November 30, 2022. ChatGPT introduces generative artificial intelligence to the general public with a bang. One million users in five days, more than 100 million in less than two months: the general public has adopted OpenAI’s chatbot at an unprecedented rate in the history of tech. In turn, ChatGPT gave the start of a frantic race towards a colossal new market. Obviously, the current tech giants are in the starting blocks.
For Microsoft, Google, Amazon and Meta, generative AI is a major strategic issue, which justifies investing tens of billions of dollars in the technology. And for good reason: generative AI could disrupt all sectors of the economy by integrating with the software that businesses need. The most optimistic do not hesitate to compare its transformative potential to that of the Internet… Since the start of the year, not a week passes without the announcement of a software offering powered by generative AI. Thanks to the precursor ChatGPT, OpenAI stays ahead of the game. But its success is also – and above all? – that of Microsoft, its privileged partner. In 2019, the tech giant invested a billion dollars to cover its staggering research expenses and provide it with the necessary computing power.
First place bonus for Microsoft
Following the success of ChatGPT, Bill Gates’ company once again invested $10 billion in January 2023. With this tidy sum, Microsoft granted itself exclusivity for the indirect sale of OpenAI artificial intelligence. In other words, if a company wants to deploy ChatGPT, it only has two choices: go directly through OpenAI – whose servers are at Microsoft – or go through the Microsoft Azure cloud platform, which provides access to other complementary services and much simpler deployment.
“ Today, the vast majority of generative AI use cases deployed in companies rely on Micro- softAzure », Notes Ghislain de Pierrefeu, AI expert from the consultancy firm Wavestone. But the war is not lost for the others: “ Since the start of the school year in September, we have seen more and more companies going to Google Cloud and Amazon Web Services, or deploying open source models like those of Meta on their own infrastructures..
Google is struggling to keep up
If Microsoft is benefiting so much from its lead, it is because the dazzling success of ChatGPT has taken the entire artificial intelligence sector by surprise, starting with the most recognized company on the subject, Google. With two world-class research divisions (Google Brain and Google DeepMind, which have since merged, editor’s note), the latter attacks the Microsoft-OpenAI duo head-on. The Mountain View giant has increased the announcements of new AI models – such as Bard, a competitor to ChatGPT – and new features – like Duet AI, an assistant for its Google Workspace office suite. But it suffers from chronic lag compared to OpenAI. Google could hope to return to the lead thanks to Gemini, a new generative AI model supposedly more efficient than GPT-4, the best OpenAI model which serves as a benchmark for the sector. But while it was expected for this fall, Gemini has still not been launched. And at the same time, OpenAI presented more powerful (so-called “turbo”) versions of GPT-4 and ChatGPT and rolled out a list of new business services. One way to drive the point home.
Billions of dollars to enter the race
Amazon Web Service (AWS), the digital arm of the e-commerce giant, is lagging behind even more. Since it does not have its own large generative AI models, at the end of September it invested more than four billion dollars in Anthropic, the largest startup competing with OpenAI. With an obvious objective: to reproduce the success of Microsoft. But in the process, Google also invested two billion dollars in the young company, thus creating a strange three-way marriage.
For his part, Meta has chosen to avoid direct confrontation with ChatGPT, and holds his own racing lane. Facebook’s parent company offers its generative AI models – such as Llama-2 – as open source, that is to say freely accessible on the Internet. It even authorizes their free use for businesses.
Finally, a small group of startups is trying to keep up with the infernal pace of the tech giants thanks to significant fundraising, like Inflection AI (1.3 billion raised) and Cohere (445 million dollars). . But they can be counted on the fingers of one hand.
“Training very large language models like GPT-4 comes with an extremely high infrastructure cost. It’s a game that very few companies can participate in », recalls Nicolas de Bellefonds, associate director at Boston Consulting Group.
The Nvidia processors needed to train AI are being sold for several hundred thousand dollars, and the world’s best specialists are being poached at high prices. In this very American-centric financial race, only one European is managing to keep up for the moment: the German Aleph Alpha, which raised 500 million euros at the beginning of the month. But will this sum, immense for the Old Continent, be enough to sit at the table of giants?