'100% Winning Probability' Lottery Revealed... Nvidia's Secret to Never Missing First Place [Tech Talk]

'Born out of necessity' GPU
From scientific chips to AI accelerators
Jensen Huang, pursuing relentless evolution

There is a term called 'Hardware lottery.' It was proposed by Google Brain researcher Sarah Hooker in 2020 to explain the popularity of NVIDIA graphics processing units (GPUs), which have rapidly risen as AI accelerators.

Processors Born Out of Necessity

Jensen Huang, CEO of NVIDIA, is showcasing a new product at the AI conference 'GTC 2024' held on the 18th (local time) in San Jose, California, USA. GTC is the world's largest AI conference hosted annually by NVIDIA.
[Photo by AFP Yonhap News]

The core idea of the hardware lottery is simple. The semiconductor hardware that becomes the 'standard' is actually determined not by the manufacturer's outstanding technology or design, but by the needs of software developers.

In other words, when AI first emerged, computer scientists needed multi-core processors capable of handling vast neural networks. However, the CPU, which was the dominant hardware in the market at the time, was not suitable, so they hastily chose GPUs instead. In short, NVIDIA GPUs simply had the most efficient architecture for machine learning (ML) among the processors readily available in the market at that time.

You can understand why Hooker likened this phenomenon to a 'lottery.' No matter how excellent a company, individual, or country is, no one can predict which software will become dominant in the future.

How NVIDIA Has Relentlessly Maintained Its Top Position

So, did NVIDIA seize the semiconductor industry's dominance today simply because of luck? Not exactly. While the initial selection of GPUs as AI processors might have been a stroke of luck comparable to winning the lottery, the momentum has been sustained thanks to the hard work and dedication of CEO Jensen Huang and NVIDIA’s engineers.

A prime example is NVIDIA’s parallel computing platform, 'CUDA.' Since 2004, GPUs have been evolving into parallel computing processors with multiple cores, and CUDA was the platform designed to control them. Today, neural network AI is also processed through parallel computing.

NVIDIA CUDA logo
Photo by NVIDIA

However, when CUDA was first announced in 2007, machine learning was an extremely unfamiliar term to the public. When CEO Huang initially developed CUDA, he intended to create a scientific computer to calculate complex phenomena such as particle movement and physical processes. But in the mid-2010s, as machine learning began gaining popularity, NVIDIA swiftly pivoted toward AI.

Over the past decade or so, GPU architecture has undergone countless changes. NVIDIA has continuously developed new technologies suitable for parallel computing, such as CUDA cores, Tensor cores, and Transformer cores, integrating them into GPU designs. As a result, the architecture of the latest generation of data center (AI) GPUs is diverging significantly from gaming GPUs.

During the transformation from a device for gaming to a scientific processor and then to an AI-dedicated processor, NVIDIA focused on only one thing: the 'feedback' from the actual consumers purchasing the computer chips.

In fact, NVIDIA’s real strength is not its technology (NVIDIA’s R&D spending relative to sales is much lower than that of most big tech companies) or foresight. Instead, NVIDIA has always been closely connected with chip users, including research labs of large companies studying AI and university labs. By constantly monitoring the direction in which software evolves, NVIDIA has been able to maintain its position as the 'hardware lottery' winner for over a decade.

CEO Personally Visiting from Big Tech to Startups

Driving AI of the autonomous driving startup 'Wave' introduced at GTC. Wave is a company developing full autonomous driving using only cameras and reinforcement learning, and is regarded as a company that has brought innovation to the generative AI field.
[Image source=X capture]

To see how consumer-friendly NVIDIA is, take a close look at the recent GPU developer conference, 'GTC 2024.' CEO Huang not only introduces new products but also meticulously checks what technologies companies are developing using GPUs. He personally visits booths of big tech companies that purchase tens of thousands of chips as well as AI startups just beginning their research.

As NVIDIA jumped to become one of the top three companies on the New York Stock Exchange, CEO Huang has been receiving spotlight attention like a 'rock star.' Some praise his vast vision and persistent business philosophy, while others express envy.

CEO Hwang took a photo with the technicians of the genome analysis device 'Oxford Nanopore.' Genetic engineering is a new technology field actively integrating GPU and AI, and Nanopore is one of the emerging technology companies gaining attention.
[Image source=Ex]

However, CEO Huang and NVIDIA have always been a company focused on 'consumers,' not 'vision.' GPUs are processors that have evolved according to immediate needs rather than a grand plan looking decades ahead.

In fact, it is no exaggeration to say that CEO Huang’s business strategy is a strategy of no strategy. In an interview last year with Sweden’s 'SANA lab,' he said about NVIDIA’s business strategy: "We do not make periodic strategies like five-year plans because the world we live in is a living, breathing place."