Winter Olympics in the Cloud: Alibaba Turns Milano Cortina into an AI Test Lab

Contents

When the Olympic Winter Games in Milano Cortina officially open on Friday evening, the most important competition for the IT industry will take place behind the scenes. It's about the authority to interpret what we see – and how it's processed. Alibaba Cloud, a firm partner of the International Olympic Committee (IOC) since 2017, has massively expanded its presence this year. While in the past it was primarily about hosting and replacing expensive satellite links, generative AI is moving to the center stage in 2026. For the first time, large language models (LLMs) are set to dominate the entire workflow from production to archiving.

The heart of this development is Alibaba's Qwen AI model family. The technology is intended to achieve what human editorial teams can hardly manage with the flood of video material: creating metadata in real-time. The "Automatic Media Description" (AMD) system scans the feeds, recognizes athletes and key moments, and writes the corresponding descriptions almost simultaneously.

What sounds like a huge increase in efficiency for broadcasters also marks the departure from purely human interpretation of sporting events. When algorithms decide which scenes are marked as "highlights," the IOC delegates the editorial pre-selection to the software of a Chinese tech giant.

Why Qwen instead of ChatGPT or Gemini?

The fact that Qwen is setting the tone at the Games in Italy, rather than OpenAI with GPT-4 or Google with Gemini, is not a technical coincidence but the result of hard-nosed sports politics. Alibaba secured a place in the exclusive sponsor's Olympus as early as 2017 as part of the Olympic Partner Program. The exclusive contract runs until 2028 and has an estimated volume of over 800 million US dollars.

Through these long-term contracts, the IOC is bound to its partners' roadmaps. This has consequences: Alibaba not only provides the computing power. By integrating into the official platforms, the company also dictates which AI standards apply to broadcasters worldwide. A switch to potentially more powerful models is legally and economically hardly feasible within the contract period.

Digital Aesthetics: When Sport Becomes a Video Game

AI will also become more visible to viewers on screen. In 17 sports, including ice hockey and figure skating, advanced "360-degree replay systems" will be used. Their goal is the most complete spatial reconstruction possible: within less than 20 seconds, the AI calculates a 3D model of the scene from various camera perspectives. New are the so-called "Spacetime Slices," which freeze motion sequences stroboscopically in a single image. Technically, this is impressive, as the algorithms must precisely separate athletes from complex backgrounds such as swirling snow or reflective ice.

With this initiative, the IOC is trying to transform sport into an aesthetic that is otherwise more familiar from high-end animations. But this perfection comes at a price: the boundary between real video material and AI-generated reconstruction becomes more permeable. In an era of increasing deepfakes and AI manipulation, trust in the "real" image is a valuable asset that is being put at risk here for the sake of show effect.

From Punch Card to Cloud

Today's cloud dependency is the final stage of a development that began decades ago. In the era of IBM – an IOC partner from 1960 to 2000 – the focus was primarily on pure data processing. Results were laboriously entered into mainframes back then. The disaster in Atlanta in 1996, when IBM's info system collapsed under the load and provided journalists with incorrect results, remained legendary. Thereafter, Atos took over and focused for decades on system integration and cybersecurity, with data centers operated physically on-site or in regional hubs.

The radical break occurred with Tokyo 2020 and Beijing 2022. Alibaba moved the core systems entirely into the cloud for the first time. While IT was previously a supporting infrastructure, it is now the central nervous system. While this centralization saves costs, it also makes the global event more vulnerable to geopolitical tensions and technological errors.

The Digital Olympic Memory

The ambitions go far beyond the live broadcast. The IOC is also entrusting Alibaba with its "memory." In the Olympic Museum in Lausanne and in internal archives, Qwen is being used as a digital librarian. Over eight petabytes of historical material are to be made searchable through AI tagging and natural language queries. What superficially appears to be a practical tool for historians actually means the complete algorithmic exploration of Olympic history.

The "Sports AI" system also automates the categorization of decades of video material. This raises questions: If an AI decides which emotional reactions are "relevant," a filtered view of the past emerges. Thus, a technological smoothing could take place, where only the image material that is optimized for the cloud and easily consumable is brought to the surface.

The economic logic behind this is simple: the IOC wants to reduce the costs of broadcasting. The "Live Cloud" is now the standard for distribution. In 2026, 39 broadcasters will access over 400 feeds without having to book expensive satellite capacity. This lowers the entry barriers for smaller broadcasters but increases dependence on a single infrastructure. Should Alibaba's cloud infrastructure falter, screens could go black for dozens of rights holders simultaneously.

(vbr)