Satya Nadella could rest on his laurels. After all, since he took office in 2014, Microsoft’s value has increased tenfold – it now stands at $2.355 billion – a performance that puts him on par with Tim Cook, Apple’s boss. But this engineer born in Hyderabad, India, who joined the company in 1992 and who obtained his MBA by correspondence, now wants to carry out a decisive transformation to make Microsoft the main player in so-called generative artificial intelligence (AI). , those of large language models (LLM) stuffed with billions of documents and capable of dialoguing as well as generating prose.
Microsoft’s offensive – along with the $40 billion in venture capital that has poured into the sector – is driving intense competition in the AI space. This fall, the Redmond, Washington-based firm was keen to overtake Google, whose Bard model is constantly gaining in performance and which is preparing a powerful new LLM called Gemini, expected before the end of the year. Less than a year ago, Satya Nadella made a decisive choice by sealing an alliance with OpenAI, the company creating ChatGPT, for an investment of 13 billion dollars. Since then, all of the brand’s office products as well as its Windows system installed on 1.6 billion PCs worldwide are in the process of being irrigated by adaptations of GPT-4, with the introduction of a spectacular series of personal assistants brought together under the name Copilot.
For its CEO, the success of this strategy will be measured over time, as was the case for the cloud. “If in 2010 you had asked me what the impact of the cloud would be [NDLR : sur notre revenu], I would have said ‘wait ten years’. Today, the diffusion of all this is much faster. It will take three years, four maybe… Our advantage comes from the fact that these large language models could not have been trained without a cloud like ours.”
Microsoft’s enormous installed base of software – 345 million paying users for the Office 365 suite – gives it considerable leverage. If just 10% of them opted for the Copilot AI module, which will be billed at $30 per month, this would translate into $12 billion in additional revenue in a single year, with Microsoft having made $198 billion in revenue. business in 2022. The idea, naturally, is to convert most of the customers in a few years.
An ever more dominant Big Tech
Satya Nadella does not hide the fact that the advent of generative AI risks further strengthening the overwhelming domination of Big Tech and the position of Microsoft. “That’s obviously our intention,” he said. On a more general level, the scale of investments will be a determining factor. “Simply,” he continues, “because of the computing power that is necessary, all ‘hyperscalars’ [Amazon, Google, Microsoft] have a fundamental advantage to be leaders in the era of artificial intelligence.”
The CEO of Microsoft also suggests that industrial policies outside the United States are not necessarily well adjusted. “The main thing is not to envy what others have, but rather to see what it is possible to build with what exists. Studies on the industrial revolution, particularly in Europe, reveal that countries that have made the most progress are those who were able to capitalize on what existed and had been imported by others, while those who sought to reinvent the wheel were left behind.” For him, in Europe, competition in tech will focus on the ability to capitalize on infrastructure imported from the United States. Enough to make the proponents of digital sovereignty faint.
“If we think about it carefully,” he insists, “for a French software developer, launching a new product [d’IA] is inexpensive and virtually frictionless thanks to the investments we have made in France. This is what the competition is all about. If there’s one thing that the history of tech has taught us, it’s to never take anything for granted. It is certain that there will always be big groups, but that we will also see newcomers emerge. Microsoft is a multinational with very heavy capital investments in France, Germany, the United Kingdom and all Asian countries. With our cloud, we are no longer just a software company, we have an industrial dimension. These investments aim to deploy locally [notre écosystème] a bit like a manufacturer who sets up factories around the world. So, for example, I want to bring these technologies to France so that great tech companies can develop there. However, what they need to develop generative AI is, above all, powerful cloud computing, with the best available AI models on which they can innovate. We already have significant infrastructure [en France] and it will continue to grow.”
Precaution versus competition
Like its direct competitors, Microsoft must manage the delicate subject of the security of these super-powerful AIs, where competition risks coming at the expense of caution. “First of all, I am delighted that all these risks are now part of the conversation. It’s a bit as if, at the start of the industrial revolution, companies had taken into account pollution, the impact on the climate or child labor. That would have spared us two pretty terrible centuries… So, it’s very good that it’s this way and it’s true that it starts with actors like us.”
For the serious risks of AI spillover, the CEO of Microsoft calls for an increased role for public authorities. “On this, a real dialogue has begun in the United States, as in France, on how to proceed in the best way. This is an area where States can take leadership. But in the end, we will collectively must arrive at a system comparable to that of the IAEA [l’Agence internationale de l’énergie atomique]to ensure that humans remain in charge.”
The question of intellectual property
On the question of regulation, Nadella emphasizes that, unlike in the past, tech companies are taking the lead this time: “All the players, and not just Microsoft, are showing great maturity on the subject.” Here again, he defends a multilateral approach: “For everything related to security, governments will have to do their part, as will the non-profit sector and universities.” He regrets that there is no strong academic authority for AI, like CERN for fundamental physics.
Apart from the thorny question of security, Big Tech will be confronted with claims on intellectual property. Today, AI models are trained with whatever they can collect: articles, books, images, documents of any kind, whether protected or not.
When we ask Satya Nadella how he intends to respond to the demands for compensation that are coming, he does not shy away. “Everyone must accept that anyone who owns intellectual property is entitled to exercise their right.” A small precaution however, Nadella reminds that in the mass of data ingested by LLMs, we will have to take into account the fair usefair use, a much more extensive concept in the United States than in Europe.
“Justice will have to decide, he said. But it is fundamental to protect intellectual property and the income of rights holders. The exchange will have to be balanced. There will be different economic models to ensure this balance: in certain cases, these will be traffic referrals [vers la source, comme pour les liens d’un moteur de recherche], or advertising revenue sharing systems, or flat-rate licensing agreements. There will be no dominant model. And then, rights holders always have the possibility of avoiding the collection of LLMs. This opt-out is easy to implement. I think that current content providers will have to get used to a world where there will be great flexibility in the mechanisms, which will go from contributions [gratuites] as we do in open source, to paid licenses, including revenue sharing on certain AI products. These are choices that we will have to make collectively, because today we can no longer put the genie back in the bottle.”