Rare prestige, its name alone evokes a technology. OpenAI, the father of ChatGPT, was the first to demonstrate the full power of generative artificial intelligence to the world. But is it really taking advantage of it? According to the media The Informationthe Californian start-up could lose 5 billion dollars in 2024. A colossal sum despite its popular success – 180 million active users – and individual subscriptions sold at a price of 20 dollars per month. This pioneer is not the only one to see red. The staff of Inflection AI, once its competitor, was absorbed by Microsoft, due to the lack of satisfactory economic prospects. Stability AI, in image generation, had to lay off staff. In recent weeks, more than 1,000 billion dollars of market capitalization has flown away across the Atlantic, a large part of which was in companies focused on this technology. Even Microsoft’s good results on Tuesday did not reassure investors. The Redmond firm, despite a 29% increase in its sales on Azure, its cloud service essential to the diffusion of AI, fell on the stock market by almost 7%, before recovering during the session.
There’s a whiff of overenthusiasm in the air. A bubble? In a sensational report, Barclays recently wrote that Wall Street analysts expect Big Tech companies to spend about $60 billion a year developing generative AI models through 2026, but they’ll only reap about $20 billion a year in revenue over the same period. Those investments—mostly physical investments in data centers, chips, or servers—would be enough to produce 12,000 tools equivalent to OpenAI’s ChatGPT, Barclays said.
An effort that seems excessive, while the monetization of these solutions is currently unclear. On the one hand, because with open source AI, often freely accessible, there is competition that rivals in relevance and precision with closed commercial solutions. And quite simply because the technology has not yet convinced its end users. “For many companies, the ROI (return on investment) is not yet very positive on their use cases of generative AI. And sometimes, even if they have good ideas, they are not ready to deploy them. Our customers deplore the lack of reliability of the technology, the excessively large volumes of models and insufficient specialization”, Chadi Hantouche, associate director at Wavestone, in charge of AI, emphasizes to L’Express. “Who needs a Ferrari to go get their bread?” The same question arises for the general public. Apart from ChatGPT, alone on its mountain, who really emerges in its wake?
The unknown of productivity gains
Since the famous “Dotcom” bubble, with the failure of a cohort of startups launched in the Internet at the turn of the 2000s, or the recent flop of the “metaverse”, Tech has been very wary of euphoria. This feeling has crystallized, in AI, with the rise of Nvidia, the king of GPUs – these chips specialized in computer calculations – whose market capitalization reached 3,000 billion dollars in June, surpassing Apple and Microsoft. The cloud services of Microsoft, Amazon or Google where AI models are distributed have also consolidated their domination and reaped juicy profits. Server manufacturers have filled their order books for several years. A small fraction of the ecosystem has therefore carved out the lion’s share. Elsewhere, for ChatGPT as for a myriad of start-ups – more than 200 already launched in France in the field – the wall of profitability remains insurmountable. Beyond the immaturity of the technology, this is explained by the enormous costs of the energy expended to run their programs.
At the same time, Daron Acemoglu, a researcher at the Massachusetts Institute of Technology (MIT), cooled the enticing prospects projected by colleagues, but also banks and consulting firms, regarding the productivity gains enabled by AI. They would only be around 0.5%, and within a ten-year horizon. The GDP would not benefit from it, contrary to what many specialists are saying, such as, in France, the duo Anne Bouverot and Philippe Aghion. The authors of a noted report on the benefits linked to generative AI were counting on gains ranging from 250 to 420 billion euros over the next ten years. “Given the direction and architecture of generative AI technology, truly transformative changes will not happen quickly and few, if any, of them will probably happen in the next ten years,” said Daron Acemoglu. A bit like in computing, whose benefits only materialized several decades after its birth. A certain optimism persists, despite everything. “In generative AI, the revolution promised a year and a half ago has not yet taken place, notes Chadi Hantouche. But it will happen.”
“The risk of underinvesting is greater”
“Productivity gains will materialize at different rates depending on the sector,” says Bastien Drut, director of strategy and economic research at CPR Asset Management, an asset management company. And in specialized sectors, such as drug research, IT development, sales force management tools, he lists, “adoption is much faster so the gains are already clear.”
More broadly, in AI, “demand is well above supply,” according to his colleague Wesley Lebeau, portfolio manager at CPR Asset Management. This is an observation based on the demand for chips, which is not weakening. Other indicators are also reassuring, such as Meta’s profitable strategy in open source AI, Apple’s arrival on this market or Nvidia’s release of a completely new ecosystem that should increase the available computing power, while reducing the associated energy costs. “There is always, at the very beginning of an innovation cycle, an implementation of the infrastructure that can be significant and very costly,” explains Wesley Lebeau. For AI, this requires in particular specialized data centers that will sometimes not see the light of day before 2027 or 2028. This will be the case in France for the Petit-Landau data center built by Microsoft, for more than 2 billion euros. “This does not call into question the medium or long term bet,” concludes the expert.
In any case, the figures behind the AI boom have no plans to change their strategy. “We are witnessing the birth and evolution of a technology that I believe is as important as electricity or the Internet. AI can be the foundation of a new industrial base that our country would be wise to embrace,” wrote Sam Altman, CEO of OpenAI, in an op-ed a few days ago in Washington Post. This prospect is priceless to him. For example, he plans to raise $7 trillion to create factories specializing in the manufacture of AI chips. His estimated losses for 2024 with ChatGPT clearly do not affect his ambitions.
“The risk of underinvesting is significantly greater than the risk of overinvesting for us,” Microsoft CEO Sundar Pichai said. In Big Tech, the fear of missing out, the fear of missing out in the original version, and the intense competition between its actors, always prevail over the rest. In a blog post published in mid-July, Sequoia venture capitalist David Cahn sees no reason for the pace of investment to slow. Even if it takes time to turn a profit. Big Tech has the money to spend. It may have no choice. “The cloud giants see AI as both a threat and an opportunity and don’t have the luxury of waiting to see how the technology evolves,” Cahn writes. “If you don’t acquire the land, power, and labor now, someone else will.”
.