According to an estimate by SemiAnalysis, ChatGPT, OpenAI's superstar interface launched in November, gobbles up about $700,000 a day in operating costs.
According to the website The Information, Sam Altman, the boss of OpenAI, which lost $ 540 million last year, has evoked fundraising reaching $ 100 billion in the coming years to finance the development of the young Californian firm.
"We're going to be the most capital-intensive startup in Silicon Valley history," the executive said recently during a panel discussion.
Google, Microsoft, with the help of OpenAI or Meta, have already invested billions to build their own generative artificial intelligence (AI) interfaces, i.e. capable of generating content on demand in everyday language.
"People don't realize that AI like ChatGPT requires enormous computing power," says Jack Gold, an independent analyst. "How many companies can afford to buy 10,000 H100 units from Nvidia?" - graphics processors (GPUs) in high demand for AI, whose price hovers around $ 30,000 each - he asks.
A world is emerging more and more clearly in which only a handful of companies have the financial strength to build, from scratch, a generative artificial intelligence model that can compete with existing interfaces.
ChatGPT, OpenAI's superstar interface launched in November, gobbles up about $700,000 a day in running © costs Marco BERTORELLO / AFP/Archives
The others have no choice but to use the technology and capabilities of these giants, as they already do for remote computing (cloud), which has become a rent for Microsoft, Google or Amazon.
AI and the cloud are also coming together and are likely to increase the dependence of companies, whose remote computing needs will increase to create variations adapted to their needs of large generative artificial intelligence models.
"Goose with golden eggs"
The cost of using the cloud is already "a very underestimated problem by many companies" and it could be further amplified with generative AI, says Stefan Sigg, product manager at software company Software AG.
According to Microsoft's annual report, the group's margins in the cloud reached 70% last year, while Amazon, the other cloud leader, generated $22 billion in operating profit through its dedicated subsidiary AWS (Amazon Web Services).
OpenAI boss Sam Altman, February 7, 2023 in Redmond, Washington © Jason Redmond / AFP/Archives
"Cloud monetization with Azure is a goose that lays golden eggs" for Microsoft, said Dan Ives of Wedbush Securities. "We are talking about 20, 30 or even 40 billion dollars per year in the long term if the bet on AI pays off."
For the analyst, Microsoft CEO Satya Nadella has a grace period of six to nine months before having to demonstrate that the strategic priority to generative AI translates into the group's profits.
"We're going to charge for these new AI opportunities," Microsoft Chief Financial Officer Amy Hood said last month, "and ultimately grow operating profit" (earnings before interest and taxes).
"Training (AI models) and ChatGPT will become a very important cloud service in the future," said Tenry Fu, general manager of Spectro Cloud, which specializes in remote computing optimization.
But after this development phase adapted to specific needs, "the company will have its own model" and its dependence on cloud essentials should decrease, he says.
U.S. regulators are closely monitoring the structuring of this still nascent market.
"As regulators, we must ensure that these opportunities for new entrants are not destroyed by the giants of the sector," the president of the US Competition Authority, Lina Khan, recently argued on CNBC.
The multiplication of AI © fields Valentin RAKOVSKY / AFP / Archives
"It is absolutely true that the number of companies that will be able to develop the most innovative models is going to be limited, simply because of the resources required," Altman admitted at a congressional hearing on Tuesday.
"Therefore," he added, "it is necessary that we and our competitors be subject to unprecedented surveillance."
© 2023 AFP