Since the dazzling success of ChatGPT at the end of 2022, a handful of companies have been racing towards generative artificial intelligence. Their goal: to become the first to develop general AI – or superintelligence – with capabilities greater than the human brain, and thus secure the key to a market with immense potential. To fuel this race, the financing tap opened wide: billions of dollars have flowed from 2023 on a dozen start-ups, the oldest of which, OpenAI, was only created in 2015.
Two years after ChatGPT, the sector is already consolidating. The race is expensive, very expensive: creators of AI models need ever more Nvidia processors. However, they are worth several tens of thousands of dollars per unit, and are essential for improving performance and training models. At the same time, companies must take out the checkbook to wrest the best scientists and engineers from the clutches of competition.
Result: beyond the technological issues, the race for AI is turning into a race for financing. And in this game where the United States excels, a leading trio stands out, in the shadow of the tech giants. The financial gap is such that an existential question arises for other companies, unable to keep up.
Three overfunded start-ups in the lead
- OpenAI, the race leader (around $20 billion raised)
In early October, the creator of ChatGPT raised $6.6 billion, bringing its valuation to $157 billion. Until now, the star AI start-up received its funding almost exclusively from Microsoft, but the new round of funding has brought several companies and investment funds into its capital. At the head is Thrive Capital, whose founder Joshua Kushner is close to the new President of the United States, Donald Trump. The operation is good news for OpenAI, whose codependence with Microsoft attracted the attention of the regulator, in addition to questions about its strategy.
Even if the competition continues to get closer, OpenAI remains ahead at all levels, whether in terms of the performance of its models or in the diversity of the functionalities of its tools. ChatGPT, the start-up's flagship, can today write, code, generate images, serve as a voice assistant or even serve as a search engine. OpenAI also launched the first reasoning model, called o1, supposed to open new development perspectives for the technology, a first step towards superintelligence.
Sam Altman, all-powerful at the head of OpenAI
- Anthropic, the challenger (around $11 billion raised)
Founded by five OpenAI alumni who disagreed with their management, Anthropic quickly established itself as the main competitor to the creator of ChatGPT. This reputation and the performance of its Claude model allowed it to obtain financial support from Amazon. The latter has injected 8 billion dollars since 2023, including 4 billion last week.
Less advanced than its competitor, Anthropic is however catching up, and remains the only one to get closer to the leader in the sector in terms of available resources. The battle between the OpenAI-Microsoft duo and the Anthropic-Amazon duo has only just begun.
Amazon and Anthropic, an open marriage to conquer the artificial intelligence market
- xAI, Elon Musk's rocket (6 billion dollars raised, soon 11 billion)
In May 2024, less than a year after its launch, xAI raised $6 billion, but it's not finished. The best-informed American media are already announcing an upcoming fundraising of an additional 5 billion dollars in the coming weeks. Founded by Elon Musk, financed by the cream of American venture capital and currently backed by the social network X (formerly Twitter), the start-up is taking a different path from its competitors.
A riskier path, where speed of execution prevails over prudence, and where political issues occupy a predominant place. The sulphurous billionaire sees his company as anti-OpenAI, which he considers too “woke” (progressive, editor’s note). Its place in the future government of Donald Trump, and its potential influence on the regulation of the sector can only be good news for xAI.
xAI, Elon Musk's monstrous project in artificial intelligence
Tech giants at the center of the game
- Google, the unofficial number two
Of the three cloud giants, only Google has – from the start – participated in the AI race on its own. To do this, it relied in particular on the expertise of its teams from its DeepMind research laboratory. Financed by the group's immense liquidity ($74 billion in profits in 2023), the Gemini model unveiled in 2023, a competitor to ChatGPT, immediately placed itself in the top places in the performance rankings.
More generally, Google remains to this day the only company to offer a competitor to each OpenAI product, with the exception of the o1 reasoning model. The Californian has also invested more than 2 billion dollars in Anthropic.
Artificial intelligence: Google asserts itself as the number one threat to OpenAI
- Amazon and Microsoft, the financiers of the race
The numbers 1 and 2 in the cloud initially preferred to finance the AI race rather than participate in it themselves. Microsoft took action in 2019 by focusing on OpenAI. For its part, Amazon waited until 2023 to position itself on Anthropic. And then despite investments exceeding ten billion dollars, the two technological giants also decided in 2024 to create their own internal divisions.
By the same process of “ to acquire » (acquisition by recruitment, editor's note), the two giants got their hands on the teams of Inflection AI (for Microsoft) and Adept AI (for Amazon), two companies with satisfactory technological results but unable to keep up with the pace set by OpenAI. But for the moment, these internalized teams are not directly participating in the race for superintelligence.
Tech giants have a new method to vampirize AI nuggets
- Meta, the merciless broom car
Historically, Meta is one of the strongholds of AI. But when the race started, the company chose to publish its models, called Llama, as open source. In other words, it put them into the hands of the market for free, where its competitors charge for the use of the models. With a terrible consequence: if an AI does not exceed the – very high – performance of Llama, its creator will have difficulty charging potential customers.
Like Google, Meta can draw on the group's tens of billions in profits to finance its efforts. Mark Zuckerberg also announced at the beginning of the year his desire to provide the company with a record number of Nvidia processors. With the danger, for the rest of the ecosystem, that the broom car takes the lead and crushes everyone in its path.
With Llama 3.1, Meta wants to shake up the AI ecosystem
Challengers in a delicate position
Behind the new AI giants, the equation becomes more complicated. The French start-up Mistral broke Europe's AI funding records one by one. A year and a half after its creation, it has raised a total of more than a billion dollars in funding. But the nugget, which prides itself on being the most “ capital efficient » of the market (that is to say the most efficient, with equal means) finds itself caught in a vice.
Ten times less funded than its main competitors, the start-up is struggling to keep up. And the nuggets of its size have either been vampirized by tech giants (Inflection AI, Adapt AI, Character AI), or have pivoted their economic model, like Cohere or the other European hope, Aleph Alpha. Only Safe Superintelligence, recently launched by OpenAI's former chief scientist, is in the same position.
Why German hopeful Aleph Alpha threw in the towel in the AI race
In other words, for Mistral, three trajectories are emerging. Either the start-up manages to put itself at the financial level of the leader, or it is bought or vampirized by a tech giant, or it will have to carve out a technological niche, synonymous with stopping the race for superintelligence. Moreover, if it has not abandoned its ambitions, the young company has already started to intensify its efforts in the design of very small models, which could be a saving path for its sustainability.
