Source: French to English Tester Published on: 2026-05-21
Source: The Conversation – France (in French)– By Mokhtar Bouzouina, Doctor in Management Sciences, University Paris Nanterre
Since the media coverage of ChatGPT, announcements about artificial intelligence have multiplied, with each digital giant developing its own solution. On the surface, it appears to be fierce competition between rival groups. The reality could be much more nuanced. The competitors may not be trying so much to dominate others as to become indispensable building blocks to their competitors’ successes.
At first glance, competition in generative artificial intelligence (AI) looks like a classic economic battle: Google against Meta, Meta against Anthropic, OpenAI against Google… This brief approach has the advantage of simplicity, but it also has the drawback of masking what makes this sector unique.
Competition in this market does not only take the form of direct confrontation and the race for innovation. It also involves the ability to make opponents dependent on oneself.
Heart of competition
For years, at least during the emergence of AI and the first years of its growth, the heart of the competition in this sector seemed obvious. In this market – to succeed – you had to have the bestAI assistant, that is a computer program designed to understand human requests, process them, and respond to them in a more or less natural manner. But this interpretation is becoming less sufficient today.
TheStanford AI Index2025indicates that nearly 90% of the so-called “current” AI assistant models of 2024 come from the private sector, compared to 60% in 2023, while academia remains the main source of research. And especially that the performance gaps between the best models are progressively narrowing: the gap between the first and the tenth has fallen from 11.9% to 5.4% in one year, while the top two are now separated by only 0.7 points.
In other words, when the models become technically closer, the only superiority of the model becomes less distinctive. From then on, the competitive advantage tends to shift: it goes less to the sole producer of the model and more to the one who controls thesupplementary assets(Adner, 2017,Teece, 1986).
Also to read:
Behind AI, an unsuspected ecosystem… of which Europe is (for the moment) too absent
Microsoft’s shrewd strategy
This is what makes Microsoft’s strategy particularly wise in this spirit. WithAzure, Microsoft does not only offer its own products or those of OpenAI, the platform also provides users with the modelsfrom Anthropic, Cohere, DeepSeek, HuggingFace, Mistral, Meta, or Nvidia. The strategic stake seems clear: Microsoft can win even if the prevailing model is not exclusively its own. Its power would then depend less on the superiority of a particular model than on its intermediary position.
Amazon follows a comparable logic withAmazon Bedrock, which allows access to models offered by Anthropic, DeepSeek, Meta, Mistral, OpenAI, Qwen, Stability AI, and other companies. Here again, the success of the strategy does not come solely from owning a model, but also from organizing access to several competing models. In such a system, an actor can monetize therivalryfrom his own opponents, that is to say to use the confrontation between his competitors as an opportunity to generate revenue. The market thus becomes less a product market than an entry point market.
This logic is not entirely new. It extends a mechanism already observed in other platform industries: app stores, marketplaces, or software ecosystems.
What seems more specific to generative AI, however, lies in the fact that competition is not only about controlling access to competing models, but also about mastering complementary assets that enable the conversion of their technical performance into solutions usable by customers: cloud infrastructures, computing power, specialized semiconductors, distribution channels, proprietary data, integration interfaces, and software ecosystems.
Become indispensable, even for one’s adversaries
Anthropic offers a good example of this unprecedented competition in this perspective. On one hand, the company competes directly with OpenAI and Google on AI assistant models. On the other hand, it relies on several technological layers that do not entirely belong to it.LeMaGitreported, on November 22, 2024, that the company usesAWS Trainium, ofTPUs designed by Google and Nvidia GPUs(“specialized chips” for artificial intelligence, designed to train and run models faster than traditional processors) in order to develop and implementClaude(an AI assistant, similar to ChatGPT, created by the company Anthropic).
This dependence on external infrastructures became clear in October 2025, whenAnthropic announcedthe extension of its use of Google Cloud technologies, including up to one million TPUs. At the same time,Anthropic has officially announced that it has signed a new agreement with Google and Broadcom regarding the production of “several gigawatts” of next-generation computing capacity (TPU) starting from 2027.
Being a competitor here does not mean emancipating oneself from others or distancing oneself from them; it most often implies learning to distribute one’s dependencies.
Oligopolies and interdependence
The relationship between Microsoft and OpenAI illustrates another variant of this logic. In theirjoint statement of February 27, 2026, the two firms emphasize that their partnership was designed to give them “space [to]… pursue new opportunities independently,” while continuing to collaborate. This sentence aptly summarizes the structure of the sector: proclaimed independence does not abolish interdependence; it overlays it. We are thus far from the quest for independence in oligopolistic markets, as the economist once pointed outAlain Cotta.
In generative AI, we are not either in acoopetitionin the sense ofBrandenburger and Nalebuff (1996), but within a competitive dynamic in which a firm seeks to increase its power not only by surpassing its rivals but also by becoming indispensable to their operation. It is therefore not only about preserving one’s autonomy against rivals, but about building positions such that this autonomy itself becomes relative, because it is mediated by assets controlled by others.
Tomorrow is not another day
The competitive dynamic related to AI is distinguished by another feature: it is increasingly a confrontation of “preemption” (Ian McMillan, 1983). The stakeholders are not only fighting to sell today; they are already preparing the conditions for their power tomorrow.
The Belgian economic dailyl’Échorevealed, a few days ago, that Broadcom has signed a long-term agreement with Google toco-develop and supply its AI chips until 2031. A few days later,Reuters indicated that Meta had strengthened its partnership with CoreWeaveby a new agreement of 21 billion dollars (more than 18 billion euros), adding to a previous contract of 14.2 billion (i.e., 12.2 billion euros), all running until December 2032. The stakes are far from insignificant; these agreements provide access to future cloud capacity and, for Meta, to futureNvidia Vera-Rubin “chips” via CoreWeave.
We therefore understand, from these examples, why the competition in generative AI does not really resemble a classic market war. In many sectors, victory would rely on the temporary or permanent elimination of the rival. Here, it often consists of becomingthe infrastructure, the intermediary or the providerwhich the opponent cannot do without easily. Power shifts, in this perspective, towards control of the “points of passage”: access to users, computing capacity, hosting, integration platforms, the “chips.”
And the regulators?
This market structure simultaneously explains why competition authorities now look beyond the economic models of each individual company.On March 24, 2026, Reuters reported that Teresa Ribera, European Commissioner for Competition, expanded its vigilance to the entire AI “stack”: the economic models themselves, thetraining data(information used to train a machine modellearning(to make predictions, to recognize patterns, or to generate content) and the cloud infrastructure.
This shift in focus is ultimately logical. If market power now lies in organized dependencies, it is no longer sufficient to simply observe who produces the best AI assistant. One must necessarily look at who controls the sector’s essential choke points. The right question may no longer be to identify who will ultimately win the AI race. The most accurate question would be: who will succeed in making themselves indispensable to others?
The authors do not work for, do not advise, do not own shares in, and do not receive funds from any organization that could benefit from this article, and have declared no affiliations other than their research institution.
–ref. Competition in generative intelligence is not a classic market war –https://theconversation.com/the-competition-in-generative-intelligence-is-not-a-classic-market-war-281441