We put the same question to six AI assistants: who will dominate artificial intelligence in 2036? Five put the United States in front. The sixth, Gemini, describes a duopoly between the United States and China. None picks Europe.
The favorite is therefore close to a consensus, but that consensus calls for caution: these assistants may have drawn on common sources or narratives, and we did not measure that overlap. Their agreement tells us what they find plausible, not what will happen. What separates them is more instructive: the scale at which they look at power (companies, the industrial ecosystem or states) and what, in their view, could make the favorite lose.
The six bets at a glance
| AI | Its bet for 2036 | What could make it fail |
|---|---|---|
| Claude | American leadership through a few companies, built on computing power | An American policy mistake that loosens export controls |
| Gemini | US–China duopoly, under the sway of a few giants | Energy, technical or geopolitical disruptions |
| Grok | American oligopoly in a world divided into two AI blocs | Taiwan, Chinese chips, American policy |
| Mistral | An oligopoly of three to five American companies | A Chinese leap in chips before 2029–2030 |
| ChatGPT | American dominance, with state-backed companies and allies | Chinese chips, Taiwan, energy, open AI |
| Perplexity | United States, through an oligopoly; opens on an unequal duopoly | An innovation that depends less on computing power |
Three bolder bets
AI labs would not stay independent. For Mistral, "decision-making power belongs to the companies, not to Washington," and OpenAI, Anthropic and their peers will be "absorbed, tied to or subordinated to the hyperscalers" that fund their computing. It sums up: "I bet on the oligopoly." Its starting point is private capital with no public equivalent. The 2024 figures point that way: according to Stanford's 2025 AI Index, the United States attracted $109.1 billion in private AI investment, China $9.3 billion. That figure leaves out public money, which China mobilizes differently. Mistral is also the only one to put numbers on four scenarios, using its own estimates: American oligopoly about 50%, fragmented oligopoly after a financial correction about 25%, Chinese catch-up about 15%, radical break about 10%.
Computing power could stop being decisive. Claude makes "the bet that computing power will remain the main bottleneck through 2036," and it is the cornerstone of its scenario. It relies on an advantage that can be measured today: according to Epoch AI (May 2025), the United States hosts about three quarters of the GPU-cluster performance the institute tracks, China 15%. But only the installations Epoch tracks are counted, and installed capacity says nothing about how it is used. Perplexity, for its part, names as a central uncertainty an innovation that depends less on computing power and would redistribute power. Both answers rest on the same question: will today's advantage still be tomorrow's?
To dominate would be to be hard to replace. ChatGPT qualifies the idea of the best model: the winner will be "not necessarily the one who will invent the most brilliant artificial intelligence." It adds: "It will be the one that can make its artificial intelligence the hardest to replace." It distinguishes five capacities (computing, science, diffusion, standards, strategic power), describes American companies "backed by the state" and supported by a network of allies, and presents its 2036 ranking as "my extrapolation, not a demonstrated result."
Where they part ways
Companies, ecosystem or states? The answers do not describe three opposing futures but three levels of analysis that can coexist: a few companies dominate the models, they are backed by their state, and they operate in a confrontation between two poles. Mistral and Perplexity ("an American oligopoly, not an American state") look first at companies; ChatGPT, at the ecosystem of companies and allies; Gemini, at the two state poles. Grok layers the levels: an oligopoly of American companies in a world divided into two AI blocs. One question the corpus does not settle: if these levels came into conflict, who would have the last word?
Is the favorite as clear for Gemini? Gemini is the only one that does not give the United States first place as plainly. It describes "a hybrid hegemony shared between a state duopoly (United States and China)" and large companies, with a China that is "the absolute master of industrial AI, of embodied robotics." That is the most categorical formula in the corpus, and the answer does not demonstrate it: it covers several fields at once without precise evidence. The closest figure comes from the International Federation of Robotics: in 2024, China installed 295,000 industrial robots, 54% of the world total, against 34,200 in the United States. It supports the idea of Chinese industrial strength, not of a lead in AI: these are industrial robots, not machines driven by recent AI.
Which risk weighs most? The answers often cite Taiwan, energy and Chinese chips, but not in the same way. Claude judges that the main risk to its thesis is not a spectacular Chinese catch-up but an American policy mistake. Mistral calls a Chinese leap in chips "the most destabilizing scenario." Grok adds that ultra-fast Chinese industrial adoption, combined with American energy constraints, would shift the advantage toward deployment rather than the frontier of models.
Europe: a place that depends on how victory is defined
None of the six models puts Europe in first place in 2036. Most give it a role in rules, standards or markets; these are scenarios, not findings. Claude speaks of "a power of standards and a usage market," Perplexity of a regulatory power, "not a frontier power." Gemini is the harshest: a Europe "on the way to technological decline." Grok and Mistral mention niches or specialization; ChatGPT and Perplexity an industrial role, Perplexity on condition that its sovereign-computing projects succeed.
That place follows first from a definition: in several answers, "to dominate" amounts to controlling the largest models and computing power. If that definition is accepted, an unfavorable verdict for Europe is consistent. But it does not settle the question of industrial uses, services or economic benefits. The cloud shows today's dependence: according to Synergy Research (2024), European providers hold about 15% of the European cloud infrastructure market, the three American giants about 70%. That measures today's cloud, not models, and predicts nothing for 2036.
What this comparison reveals
The American bet rests, in several answers, on a single assumption: the resources that give an advantage today will remain decisive in ten years. Their break scenarios challenge precisely that continuity, whether in computing power, chips or policy choices. The favorite is therefore clear; how long its advantage will last is far less so.
For Europe, these answers mostly give it a secondary place in producing the most powerful systems. They say much less about the possibility of succeeding elsewhere: in uses, services and benefits for the population. That is not proof that it will succeed there. It is a question this first exercise leaves open.
Method
These are forecasts produced by AI assistants, not established facts. That several assistants agree does not show that a forecast is correct. The full answers are not published.
Common prompt (original, in French). All six received the same text:
Qui dominera le monde de l'intelligence artificielle en 2036 ?
Défends une thèse claire : un pays, un bloc, quelques entreprises, ou une autre configuration si tu considères cette lecture géopolitique comme trompeuse.
Explique ce que signifie « dominer », les raisons de ton choix et les événements qui pourraient invalider ton scénario. Précise notamment la place que tu prévois pour les États-Unis, la Chine et l'Europe.
Distingue les faits sourcés de tes prévisions. Ne cherche pas à présenter tous les scénarios comme également probables : donne ton verdict.
Réponse : 700 à 1 000 mots.
English translation (ours, for reading only). Who will dominate the world of artificial intelligence in 2036? Defend a clear thesis: a country, a bloc, a few companies, or another configuration if you consider this geopolitical reading misleading. Explain what "dominate" means, your reasons, and the events that could invalidate your scenario. In particular, state the place you expect for the United States, China and Europe. Separate sourced facts from your forecasts. Do not present all scenarios as equally likely: give your verdict. Answer: 700 to 1,000 words.
Collection conditions. All answers were collected on 1 November 2026 through each assistant's web interface, using personal accounts. Each assistant answered once. "Not confirmed" means we did not record the information with enough certainty to state it.
| Assistant | Model or mode shown | Web access |
|---|---|---|
| Claude | Haiku 5.5, medium effort | Not confirmed |
| Gemini | Not confirmed | Not confirmed |
| Grok | Fast mode | Not confirmed |
| Mistral | Fast mode; underlying model not confirmed | Not confirmed |
| ChatGPT | Not confirmed | Not confirmed |
| Perplexity | Search mode | Not confirmed |
Differences between runs. Modes and models differ across assistants, web access may have differed, and no temperature was fixed, so the runs cannot be reproduced exactly. Actual lengths run from about 620 to 1,400 words, against the 700 to 1,000 requested. The quotations in this article are our English translations of the French answers. There is no Chinese model in this panel: that is a limitation. This is neither a scientific study nor an exhaustive survey, but a forward-looking exercise.
Reading notes. "Claude predicts" marks a model's opinion; outside figures are attributed to their source, with its year and scope; the last section, "What this comparison reveals," is the editors' reading.