Key Takeaways
- Professional cycling teams are increasingly embedding artificial intelligence (AI) into training, race strategy, nutrition, and talent identification, moving beyond legacy platforms like TrainingPeaks.
- AI models that focus on critical power and ‘W’ prime—rather than functional threshold power (FTP)—provide a finer‑grained view of a rider’s endurance and sprint capacity, especially under fatigue.
- Real‑time data from wearables (Whoop, Oura) and course‑specific GPX files enable predictive modelling that can forecast stage performance and guide in‑race decisions.
- Nutrition‑planning apps, such as the one co‑developed by Team Picnic‑PostNL and IG&H, use AI to align energy intake with predicted expenditure, improving fueling accuracy from ~52% to as high as 82%.
- AI‑driven “digital twins” and big‑data scouting are helping WorldTour squads spot future stars and optimise equipment choices before a rider even hits the road.
- Women‑specific AI research—exemplified by Kristen Faulkner’s personal model that incorporates menstrual‑cycle data—shows promise for tailoring training and recovery to female physiology.
- Experts like Frank Overton believe that, within two years, every WorldTour team will field an AI programme, and those that build it well will gain the marginal edge needed to win Grand Tours.
Frank Overton’s Skepticism Sets the Stage
“I have a theory that it’s all smoke and mirrors. UAE Team Emirates‑XRG has reportedly been using it for years. Visma‑Lease a Bike has a link with Mistral. Ineos has brought Netcompany on board. But apart from their press releases, we have no idea how they’re using it, which is odd, because people in cycling are a) bad at keeping secrets and b) have something to sell.” That candid assessment from Frank Overton, founder of FasCat Coaching, captures the lingering mystery around how professional squads actually apply artificial intelligence (AI) despite the hype. Overton, who has been feeding his own proprietary AI with rider data for years, notes that many teams’ AI requests fell on deaf ears—until the progressive staff at Team Picnic‑PostNL opened their doors to a deeper look.
Picnic‑PostNL’s Data‑Driven Foundation
At Team Picnic‑PostNL, data analysis and AI play an increasingly important role in optimising both performance and development, explains Narelle Neumann, the team’s head of science. “Every rider uploads daily training files containing thousands of rows of data, including power output, heart rate, cadence, speed and altitude information. During a season, this results in millions of data points.” The team processes these files through analytical models that monitor relationships such as power versus heart rate over time, correcting for temperature and altitude to reveal true fitness trends. Subjective logbook entries—perceived recovery, fitness, and RPE—are also harvested, creating a rich database that machine learning can mine for actionable insights.
Vekta’s Shift from FTP to Critical Power
Decathlon CMA CGM, Lidl‑Trek, TotalEnergies and Jayco‑AlUla have ditched the long‑standing TrainingPeaks platform in favour of the Vekta AI coaching system. According to Vekta co‑founder Paul‑Anton Girard, “Vekta has multiple uses. Refining training is one of them.” He illustrates the benefit with interval workouts: after a rider completes six 10‑minute efforts, Vekta instantly compares the session to past equivalents to gauge improvement and advise adjustments. Crucially, Vekta dispenses with functional threshold power (FTP) as the basis for training zones, instead focusing on critical power—the highest sustainable output—and ‘W’ prime, the finite anaerobic reserve that fuels sprints. Girard argues this gives “a more complete view of performance than FTP,” allowing coaches to see how deep a rider went and how deep they can go.
Durability and Fatigue Resistance
Vekta also highlights a rider’s durability—a buzzword in modern peloton circles. Research by coach James Spragg shows that while under‑23 riders can match professionals when fresh, the real gap emerges after significant fatigue accumulates. Elite Grand Tour contenders maintain high power outputs even after thousands of kilojoules of work, a trait rooted in nutrition, functional strength, aerobic endurance, and genetics. Vekta lets users analyse power files across fatigue levels (0–50 kJ/kg) and displays current kilojoules expended before each interval, enabling precise, fatigue‑aware training prescription.
Predictive Modelling for Race Day
Beyond refining workouts, Vekta helps predict performance. Teams upload GPX course files for each Tour de France stage; combined with a rider’s power data, the AI can estimate what they are capable of on that specific parcours. During a race, recovery metrics from wearables like Whoop or Oura—heart‑rate variability, sleep, wellness questionnaire responses—are fed into the model, giving coaches a real‑time picture of a rider’s current capacity. While tools such as Best Bike Split have long offered pacing optimisation, Vekta’s integration of live physiological data marks a step toward dynamic, AI‑driven race strategy.
Overton’s Layered Approach to the Queen Stage
Frank Overton is a keen advocate of AI for modelling race demands. Take stage 20 of the 2026 Tour de France—the Queen Stage, a 171 km slog from Bourg d’Oisans to a second successive Alpe d’Huez finish, featuring Croix de Fer, Télégraphe, Galibier and Col de Sarenne. By the time GC contenders reach the base of Alpe d’Huez, they will have roughly 4,500 m of climbing and 3,032 km of racing in their legs. Overton proposes analysing the stage in three layers:
- Peak power up Alpe d’Huez – model each rider’s best 30‑ to 45‑minute power output, scaled to watts per kilogram.
- Power output under fatigue – determine what a rider can sustain after 19 stages and four hard climbs, i.e., their durability at threshold and above when 3,000–4,000 kJ have been expended.
- GC modelling – combine layers one and two with the race situation entering the stage to predict the power needed to close or defend a specific time gap, incorporating prior climb dynamics, energy costs, and historical fatigue profiles.
“These are exactly the types of questions that properly trained AI models can answer very well and in real time,” Overton adds, noting that while peak power is straightforward, prescribing the exact training required 11 months in advance is the true “secret sauce.”
The Importance of Quality Data
Overton cautions that an AI system is only as good as the information fed into it. “There’s an illusion that the likes of ChatGPT and Claude deliver good training advice. But it’s not trained, meaning it doesn’t answer like a coach or AI using a relevant dataset. It takes information from the general internet… there’s no thought leadership.” His own CoachCat app is the product of years of uploading rider training data, blogs, podcasts, and training plans—proof that “for AI, data is rocket fuel. The more you have, the more insight it can provide, the better it is. But it must be the right data.”
AI‑Enhanced Nutrition Planning
Picnic‑PostNL has taken AI beyond the power meter, developing an app with IG&H to support nutrition planning during training and races. Neumann explains: “Using data models, the app predicts riders’ energy expenditure during races based on parcours information such as elevation and distance. Similar models estimate energy expenditure during training sessions based on workout descriptions. Based on these predictions, the app helps nutritionists and cooks to plan strategies and daily meals, ensuring that energy intake is aligned with the demands of training, racing and recovery.” The approach mirrors Visma‑Lease a Bike’s method, which, with AI, reports nutrition prediction accuracy ranging from a moderate 52 % to an excellent 82 % when weather conditions are factored in.
Talent Scouting and Digital Twins
AI’s potential extends beyond performance optimisation to talent identification. Girard announces an upcoming academy with a WorldTour team that will use a big‑data approach to scout juniors and predict their future performance two or three years out. Ineos Grenadiers (now Netcompany‑Ineos) have partnered with Swansea University to develop “digital twins”—data‑driven rider profiles built from internal metrics and publicly available race data. The goal is an automated system that flags standout junior performances, helping British squads unearth future stars. These digital twins also serve in equipment selection, simulating helmets, wheel depths, riding positions, and weather conditions in a virtual wind tunnel before a rider ever hits the road.
Women‑Specific AI Research
Seeing a unique opportunity, teams are turning AI toward the Tour de France Femmes. Girard notes collaboration with FDJ United‑Suez to understand how the menstrual cycle influences power output and recovery. Kristen Faulkner, EF Education‑Oatly athlete and Harvard computer‑science graduate, built her own AI model using nine years of personal data—4,440 hours of training, heart‑rate, HRV, sleep, weight, power, temperature, menstrual‑cycle phases, bloodwork and DEXA scans. “Every model is trained on my body,” she wrote. “Every finding is specific to my history. And every output is actionable, not just interesting.” Faulkner used the model to prepare for the Pan Am Championships, where she won three gold medals, and recently set a new personal best 20‑minute power output, underscoring AI’s potential to elevate women’s performance research from the ground up.
Conclusion: The Inevitable AI Future in Cycling
All told, artificial intelligence is poised to accelerate a sport where data already reigns supreme. By rapidly analysing every metric generated—from power files to sleep scores—AI equips WorldTour coaches with specific workout prescriptions, clearer race predictions, and bespoke feeding advice. As Overton signs off, echoing Sir Dave Brailsford’s optimism, he predicts that within 24 months every World Tour team will field an AI programme; those that build them well will win races, while those that lag will be dropped. The era of AI‑driven marginal gains has arrived, and the peloton is already feeling its shift.
https://www.bikeradar.com/features/long-reads/inside-the-ai-arms-race