An AI model joins the Olympic Games
Days before the Milano Cortina Winter Olympics opened, the International Olympic Committee announced that its official Olympic model is built on a Chinese large language model, and the Games that began on 6 February are using it across scheduling, translation and broadcast services. It is the first time a foundation model has been embedded in the core running of the Games.
On 5 February, two days before the opening ceremony, the International Olympic Committee announced that the Milano Cortina Winter Games would be run with an official Olympic large language model, built on the Qwen foundation model developed by Alibaba. The Games that opened on 6 February are the first in which a foundation model sits inside the operational core of the event: scheduling and resource coordination, multilingual assistance for athletes and delegations, and the flow of information to spectators all pass through a system trained on language rather than written as fixed rules. The announcement was made by the IOC president at the International Broadcast Centre, and the framing was deliberate: this is not a side application but a piece of infrastructure.
The scale of the deployment is the story. Olympic operations have always been information-intensive, with tens of thousands of accredited participants, dozens of languages and venues spread across a mountainous region, and a broadcast operation reaching a global audience. Previous Games have used machine translation and algorithmic scheduling as discrete tools. What is different here is that a single model is the connective tissue, understanding queries in many languages, coordinating between systems that previously did not talk to each other, and improving as the Games generate more data. The Olympic model is, in effect, the Games’ memory and switchboard in one.
The choice of a Chinese model is also a reminder of where foundation-model capacity now sits. A decade ago, the technology behind such a system would have come from an American laboratory as a matter of course; the fact that the IOC selected a model from Alibaba, and that the selection was uncontroversial enough to be announced in a routine press event, reflects how the field has internationalised. For the model’s developers, the deployment is a public test at maximum visibility: if the system holds up across seventeen days of peak load, the demonstration is worth more than any benchmark.
There is a quieter consequence for the spectators. The broadcast innovations around the Games, from AI-assisted replay systems that break down movement into clear sequences to real-time visualisations that explain complex sports to casual viewers, depend on the same infrastructure. The athlete sees the model as a scheduler and translator; the viewer sees it as the reason a biathlon race becomes legible on screen. The technology is doing its best work when it is invisible, which is to say when the Games simply run smoothly and no one has to think about the model at all.
The experiment is not risk-free, and it deserves scrutiny as much as praise. A foundation model that coordinates a global event raises questions about errors, bias and accountability that a traditional software system also raises, but with less predictability in the answers. The organisers have framed the deployment as carefully managed, and the Games themselves will be the review. But the direction is clear: the Olympics, an institution almost as old as the modern era of international sport, has chosen to run on language models, and if this fortnight succeeds, no future Games will be organised the old way again.