They gave pretty colors, decent shapes, and even emotions to the pixelated heroes of video games. Graphics cards from the American company Nvidia, with its famous GeForce range, have dominated this sector so dear to gamepad and mouse addicts for more than twenty years. The latest models are always snapped up at high prices on specialized e-commerce sites or in physical stores, such as in rue Montgallet in Paris (12th arrondissement), an eternal cluster of computer shops.
But if the company founded by the American-Taiwanese engineer Jen-Hsun “Jensen” Huang exceeded Tuesday May 30 the 1,000 billion dollars of market capitalization in the United States, an honor so far reserved for less than a ten firms like Apple or Microsoft, it owes it not only to the power gained by the gaming in our leisure time. The artificial intelligence (AI) revolution has also been there.
Nvidia’s stature can be summed up in three letters: GPU, for graphics processing unit. These graphics processors, nestled within their cards, have become more and more powerful, benefiting in particular from the progress made in the fineness of semiconductor engraving. The turning point is difficult to locate precisely. But in 2006, Nvidia made its GPUs programmable. Shortly afterwards, in 2009, several groups of researchers, including one led by the eminent Andrew Ng from Stanford University, are beginning to use them for deep learning (deep learning), a growing subfield of AI. Yet slower and less efficient than their CPU cousins - for central processing unit, the real brains of computers – GPUs have the advantage of carrying more memory and a better ability to execute tasks in parallel. What, again according to Andrew Ng, make data learning up to 100 times faster than before. Nvidia is rubbing its hands. “The AI found us,” said Ian Buck, one of the senior executives, a little later.
“iPhone Time”
During the following decade, GPUs fueled research in artificial intelligence, which experienced spectacular leaps. In 2012, AlexNet, an image classification AI trained with Nvidia GPUs, fascinated the scientific community. In 2017, the research paper Attention is all you need, devoted to neural networks “Transformers” paves the way for generative AIs like ChatGPT. Nvidia naturally stands out as the favorite brand for XXL computers and the training of large models that allow these AIs to invent stories, images or music.
Near Paris, on the Saclay plateau, the Jean Zay supercomputer, recently visited by L’Express, contains several thousand of them, having been used in the development of Bloom. OpenAI has also trained GPT-3 as well as GPT-4, the origin of its chatbot, thanks to these GPUs. Recently, Nvidia launched its newest line of chips, the H100, for $40,000 apiece, intended to “speed up the most complex language models up to 30 times over previous-generation products,” the company boasts. on his website. Jen-Hsun “Jensen” Huang believes she’s in an “iPhone moment,” comparing the current growth of AI to that of smartphones after Apple launched its breakthrough phone.
Unlike the cryptocurrency market, in which Nvidia has invested by offering bitcoin “miners” its imposing computing capacities before backing down, the generative AI bubble seems to have a very bright future ahead of it. The global generative AI market could generate 300 million jobs and increase global GDP by 7% in the next ten years, according to Goldman Sachs. Nvidia holds neither more nor less than the main technological brick. The firm now weighs five times more than its competitor AMD, and more than eight times Intel, in full questioning. “What Nvidia represents for AI is almost equivalent to what Intel represented for PCs”, analyzes bluntly Dan Hutcheson, analyst at TechInsights, with the bbc. The other challengers, like Google or Britain’s GraphCore, which relies on another technology, Intelligence Processing Units (IPU), are currently lagging behind in terms of performance. “There are several variants on the market, but few are ready to move Nvidia or eat into its market share. This is the default option,” observes Alan Priestley, analyst at Gartner, quoted by Bloomberg. The company could pocket $11 billion in revenue in the next quarter. That’s as much as the whole of 2020. Thank you gamers.