", under the attack of Qualcomm and Samsung, Tegra series chips were quickly left out in the cold, and Nvidia, which was originally full of confidence, unfortunately passed by the trillion mobile market.
In 2012, when Nvidia failed in the mobile market, Alex Krizhevsky, a researcher at the University of Toronto, trained the neural network AlexNet on two GTX 580 graphics cards, and won the ImageNet challenge by a significant margin in just one week.
This unexpected surprise has led to a wave of using GPUs for deep learning model training, and at the same time, it has also opened the door to a new world of AI for NVIDIA: deep learning and brute force computing with GPUs "hit it off", opening a race game of data and stacking hardware.
Since the GTC conference in 2014, Huang has officially made AI the most critical business of NVIDIA, and almost every year he will sing Moore's Law and talk about the explosive rise of GPU-accelerated computing. In 2016, when NVIDIA launched the Pascal-based GPU designed for AI, Huang personally sent the world's first DGX-1 supercomputer to a start-up called OpenAI.
Nvidia has high hopes for the future of artificial intelligence, but the market is not cold.
Before 2016, Nvidia's market capitalization was still stable around $15 billion, and its revenue in fiscal 2016 was only $5.01 billion, less than one-tenth of Intel's. The so-called wave of deep learning can be said to be a typical "thunder is loud and rain is small".
In those long days, Microsoft and Google were the giants of the technology industry, Intel and Qualcomm were the leaders of the semiconductor industry, and Nvidia was just the leader of the game market segment, a "small chip company" that rose and fell with the rise and fall of the game market.
04 Accelerating the transformation to "data center", but the market "didn't understand"
The story of NVIDIA began in 1993, but it really came into the public eye in 2018, with a gap of 25 years.
As Huang later described it, 2018 was "a near-perfect year that ended in a tumultuous fashion." In October of that year, Nvidia's market capitalization reached its peak since its listing, reaching $218.9 billion. It wasn't long before Nvidia's market capitalization fell in half.
Looking back on that period of history, many people habitually focus on Nvidia's pivot in 2017: At GTC in 2017, Huang unveiled the Tesla V100 GPU for AI, which he called "the world's most expensive computing power project" with a $3 billion R&D budget. At the same time, it also marks that the data center business is defined as the second curve of NVIDIA, and the market value rushed to $71.4 billion on the same day.
Time has once again confirmed the correctness of NVIDIA.
In the summer of 2020, the revenue of the data center business surpassed that of the gaming business to become Nvidia's largest source of revenue. It is the surging computing power provided by NVIDIA that OpenAI trained the world-renowned ChatGPT and opened the curtain of generative AI, and Nvidia's market value quickly crossed the $1 trillion level and soared to $3 trillion, once becoming the world's most valuable company.
But back to the context at that time, compared to the slightly jerky word "data center", what is most tied to NVIDIA is actually "mining".
Mining bitcoin requires a "mining machine", and a "mining machine" to run a hash algorithm requires computing power, and the source of computing power is NVIDIA's GPU. "Mining" grew in popularity in 2016 and reached its climax in early 2018, during which the demand for GPUs continued to grow. The Nvidia GTX 1080 retails for $550 and has been speculated above $1,000 in the market, so much so that Nvidia has to introduce a "purchase restriction policy" to prioritize the sale of graphics cards to gamers rather than miners.
However, the gradual decline in the price of Bitcoin since the end of 2017, coupled with the rising cost of mining new coins, has accelerated the decline of the "mining boom". Miners continue to sell graphics cards, the order of the market is disrupted, graphics cards prepared for "mining" are placed in piles in warehouses, and Nvidia's channel inventory is also in surplus. The market was eventually swept up in negative sentiment, and Nvidia's stock price immediately fell, only to return to its highs in early 2020.
Unlike Huang's belief in computational value, markets tend to be short-sighted, accustomed to seeing problems through appearances rather than the underlying logic behind phenomena.
Digital currency, metaverse, generative AI and other outlets have stimulated NVIDIA's growth again and again, giving the outside world full confidence. And when the bubble bursts or the tuyere weakens, Nvidia will inevitably be caught in it. No one will delve into how many years Nvidia has endured the cold window, how many years of cold bench it has sat on, and how many attempts it has made on the wrong route in order to support the new outlet.
This is also the fate of all "shovel sellers", the gold rush does not exist, who else will the shovel be sold to?
Even though Nvidia is already a behemoth with a market capitalization of more than $3 trillion, there are still many analysts who question its value. For example, the warning issued by "Sister Wood": Will Nvidia be the next Cisco? After all, looking at the curve of market capitalization rise, today's Nvidia is not indistinguishable from Cisco around 2000, when the dot-com bubble peaked, Cisco's stock price rose 71 times, plummeted by 90% in the following years, and never returned to its highs since.
Huang may not be able to answer these questions. Cloud service providers such as Amazon and Microsoft have bought about half of Nvidia's GPUs in the past few quarters, and Nvidia's fate remains in someone else's hands. Compared with the situation in 2018, Nvidia has not changed much, but the outside world's expectations of Nvidia have changed.
How to measure the value of NVIDIA? We're here to offer a new perspective on Huang's Law, which was proposed by Huang in 2018: GPU performance will more than double every two years. Nvidia's stock price and GPU performance growth over the past five years have been surprisingly consistent, and when the "Huang's Law" no longer works, it may be the time when Nvidia's market capitalization peaks.
05 Conclusion
Compared to most companies in the world, Nvidia is a completely lucky one.
To make a simple conversion: about 90% of companies fall in the start-up stage, the reason may simply be to choose the wrong technical route that the giants bet on; 90% of companies that have survived the start-up stage may fall into the trap of long-term unprofitable, with almost no capital and shareholders allowed to invest in unprofitable businesses for a long time; The third stage is often the status quo of many medium and large enterprises, such as Intel, which repeatedly miss new windows of opportunity and are forced to struggle in the old track.
Nvidia is currently the only company whose core business value has not been fully verified, but has pushed up its market value with external dividends again and again.
At the same time, I have to admit that NVIDIA is a very pure company, and behind every "turning the crisis into safety", it is relying on technology and products to win the battle. Huang is also a soft-standing manager, who is open to self-reflection and willing to bow to Microsoft and curry favor with Xiaomi in exchange for publicity opportunities, even if it is the newly founded OpenAI, and gives him enough respect...... will play to the extreme with the way of overcoming rigidity.
As Jensen Huang gave at the National Taiwan University graduation ceremony: Embrace failure, be brave enough to ask for help, and persist in achieving your dreams in the face of pain, while focusing on your true strengths and daring to retreat and give up strategically. And cheer yourself up with grace: Run! Whether it's running for food, or running without being treated as food by others.
A Nvidia who knows his strengths, has clear goals, is modest and cautious, and is not arrogant or impatient, even if he has been "suppressed" for 25 years, he can climb up from the trough again and again and shine at the right time.