In 2016, Jensen Huang, the co-founder and CEO of NVIDIA, unveiled the company's flagship deep learning supercomputer, DGX One, alongside none other than Elon MuskThis event marked a significant moment in the tech industry, especially considering the fact that Huang donated this $129,000 machine to a fledgling startup that would later become globally influential: OpenAI, co-founded by Musk himselfFast forward seven years, and NVIDIA's A100 chip became so sought-after that it was practically sold out everywhere, and by May 2023, the company’s market capitalization surpassed an astonishing $1 trillion, making it the fifth most valuable company in the United States, effectively eclipsing giants like Intel and AMD.

But how did NVIDIA, once just a card seller in the gaming industry, go through a remarkable transformation to become the cornerstone of AI technology and development? The journey is not just about technology; it's about a relentless pursuit of innovation, strategic decision-making, and also a bit of luck.

The year 1993 was particularly pivotal for both the global tech landscape and the future of NVIDIA

This was the year that China officially connected to the internet, marking the beginning of a technological revolution in one of the world's largest markets, while in the U.S., the personal computer revolution was in full swingMajor players like Intel and AMD were basking in the limelight with their Pentium CPUs, leading the charge in a rapidly evolving industry.

At this time, Jensen Huang, who had spent a decade honing his skills in the chip industry, worked at AMD on chip design and LSI Logic in graphic processingHe was well-versed in technical intricacies, management, and sales—essential skills that would later serve him well as an entrepreneurBorn in Taipei in 1963, Huang was shipped off to live with his uncle in Washington State at just nine years of age and attended a rural boarding school where he adapted quickly despite the challenging environment.

While he engaged in teenage rebellion—such as sneaking out and smoking—Huang managed to rediscover his academic prowess, eventually earning admission to Oregon State University to study electrical engineering

It was during his time in the lab that he made a promise to his girlfriend, Lori: by the age of 30, he would own a company.

Fast forward to 1993: Huang kept his word and, together with two technical partners, co-founded NVIDIA, marking his 30th birthday with the company's inaugural operations.

NVIDIA focused on developing graphics processing chips, aligning itself with the rising tide of PCs, gaming, and multimediaHowever, the startup faced a significant hurdle early onIn 1995, NVIDIA launched its first multimedia accelerator NV1 aimed at gaming consoles, securing a noteworthy $7 million contract with Sega for the development of NV2. But shortly after, with the release of Windows 95 from Microsoft, the gaming market rapidly shifted from consoles to PCs, and NV2 found itself incompatible with the new Windows standard.

In a moment of vulnerability, Huang took the brave step of confessing the predicament to Sega, asking to halt the development and requesting full payment

To his surprise, Sega agreed, reflecting the sincerity and boldness of Huang's approach.

By 1997, NVIDIA launched the Riva 128 graphics accelerator, designed specifically for the PC marketThanks to its high cost-performance ratio, the unit sold over a million units in just four months, marking the beginning of NVIDIA's ascent as it set its sights on the then reigning champion, 3Dfx.

To surpass 3Dfx, Huang adopted an aggressive approach, committing to a model of three teams producing updates every six monthsHe introduced the "Huang's Law," claiming NVIDIA's chip performance would double every half-year.

In 1998, NVIDIA released the even more powerful Riva TNT, establishing itself as a formidable competitor to 3DfxThe company’s successes can be traced back to two pivotal decisions: first, closely aligning with Microsoft's burgeoning technology standards during a time of contention in the graphics processing sphere, and second, forming a strategic partnership with TSMC to focus on design while leveraging TSMC’s manufacturing capabilities.

By 1999, NVIDIA went public and introduced the world’s first GPU, the GeForce 256, delivering unprecedented performance and outpacing 3Dfx in market share

alefox

A year later, NVIDIA acquired 3Dfx, consolidating its dominance in the graphics card realm.

However, as the top graphics card producer, NVIDIA soon faced a new challenger: the venerable ATI, founded by the Hong Kong-born entrepreneur Benny HsuIn 2000, ATI announced the Radeon 256 to compete against NVIDIA’s GeForce 256 with superior performance, igniting an intense rivalry that would last for years.

By 2003, when China was established as the world's second-largest PC market, ATI launched its high-end Radeon 9800 XT in China, prompting NVIDIA to quickly counter with the GeForce FX 5900 just four days later.

ATI struggled amidst this fierce competition as its founder, Hsu, was ultimately sidelined due to financial controversies, leading to ATI's progressive declineConversely, NVIDIA maintained an upper hand, controlling the mid-range market profits.

ATI was acquired by AMD in 2006, but the latter's financial struggles post-acquisition left it vulnerable to NVIDIA’s persistent onslaught, allowing NVIDIA to claim over 80% of the desktop graphics card market.

Flushed with success, NVIDIA adopted "Huang's knife technique," systematically slicing its graphics card performance profiles.

Yet, Huang had the foresight to recognize that relying solely on gaming graphics wouldn’t ensure long-term viability; the company needed to evolve

At this point, GPUs were primarily utilized just for 3D graphical rendering and were often underutilized.

In 2006, NVIDIA’s chief scientist, David Kirk, proposed a bold idea: to generalize GPU technology and make it programmableThough met with opposition from the board due to the hefty costs and increased complexity, Huang decided to push ahead with the development of the CUDA platform.

This decision was a high-stakes gamble that faced significant pushbackThe initial years saw NVIDIA struggling, with profits plummeting and stock prices stagnating for five long years.

Fortunately, NVIDIA’s gamble on CUDA would eventually pay off handsomelyIn 2012, Geoffrey Hinton, one of deep learning's three pioneers, used two GTX 580 graphics cards to train a deep neural network known as AlexNet in just six days, clinching victory at the ImageNet large-scale visual recognition challenge

This breakthrough illustrated the immense potential of GPUs in deep learning, prompting NVIDIA to embrace the AI wave.

The appeal of GPUs in the AI era hinges on their operational differences from CPUsWhile CPUs have fewer cores designed for complex logical operations, GPUs boast thousands of simpler cores suited for parallel processing, thus gaining a clear edge in handling substantial data workloads.

NVIDIA’s CUDA platform played a decisive role in unlocking GPU capabilities, lowering programming barriers significantlyEngineers and scientists could now harness GPU power without steep learning curvesAlthough AMD later rolled out a similar platform known as RCM, it came a decade too late.

With their first-mover advantage, NVIDIA built a comprehensive eco-system of hardware and software standards, dominating the AI landscape