Image Credits - The Brand Hopper
According to the Economist, no company has gained as significantly from the artificial intelligence (AI) explosion as Nvidia. The chipmaker’s stock value has experienced a remarkable increase of nearly 450% since January 2023. Nvidia’s market capitalisation is nearing $2 trillion, positioning it as the third-largest company in the United States, trailing only behind Microsoft and Apple.
For the latest quarter, Nvidia reported revenues of $22 billion, a significant increase from $6 billion in the corresponding quarter of the previous year.
The majority of analysts anticipate that Nvidia, holding over 95% of the specialist AI chip market, is poised for rapid expansion in the coming years. What sets its chips apart?
Originally developed for video gaming, Nvidia’s AI chips, commonly referred to as graphics processor units (GPUs) or “accelerators,” have found a broader application.
Nvidia’s chips operate using a method known as parallel processing. This technique involves splitting up a complex calculation into smaller pieces, and then simultaneously processing these pieces across numerous “cores” — which can be thought of as the chip’s mini-brains. This approach allows the chip to handle multiple tasks at the same time, significantly speeding up the processing of data and making the chips highly efficient for tasks requiring intense computation, such as AI and gaming.
This approach enables a GPU to perform calculations much quicker than if it were to process tasks one after the other. Such a method is perfectly suited for gaming, where creating realistic graphics necessitates the simultaneous rendering of numerous pixels on the screen. Nvidia’s top-tier chips currently make up 80% of the gaming GPU market.
Fortunately for Nvidia, its chips have discovered a broader array of applications beyond their original purpose, including cryptocurrency mining, autonomous vehicles, and most notably, the training of AI models.AI is powered by machine-learning algorithms that utilize a subset of deep learning known as artificial neural networks. Within these networks, computers can identify rules and patterns by analyzing extensive datasets.
Algorithms are step-by-step instructions or rules designed to perform specific tasks or solve problems, used in computing to process data and make decisions.
Deep learning is a type of machine learning where computers learn from examples in a way similar to human learning. It uses artificial neural networks, which are designed to imitate the human brain, to help computers understand data, recognize patterns, and make decisions.
Educating a network requires vast amounts of computations. However, since these tasks can be divided into smaller segments, utilizing parallel processing significantly accelerates the process. A high-end GPU possesses over a thousand cores, enabling it to execute thousands of calculations concurrently.
Upon recognizing the high efficiency of its accelerators in training AI models, Nvidia shifted its focus towards tailoring them for this specific market. The company’s chips have evolved to match the growing complexity of AI models, achieving a 1,000-fold enhancement in computational speed over the decade leading up to 2023.
Nvidia’s skyrocketing value isn’t solely due to the enhanced speed of its chips. Its advantage also spans two additional domains, including networking. As AI models become increasingly complex, the data centers operating them require thousands of GPUs linked together to amplify computational capacity, whereas most computers typically utilize just a few.
Nvidia links its GPUs using a high-speed network built on technology from Mellanox, a networking technology provider it purchased in 2019 for $7 billion. This integration enables Nvidia to enhance the performance of its chip network in a manner unmatched by its rivals.
Another key advantage for Nvidia is CUDA, a software platform that enables users to precisely adjust the performance of its processors. Nvidia has been developing this software since the mid-2000s and has actively promoted its use among developers for creating and experimenting with AI applications. As a result, CUDA has become the recognized industry standard.
Nvidia’s substantial profit margins, coupled with the swift expansion of the AI accelerator market—expected to hit $400 billion annually by 2027—have drawn the attention of competitors. Companies like Amazon and Alphabet are developing their own AI chips for use in their data centers.
Other major chip manufacturers and startups are eyeing a portion of Nvidia’s market share. In December 2023, Advanced Micro Devices (AMD), a rival chipmaker, launched a chip that, by certain metrics, is approximately twice as potent as Nvidia’s most sophisticated chip.
However, surpassing Nvidia in hardware alone might not suffice. Nvidia leads the AI chip industry due to its superior chips, networking equipment, and software. For any rival aiming to overtake this semiconductor giant, excelling in all three domains is necessary. Achieving this is a formidable challenge.
(The author Girish Linganna of this article is a Defence, Aerospace & Political Analyst based in Bengaluru. He is also Director of ADD Engineering Components, India, Pvt. Ltd, a subsidiary of ADD Engineering GmbH, Germany. You can reach out to him at: [email protected])