Nvidia concluded with great fanfare the year which established it as the kingmaker of artificial intelligence (AI), with record revenues and profits, driven by strong demand for electronic components of which it is the undisputed master.
The American AI chip giant achieved nearly $61 billion in revenue and multiplied its net profit six-fold, to nearly $30 billion, over its entire annual financial year ended at the end of January, d ‘after a results press release published on Wednesday.
“Nvidia has enabled the emergence of a new era of computing, that of generative AI, where software can learn, understand and generate all kinds of information: human language, biological structures or even 3D universes” , welcomed Jensen Huang, founder of Nvidia, during a conference for analysts.
Processors called “GPUs”
Unknown to the general public until a year ago, the Santa Clara (California) group manufactures graphics cards, processors called “GPUs”, initially designed for the needs of video games. Their computing power, necessary for processing massive quantities of data, has made them essential to developers of generative AI models (OpenAI, Meta, Google, etc.)
This technology, which makes it possible to produce text, images and other data on a simple query in everyday language, has aroused extraordinary enthusiasm since the successful launch of ChatGPT by OpenAI at the end of 2022.
From November to January, Nvidia achieved $22 billion in revenue and $12.3 billion in net profit, results well above its forecasts and market expectations.
“AI Festival”
And the demand is not weakening. The American company expects $24 billion in turnover for the current quarter. Enough to impress Wall Street: Nvidia’s stock jumped more than 9% during electronic trading after the close of the New York Stock Exchange.
“Nvidia remains the only player capable of producing the GPUs that power the AI revolution,” recalled Wedbush analyst Dan Ives on Wednesday. “The AI revolution started with Nvidia, and we believe the AI party is just getting started.”
According to him, the year 2023 marked “the start of the greatest technological transformation since the first days of the internet in 1995”.
Nvidia’s graphics cards – like its flagship GPU, the H100, which are worth tens of thousands of dollars a piece – power servers in data centers specializing in AI. This sector has become the group’s main customer. Nvidia’s components are essential for training language models with mountains of data, and also for their practical application, or “inference,” when a user formulates a query and the model responds with the logical sequence of words /pixels/sounds/code most likely.
“Almost every time you interact with ChatGPT, every time you use Midjourney (image generator), every time you generate these incredible videos with Sora (latest from OpenAI) or Runway, Nvidia is doing inference “, emphasized Jensen Huang. “The share of inference in our activity has seen considerable growth, of around 40%”.
Chinese shadow
The demand for its components is such that the company must arbitrate between its customers, who are often competitors. “We do our best to allocate fairly and avoid allocating unnecessarily” when a data center is not yet ready, explained the US-Taiwanese leader. He also mentioned “sovereign AI”, that is to say the desire of many countries not to depend on American AI models.
“They want to transform their own data, which belongs to them – their language, their history, their culture – to create their own digital intelligence,” he detailed, betting on the development of “sovereign AI infrastructures ” elsewhere than in the United States, but still with Nvidia chips.
The only downside for the group: the restrictions on exports to China imposed by the American government. “We had to pause in this market, to reconfigure our products so that they absolutely cannot be hacked,” admitted the engineer. “Our business suffered, but after the current quarter we hope to be able to return.”