Inside the 2026 World Cup: How Lenovo's AI is Redefining Refereeing and Tactical Analysis
The partnership between Lenovo and FIFA, established in October 2024 at Tech World in Las Vegas, is not a classic sponsorship. Lenovo is the Official Technology Partner of the 2026 World Cup, scheduled to take place in the United States, Canada, and Mexico, as well as the Women's World Cup in 2027. The contractual scope covers IT infrastructure, devices, AI, data analysis, and managed services. The tournament's technological machinery is already operational: over 17,000 devices distributed across 16 stadiums and other competition venues, operational intelligence systems, digital twins of the facilities, and two AI projects that go directly to the heart of football decision-making.
In Milan, at a roundtable with the press, Pierluigi Collina, Chairman of the FIFA Referees Committee and Chief Refereeing Officer of FIFA, and Valerio Rizzo, AI Senior Manager & Solution Architect at Lenovo and technical lead of the 3D Digital Avatars project, described how these systems work, what industrial constraints they imposed, and what trajectory they might follow once they leave the World Cup perimeter.
Hyper-realistic 3D Avatars and Connected Ball
The first project is the 3D VAR, the visual evolution of refereeing technology. The algorithmic part, namely triangulating positions and detecting events on the field, remains owned by Hawk-Eye Innovations. Lenovo has worked on the next level, the visualization and generation of three-dimensional avatars of the players, integrated into semi-automatic offside reconstructions.
"We haven't changed the underlying technology, the measurement algorithms or event detection, we have changed the usability and perception of the data," explained Rizzo while describing the scope of the work.
According to Rizzo, the pipeline set up by the Lenovo team is the first in the history of cinema, media production, and video games to be fully automated. The characters in a football game come from the manual work of 2D and 3D designers, while here the entire chain is automatic. It starts with a scanner equipped with 36 high-resolution cameras (the initial version had 26), far fewer than the 120-180 sensors used in film scanners, but enough because the reconstruction is guided by AI. The pipeline includes five AI models (down from twelve) and is based on 3D Gaussian Splatting, a technique that generates a volumetric cloud of particles from a few two-dimensional images. On the player's body, segmented into parts by an AI model trained on millions of images, jerseys, physical conformations, and morphological details are mapped. On hair, one of the most challenging problems in 3D graphics, a statistical algorithm trained on millions of morphological models operates, capable of reconstructing the skull even when covered. All models used are open source, from European or U.S. origins, a choice dictated by adherence to international export and compliance directives.
The industrial constraint imposed a second level of automation. Players must be scanned in an A-pose with precise metrics because any angular variation of the arms, head, or fingers creates cascading errors. The scanner therefore contains an internal pipeline of AI models that reads in real-time the subject's position, corrects the head, eyes, hands, fingers, legs, and trunk, and only starts the acquisition when the pose is compliant. The 36 cameras shoot in parallel over a hundred milliseconds, and the entire passage of the player through the scanner usually lasts between ten and thirty seconds. The 1,248 athletes from the 48 participating nations will be scanned at base camps during media day, with 28 scanners distributed across the 48 locations.
The generated avatars feed into a centralized platform, a kind of 3D marketplace built from scratch by Lenovo, with version management, texture exploration, and team grouping. Hawk-Eye's micro data centers distributed in the stadiums download the assets before the matches. The infrastructure is owned by FIFA and managed by FIFA. The architecture is hybrid, in line with Lenovo's approach in AI-dedicated infrastructures in other sectors, with part of the processes in the cloud for logistical and cost reasons, and part kept close to the scanner’s main controller for latency reasons.
The declared accuracy of the scanner is less than a centimeter. The data combines with the connected ball introduced in Qatar 2022, which, thanks to an internal chip, sends 500 measurements per second, detecting the exact moment of the touch and the position relative to the players. Based on the sensor data, the algorithms decide the geographic offside, without relying on a still image selected by an operator. A new feature of the 2026 World Cup is the threshold of 10 cm above which the system notifies the assistant referee that the offside is clear, allowing the assistant to immediately raise the flag. This avoids unnecessary sprinting by attackers for thirty meters and the risk of injuries, a contraindication that VAR had introduced by asking assistants to keep the flag down in doubtful cases.
The logic that ties together this technological stratification, argued Collina, lies in the relationship between understanding and acceptance. When goal-line technology entered football, its graphic rendering was already clear to the audience, with the ball, line, and space in two dimensions, goal or no goal. The previous offside technology, based on 2D lines drawn on the field and on a perpendicular projection chosen by an operator, remained incomprehensible for good portions of the spectators.
"If we can make the decision understandable because I see what happened, I am more inclined to accept it. If I don't understand, I reject it a priori. Not because I don't see it, but because I don't understand it," summarized Collina. Hence the choice to move to three-dimensional visualization with recognizable bodies instead of abstract lines.
Football AI Pro, Coordinated Agents and Knowledge Graph
If the 3D VAR operates on the perception of the match event, the second intervention axis of the partnership shifts the AI towards tactical analysis. It's called Football AI Pro. It is a system based on generative AI that FIFA will distribute at the start of the tournament to all 48 national teams and, separately, to the refereeing team. It operates like a conversational assistant similar to those many business users use daily but is based on Lenovo's enterprise platform for knowledge management and is specialized in the football domain. Beneath the surface is a multi-agent architecture, with an orchestrator coordinating multiple specialized agents, each authorized to use a subset of tools and query their own data. The possible queries are not the classic ones of an SQL system, but semantic explorations and causal reasoning: how would a formation change behave against a specific opponent, which player with certain backward passing characteristics is most suited for the center-back role, at what phase of the match does a team show a recurring performance drop.
The enabling element is the FIFA Football Language, the semantic standard built by FIFA's innovation team to uniquely and transversally define concepts such as tackle, dribbling, or shot. It is the backbone of the knowledge graph that connects events, data, and tactical concepts, and without this ontology work, no natural language query could return coherent answers on the petabytes of tracking, statistics, videos, and historical analyses that FIFA retains. The elaborated metrics are over 2,000. The responses return in the form of text, graphs, video clips, and 3D reconstructions, with the possibility to directly query the footage and receive the relevant sequences.
The platform will only be accessible during preparation and post-match analysis phases, never during, and Collina positioned the tool within his career path as an international referee. In 2002, appointed for the World Cup final in Japan, he requested the local federation for video tapes of Brazil and Germany to prepare for the expected rivals, and had to justify the request. In 2015, as UEFA’s referee manager, he permanently included two match analysts in the refereeing staff. Today, that same work, which teams perform with tools like Wyscout, is integrated into a generative AI system accessible in seconds. The refereeing team is the forty-ninth user of the platform, and leading the training of the FIFA video match officials is Massimiliano Irrati, a former Italian referee who later joined the federal staff.
A Pioneer for Professional Football
On the replicability in lower-level leagues, prompted by Edge9, Lenovo and FIFA confirmed the scheme already followed with VAR and the semi-automatic offside technology of Qatar 2022. FIFA pushes the technological frontier to its limit at the World Cup, the industry derives less expensive versions suitable for the Champions League, Premier League, La Liga, and Serie A. Once the avatar generation process is automated, the unit cost becomes low, and the constraint is primarily organizational. The same automatic generation of avatars and pose estimation without skeletons has now also taken root outside football, thanks to public foundational models, from medical imaging to cinema, and basketball.
Among the collateral projects presented is also the new generation of the Referee View, the referee's body cam that FIFA had already experimented with at the Club World Cup 2025 and which for the 2026 World Cup is equipped with an AI stabilizer developed together with Lenovo. The new feature compared to the Club World Cup is the stabilization algorithm, which intervenes on motion blur and vibrations caused by rapid head movements, providing images fluid enough to be used in live broadcasts. The footage, tied to the referee's gaze, offers a viewpoint closer to the match decision, and with the new stabilization, it becomes usable even for a general audience.
Taken together, the 3D VAR, Football AI Pro, and Referee View define Lenovo's profile as FIFA’s technology partner, not merely as a sponsor, with its teams working alongside those of Hawk-Eye, the broadcasters, and the Federation's innovation unit. None of the three projects would have been possible without this operational integration, and it is in their coherence, more than in the individual functionalities displayed in Milan, that the maturity achieved by the technological partnership between the two organizations is measured.