(BFM Bourse) – The famous graphics processor manufacturer unveiled its new product on Monday during its developers conference. The event was probably too anticipated to fully satisfy the market.
Star of the American stock market, Nvidia has become so much “THE” big stock on Wall Street that its conferences now constitute a full-fledged event for the market, like Apple’s “keynotes” in the past.
The group held its developers conference in San Jose, California on Monday. Wedbush analyst in charge of tech, Dan Ives, even described this meeting as the “Woodstock of artificial intelligence (AI)”. “For Nvidia and Jensen (Huang, the group’s CEO, Editor’s note) all eyes will be on the details and price of the new Blackwell B100 AI chip and architecture, as it represents another pioneering moment for the gaming industry. ‘AI in the future,’ he warned.
Nvidia, in fact, lifted the veil on Monday on this new graphics processor (GPU). This GPU based on the Blackwell architecture has superior characteristics to the H100, the group’s flagship product released in 2022.
>> Access our exclusive graphic analyses, and gain insight into the Trading Portfolio
Bigger, more powerful
Named in homage to David Blackwell, the first academic to join the Academy of Sciences and a specialist in game theory and statistics, the graphics processor would be, according to Jensen Huang cited by the Financial Times, twice as powerful to train the generative AI models. It would also be five times more efficient in “inference”, that is to say the speed at which AI models such as ChatGPT can answer queries, continues the “FT”. This GPU also combines 208 billion transistors compared to 80 billion for the H100, according to the British daily.
Clearly, Blackwell provides much more computing power, which is essential for developing large advanced language models (LLM), the foundations of generative AI. Blackwell will serve as the basis for creating Nvidia’s “GB200 superchip” which will combine two Blackwell GPUs with the company’s “Grace” CPU.
“We need bigger GPUs,” Jensen Huang said at that conference. This is in order to be able to cope with exploding demand as well as the need to reduce costs.
Important point: according to Agence France Presse, Nvidia claims that this “superchip” would be 25 times more efficient in terms of energy efficiency. Which proves to be imperative for Nvidia to strengthen a leadership which will end up being contested by its rivals, such as AMD, or by Gafam themselves.
“There is an important issue in terms of the price of GPUs and their electricity consumption, we must change the paradigm in the long term,” warned Jean-Edwin Rhea, fund manager at Quadrille Capital, in February in BFM Bourse. An Nvidia H-100 GPU currently costs $40,000 per unit and ChatGPT needs 10,000, “which is expensive” added the specialist.
It is therefore important for Nvidia to tackle the issues of price and energy, especially as its customer base remains very concentrated. “Nvidia sells chips to around ten customers who are mega global tech companies. This is not the same as Apple which sells to billions of individuals or the other Gafam which have a huge number of end users,” added Jean -Edwin Rhea.
The glass ceiling reached on the stock market?
Ultimately, Nvidia’s presentation hardly moved the market. On the contrary. Nvidia shares gained barely 0.7% on Monday and dropped 3% this Tuesday at the start of the session on Wall Street.
Given the anticipation generated by the event and Nvidia’s stock market performance (the stock has gained 78% since the start of the year), profit-taking seemed inevitable. Considering these two elements, “It was therefore difficult to impress investors with the details of the presentation,” writes Bloomberg.
For his part, Daniel Newman, managing director of The Futurum Group, a consulting firm, told the Financial Times that the market reaction could indicate that the frenzy around Nvidia “has reached its peak.”
“The market is very forward-looking. It already sees the massive market share, the dominant position with the hyperscalers (the heavyweights of the cloud, like Azure or Amazon Web Services, Editor’s note), the real lead that it has on the market thanks to [logiciel] Cuda” from Nvidia, he adds.
The continuation of Nvidia’s stock market rally will depend above all on its ability to sustain its breakneck growth. For the first quarter of its current fiscal year, the company anticipates revenues of $24 billion, which would represent growth of 233% compared to the same period of the previous fiscal year.
Bank of America nevertheless reiterated its buy advice on the stock on Tuesday, noting that the announcements turned out to be “in line with expectations”.
“With Blackwell, Nvidia should not only consolidate its lead in the field of LLM training, but also take a big step forward in AI inference,” explains the American bank.
“Specifically, Blackwell is capable of four times faster training than H100, and up to 30 times faster inference thanks to its new second generation transformation engine,” adds the establishment.
Julien Marion – ©2024 BFM Bourse