AI at Scale: How the Olympics Stress Test Infrastructure

How AI Is Powering the 2026 Winter Olympics in Real Time

Large popular events like the Olympics used to be displayed exclusively through TV networks, with carefully placed TV crews capturing the moment and displaying it live around the world.

This way of capturing the Olympics on cameras is changing very quickly, with a mix of drones, AI, and cloud-based infrastructure radically changing the technological backbone of the event.

While this was already partially true for the 2024 Paris Olympics, this is even more true for the Milano Winter Olympics of 2026.

Multiple-angle cameras, a cloud-first approach, and powerful AI are building, almost in real time, a 3D representation of the competition in 17 different sports, and helping reduce the work required, the energy consumed, and the delays in global broadcasts.

Behind this technical revolution is Alibaba, the Chinese giant of e-commerce and cloud computing, with a solid position in AI as well.

AI Taking Over The Olympics

Cloud Broadcast

The Olympic Broadcasting Services (OBS) and the International Olympic Committee (IOC) have partnered with Alibaba to deploy cloud computing and AI at a scale never seen before for the 2026 Winter Olympics.

This is the latest step following previous deployments in Tokyo 2020, Beijing 2022, and Paris 2024 of cloud-based, AI-enabled broadcasting.

Since its introduction at Tokyo 2020, OBS Live Cloud has evolved from an optional service to a core distribution platform. At the Paris 2024 Games, it became the primary method for remote broadcast delivery when it enabled 400 live video streams and 3D rendering.

In Milano, the company deployed support for 39 broadcasters, delivering 428 live video feeds, including 26 in ultra-high definition streams, along with 72 audio feeds.

While select premium feeds and capture systems operate in ultra-high-resolution formats such as 8K and UHD, the broader broadcast mix includes multiple resolutions optimized for workflow efficiency and distribution requirements.

The cloud-based delivery has been replacing traditional satellite links and dedicated transmission lines to reduce cost, setup time, and technical complexity, while improving flexibility and resilience.

This also enables smaller broadcasters to access professional-grade broadcast capabilities without heavy upfront investment.

Besides live recording of the sports events themselves, 5,000+ short-form pieces, including behind-the-scenes footage, highlights, and emotional reactions, will be distributed through OBS Content+, a cloud-based platform powered by Alibaba Cloud.

Instant Understanding And Tagging

Alibaba’s Automatic Media Description (AMD) System is a system powered by the company’s AI Qwen, an advanced large language model (LLM).

The system automatically identifies athletes and key moments, generates event descriptions, and tags video assets within seconds, significantly reducing manual processing time.

The OBS team can also communicate with the AI in natural language to process the live videos more quickly. For example, a request for “find the figure skating gold medal performance” will retrieve the right information almost instantly.

This way, OBS personnel can find, develop, and distribute Olympic stories across platforms more easily.

Instant Intelligent Replays

The company’s Real-Time 360º Replay systems automatically create immersive replays with fluid camera movement and stroboscopic visual effects.

To do so, the AI algorithm separates athletes from complex backgrounds such as snow and ice, and then creates 3D reconstructions of key moments in as little as 15–20 seconds, fast enough for live broadcast use.

This combines with previously used effects like the BulletTime first introduced at Beijing 2022 to provide frame-freeze and slow-motion views, or the new Spacetime Slices capability, which visualizes multiple phases of an athlete’s movement in a single composite image, allowing viewers to better understand technique and performance.

AI Olympics Agents

This year, AI was not just deployed at the broadcast level, but also in the Olympic village and among the attending public. For example, Qwen was powering a series of AI agents called the “Olympic AI Assistants.”

One of the agents’ tasks was to provide multilingual conversational support and real-time event information, allowing fans to access official Olympic Games content through a chat-based interface.

The same technology will be permanently deployed at the Olympic Museum in Lausanne, where visitors will have access to personalized AI audio guides that enhance the museum experience.

Qwen was also deployed in the secure portal for National Olympic Committees (NOCs). There, it is used to locate documents, policies, and grant guidelines through natural language queries, with built-in multilingual translation support.

Qwen will also improve access to Olympic sport archives through “Sports AI,” a cloud-based media archiving solution that includes AI tagging, video search, and conversational search.

Maybe less impactful, but illustrative of how much the Olympics are becoming infused with “tech,” visitors could even move a robot arm by just moving their own hand and have an AI interpret the moves to give them an Olympic memorabilia.

Massive Data Volume

While Qwen is definitely making the most visible change in the way Olympic videos, images, and data are handled, another invisible task is the supporting infrastructure that makes it possible.

There is no less than eight petabytes of historical Olympic media that are now hosted on Alibaba’s cloud computing systems.

From AI Training To AI Inference

Besides the niche use case of broadcasting and archiving, the Milano 2026 Olympics reflect a massive shift in how AI should be considered.

Until now, the focus has been on AI training and what new ability it acquired in this latest iteration of the model.

This is shifting to AI inference, where the already trained model is now being deployed for narrower real use cases. For example, instant translation, or making a replay of athletes in less than 20 seconds, or making the massive Olympics archives much easier to search.

AI inference is not only more useful, but it is also a lot less compute and energy-intensive, as only a specific set of the AI ability is used at a time. This should help reduce the constraint of energy supply and hardware availability that has been limiting AI speed in the past few years.

How to Invest in AI Infrastructure Behind the Olympics

Alibaba

OpenAI and Anthropic, as well as most “big tech” US companies, are racing to make AGI (Artificial General Intelligence), and might succeed in doing so. But the Chinese AI industry is adopting a slightly different approach, already deploying AI in applications useful today.

The idea is that instead of aiming for AI to replace workers, it should be improving productivity across the entire economy at once.

As China has been limited by export controls over advanced chips, with its own domestic production only slowly catching up, it has also made an effort in training its AI more efficiently and focused more on AI inference and its deployment.

Qwen is currently one of China’s (and the world’s) leading AI, performing as well as or better than leading Western AIs on multiple benchmarks.

Source: Qwen

Alibaba’s AI progress is helped by the cloud computing capacity of Alibaba (1/3rd of the Chinese market and ranked #1) and the cash flow from the e-commerce business (twice as large as Amazon by gross merchandise value).

Source: Alibaba

In that context, the massive role of Alibaba in the Olympics should be understood not as a move to generate revenue, but to raise the profile of the company at a moment when Western AI companies dominate the global discussion and invest massively in advertisement like at the Super Bowl, which was dominated by AI-related ads.

(You can also read our dedicated investment report on Alibaba for more information on the company)

Latest Alibaba (BABA) Stock News and Developments